--- title: "I Used AI Agents for Every Go-To-Market Role (Sales, Marketing, CS, RevOps)" episode: 74 podcast: "The LeanScale Podcast" publisher: "LeanScale" guest: "Yasin" guest_title: "Head of AI (former), LeanScale" date_published: 2026-05-15 date_modified: 2026-07-22 duration: 00:48:03 word_count: 12283 topics: ["ai-in-gtm", "revenue-operations", "gtm-strategy", "forecasting", "sales-leadership", "demand-generation"] canonical_url: https://leanscale-knowledge-hub.netlify.app/podcast/ai-agents-for-every-go-to-market-role/ source: "LeanScale Podcast Knowledge Hub — https://leanscale-knowledge-hub.netlify.app" license: "Free to quote and cite with attribution to The LeanScale Podcast." --- # I Used AI Agents for Every Go-To-Market Role (Sales, Marketing, CS, RevOps) _A live build-along: AI agents for sales, sales management, marketing, customer success, and RevOps — plus the 2026 agent-platform landscape_ **Episode 74 · The LeanScale Podcast** Yasin, Head of AI (former), LeanScale Published May 15, 2026 · Updated July 22, 2026 · 00:48:03 Canonical: https://leanscale-knowledge-hub.netlify.app/podcast/ai-agents-for-every-go-to-market-role/ **Topics:** AI in GTM · Revenue Operations · GTM Strategy · Forecasting · Sales Leadership · Demand Generation ## Executive summary In this build-along episode of The LeanScale Podcast, Yasin — LeanScale's Head of AI — puts AI agents to work live across every go-to-market function, not as theory or toy prompts but as agents actually running in the background while he narrates. Five scenarios anchor the hour: a sales rep recovering from a call that went sideways with Notion, a sales manager deciding whether a rep named Michael is coachable, a marketing team turning six prospect calls into a messaging verdict and a branded slide deck, a customer success manager mapping stakeholders after a champion leaves Amplitude six weeks before renewal, and a RevOps manager pressure-testing a $2.3M forecast before it goes to the board. Each agent launches sub-agents, pulls transcripts, researches the market, writes files to disk, and (where connected) updates Salesforce via MCP. The central reframe is that the thing most people call 'AI' — the browser chatbot, the consumer app — is the wrong product for real work. Each major lab (OpenAI, Google, Anthropic) ships three products: the model (GPT-5.2, Gemini 3, Claude Opus 4.5), the consumer app for general Q&A, and the agent platform for professionals who need to get work done. Yasin's demo runs inside Claude Code as a VS Code extension, and he draws a hard line between the two surfaces: only agent platforms can maintain a persistent to-do list, spin up sub-agents, read and write files on your machine, switch between plan and bypass-permission modes, queue messages, and compact their own context. Setup, he shows, is three steps and takes five minutes: download VS Code, add the official Claude Code extension, log in with a $20-a-month Claude subscription. The moat, Yasin argues, is context. Feeding the agent your own playbooks, objection-handling docs, personas, sales frameworks, and call transcripts is what turns raw model intelligence into department-specific output — the difference between asking a chatbot a question and operationalizing a prompt that does the work. And the prompt itself is architected deliberately: give the agent context, tell it to launch sub-agents, have it track a to-do list, instruct it to be token-efficient, direct it to write files, and — the throughline of the show notes — tell it to think through the problem like a strategist, not a chatbot. He layers in the practical literacy that makes this durable: what a token is, why the 200,000-token window is working memory, why bigger isn't always better, and how to 'compact' one agent's context into a clean hand-off to a fresh one. The endgame he points to is a world where you connect MCP servers (Salesforce, HubSpot, Teamwork) and stop logging into SaaS UIs entirely, while a shared file ecosystem — GitHub replacing Google Drive — becomes the team's brain of skills, SOPs, and context. Adoption at LeanScale is already real: the whole team burned 1.86 billion tokens in a couple of weeks, tracked on a live leaderboard. Who should listen: any sales, marketing, CS, or RevOps operator who keeps hearing 'AI agents' and wants a concrete, reproducible picture of what they actually do inside real workflows today. The takeaway is that the leverage isn't the model — it's the agent platform plus your own context, wielded by an operator who prompts like a strategist. ## Key takeaways 1. **AI agents can now do real work in every GTM function — live, not in theory** — Yasin runs five agents concurrently across sales, sales management, marketing, customer success, and RevOps. Each ingests real context, launches sub-agents, researches the market, and writes usable deliverables (call analyses, recovery emails, coachability assessments, messaging verdicts, stakeholder maps, forecast reality checks) while he keeps narrating. _Why it matters:_ Stop evaluating agents on chat demos. The bar is whether an agent can take your context and produce an output you'd actually ship — and across GTM, that bar is already being cleared today. _For:_ RevOps Leaders, Sales Leaders, Marketing Leaders, Customer Success, Founders 2. **The unlock is the agent platform — not the chatbot most people call 'AI'** — Every lab ships three products: the model, the consumer app (the browser chatbot for general Q&A), and the agent platform for getting work done. The chatbot is what most people mistake for AI; the agent platform is a different product entirely, and it's where GTM work belongs. _Why it matters:_ If your team's mental model of AI stops at ChatGPT-in-a-browser, you're evaluating the wrong product. Move the professional use cases onto an agent platform. _For:_ RevOps Leaders, Founders, Revenue Executives 3. **Context is the moat — feed the agent your own playbooks and transcripts** — The difference between raw model intelligence in a browser and a useful output is the context you supply: objection-handling docs, sales frameworks, personas, current messaging, account history, and real call transcripts. That's what makes the output specific to your company and department. _Why it matters:_ Invest in building a clean, reusable context library (playbooks, SOPs, frameworks). Your proprietary context — not the model — is the durable advantage. _For:_ RevOps Leaders, Sales Leaders, Marketing Leaders, Customer Success 4. **Architect the prompt like a strategist, not a chatbot** — Yasin builds each prompt as a system: give context files, tell it to launch sub-agents, have it maintain a to-do list, instruct it to be token-efficient (pull relevant quotes, don't dump), direct it to write files, and end with 'think through this like a strategist / thought partner.' That structure is what produces strategic output instead of a summary. _Why it matters:_ Treat prompting as agent architecture. The repeatable pattern — context, sub-agents, to-do list, token efficiency, file output, strategic framing — is more valuable than any single clever prompt. _For:_ RevOps Leaders, Founders, Sales Leaders 5. **Sub-agents and persistent to-do lists let one agent work autonomously for a long time** — A main agent can spin up sub-agents that research in parallel and report back, and it maintains an internal to-do list so it holds intent over long runs. Yasin says agents routinely work 10–40 minutes autonomously, and he's had one run for four hours on a task. _Why it matters:_ Design work for asynchronous execution: hand the agent a big, well-specified job and let it grind while you do something else. Calibrate expectations — good work takes the agent real time, just as it would a human. _For:_ RevOps Leaders, Founders, Revenue Executives 6. **Setup is three steps, five minutes, and $20 a month** — Download VS Code, install the official Claude Code extension from Anthropic (the building-blocks icon), and log in with a $20/month Claude subscription. That's the entire on-ramp to run everything demoed in the episode; upgrade prompts only appear once you exhaust the base plan's tokens. _Why it matters:_ The barrier to getting started is near zero. There's no reason for a GTM operator not to have an agent platform running this week. _For:_ Founders, RevOps Leaders, Sales Leaders 7. **Sales rep: turn a lost call into an analysis, a recovery email, and honest CRM notes** — Given the Notion call transcript plus the sales playbook and objection-handling docs, the agent identifies where the call derailed (abandoning discovery for a feature demo in the first 25 seconds), drafts a recovery email addressing the real concerns, writes CRM notes, and appends live market research on why the pricing ask missed. _Why it matters:_ Every hard call becomes a coaching artifact and a next step. Reps can debrief properly in minutes with market context they didn't have in the room. _For:_ Sales Leaders, RevOps Leaders 8. **Sales manager: get a coachability verdict from pattern analysis across many calls** — Fed a rep profile, the team's sales framework, and Michael's last five transcripts, the agent analyzes the pattern (not each call separately), rates him on the framework's dimensions, cross-references prior coaching notes, and returns a coachability call plus a 30-day plan — or the outline of the hard conversation if he's not. _Why it matters:_ Managers can make evidence-based people decisions grounded in their own rubric, spotting persistent weaknesses versus one-off mistakes instead of relying on gut and recency bias. _For:_ Sales Leaders, RevOps Leaders 9. **Marketing: prospect calls → messaging verdict → blog → branded, editable slides** — From six mixed win/loss/stall calls plus current messaging docs, the agent says what's landing ('implementation in weeks not months' — keep and amplify), what's mis-said ('one platform vs. five tools' resonates less than 'a focused tool that does one thing well'), with transcript citations. It then drafts an on-brand blog post, generates wireframes for approval, and outputs a PowerPoint in LeanScale branding where every element is real, editable text — not an image. _Why it matters:_ Messaging research, content, and design collapse into one operator-driven workflow. The slides come out editable and on-brand, ready to drop into Google Slides. _For:_ Marketing Leaders, RevOps Leaders 10. **Customer success: build a stakeholder map before a champion-churn renewal** — When the champion (Sarah) leaves Amplitude six weeks before renewal, the agent researches new leadership via sub-agents, mines past calls for who else attended and what they cared about, maps the three to four stakeholders to engage, flags internal politics and budget pressure, and gives an honest read on whether the account is likely to churn. _Why it matters:_ CS can move fast on silent risk — reconstructing the relationship graph and a week-by-week action plan from CRM history and call transcripts instead of starting blind. _For:_ Customer Success, Revenue Executives 11. **RevOps: pressure-test the forecast against pipeline red flags before the board sees it** — Given red-flag criteria (skipped stages, no activity in two weeks, close dates pushed more than once, no logged customer meeting in 30 days, single-threaded deals), pipeline, forecast export, and rep activity, a purpose-built forecast-analyzer agent audits the $2.3M commit and returns a realistic $1.2M–$1.75M range, names the rep carrying dead deals, surfaces the risk patterns, and drafts what to tell the board. _Why it matters:_ The forecast conversation becomes evidence-backed. RevOps can walk into the board meeting with a defensible number and the deal-level reasons behind the gap. _For:_ RevOps Leaders, Revenue Executives, Sales Leaders 12. **Token literacy is the new operating skill: understand the window and compact it** — A token is the atomic unit of how AI thinks (~3–4 characters). The context window (e.g., 200K for Claude 4.5, 1M for Gemini 3) is working memory that holds all inputs and outputs; hit the limit and the agent forgets earlier context. Agent platforms show your usage and let you 'compact' — summarize and hand off to a fresh agent with a clean window, like an employee ending a shift. _Why it matters:_ Operators must manage context like a resource: bigger windows aren't always better (more context, less attention to detail), and compaction lets you preserve intent across long, token-heavy jobs. _For:_ RevOps Leaders, Founders 13. **Skills let you download a capability the way Neo downloads kung fu** — Anthropic and the other labs adopted a shared skills standard: a folder of files that teaches an agent how to do a task (make a PowerPoint, an SOP, a PDF, wireframes). You point the agent at the skill's path and it executes — and you can build your own from your team's repeated work. _Why it matters:_ Package your team's recurring outputs as reusable skills so a thin team can run sophisticated workflows on demand, and compound learning by saving good outputs as new SOPs. _For:_ RevOps Leaders, Marketing Leaders, Founders 14. **The endgame: connect MCPs, stop logging into SaaS UIs, replace Drive with a shared file brain** — Beyond internal files, agent platforms connect to systems via MCP — Salesforce, HubSpot, Teamwork — so you can ask the agent to update an opportunity or summarize a project without opening the app. Yasin's stated goal: never log into a SaaS UI again. At LeanScale, a shared file ecosystem on GitHub is starting to replace Google Drive as the team's context store. _Why it matters:_ The interface to your stack becomes the agent, not fifteen browser tabs. Standing up MCP connections and a shared context repo is the infrastructure that makes agent-first GTM real. _For:_ RevOps Leaders, Founders, Revenue Executives ## Frameworks ### The Three AI Products: Model, Consumer App, Agent Platform (12:41) **Definition:** Each major lab (OpenAI, Google, Anthropic) ships three distinct products: the model (baseline infrastructure — GPT-5.2, Gemini 3, Claude Opus 4.5), the consumer app (the browser chatbot for general-population Q&A), and the agent platform (for professionals to get work done). Yasin's core reframe: what most people call 'AI' is only the consumer app, built for general use cases like Q&A, therapy, or health advice. Real GTM work belongs on the agent platform, a separate product with a different feature set. Bigger models aren't automatically better — larger context windows trade detail-attention for ingestion capacity. ### Agent Platform vs. Consumer App: The Feature Divide (14:25) **Definition:** The capabilities that only agent platforms have and consumer apps lack: a persistent internal to-do list (so the agent works 10–40+ minutes autonomously), the ability to launch sub-agents, reading and writing files on your local machine, permission/plan modes, queued messages, and context compaction. These features are why the agent platform can operationalize a prompt into real work rather than just answering a question. The main agent orchestrates sub-agents that research in parallel and report back, maintains intent via its to-do list, and produces file deliverables you can open and edit. ### Agentic Prompt Architecture (Think Like a Strategist, Not a Chatbot) (03:04) **Definition:** A repeatable structure for building agent prompts: (1) supply context files, (2) instruct it to launch sub-agents, (3) have it maintain a to-do list, (4) tell it to be token-efficient, (5) direct it to write outputs to files, and (6) frame it to think like a strategist / thought partner. Across all five use cases Yasin assembles the prompt the same way, layering context and instructions so the agent behaves like an operator, not a Q&A bot. The strategist framing at the end is what elevates the output from a summary to a decision-grade deliverable. ### The Token Window & Context Compaction (32:15) **Definition:** A token is the atomic unit of how AI thinks (~3–4 characters). The context window is finite working memory holding all inputs and outputs (e.g., 200K for Claude 4.5, 1M for Gemini 3); once full, the agent forgets earlier context. 'Compacting' summarizes the current context and hands it off to a fresh agent with a clean window. Agent platforms surface live usage (e.g., '73% used, 27% remaining') and let you compact before a big job — like an employee handing off shift notes to the next person. Managing the window is a core operating discipline, and bigger isn't always better because more context dilutes attention to detail. ### Agent Skills (Download a Capability) (29:35) **Definition:** A skill is a folder of files that teaches an agent how to perform a task (make a PowerPoint, an SOP, a PDF, wireframes). The labs adopted a shared skills standard, so you can download skills from the internet or build your own and point the agent at the skill's path to execute it. Yasin's analogy: it's like Neo in the Matrix instantly learning kung fu — the skill gives the agent a new capability on demand. This lets a small team run sophisticated, branded workflows and compound learning by saving strong outputs as reusable SOPs/skills. ### The Three-Step Agent Platform Setup (14:54) **Definition:** Getting an agent platform running in ~5 minutes: (1) download VS Code, (2) install the official Claude Code extension from Anthropic, (3) log in with a $20/month Claude subscription. Restart, click the orange icon, authorize, and the agent is enabled. Yasin frames the near-zero barrier deliberately: everything demoed in the episode runs on the base $20 plan until you exhaust its tokens. The same folder-based workspace then becomes a shareable ecosystem of files, skills, and context for the whole team. ## Quotes _Speakers inferred from an undiarized transcript — verify before attributing._ > "There's a lot of hype around AI and go-to-market. What does it even mean to be using AI agents? What the heck even is an AI agent?" > > — Yasin, The LeanScale Podcast Ep. 74 (00:00) > "I want you to think through this like a strategist." > > — Yasin, The LeanScale Podcast Ep. 74 (03:04) > "The agent platform is for professionals. It's actually to get work done." > > — Yasin, The LeanScale Podcast Ep. 74 (13:59) > "I've actually had agents that have worked for four hours asynchronously on a task." > > — Yasin, The LeanScale Podcast Ep. 74 (14:54) > "You can run this entire system, and everything that I just showed you, for $20 a month. How insane is that?" > > — Yasin, The LeanScale Podcast Ep. 74 (16:12) > "This is the big difference between just using the raw intelligence of AI inside of a browser, and then actually feeding it the context that's going to allow it to be personal for each and every single department." > > — Yasin, The LeanScale Podcast Ep. 74 (22:32) > "This is where you've got to have some realistic expectations, because if you expected a human to do this, it would take them some time too." > > — Yasin, The LeanScale Podcast Ep. 74 (24:55) > "It's kind of like Neo in the Matrix, where he's like, 'I don't know kung fu,' and then, boom, now you know kung fu. That's what skills are to AI agents." > > — Yasin, The LeanScale Podcast Ep. 74 (29:35) > "A token is the atomic unit of how AI thinks — one token is approximately three to four characters." > > — Yasin, The LeanScale Podcast Ep. 74 (32:15) > "When you get to that 200,000 limit, it's going to start to forget what happened before that." > > — Yasin, The LeanScale Podcast Ep. 74 (34:12) > "No, no, no — the future is going to be utilizing agent platforms." > > — Yasin, The LeanScale Podcast Ep. 74 (43:55) > "I have a joke that I make here at LeanScale where I say I do not want to log into any single SaaS UI ever again. Everything that we do going forward is happening within the agent platform." > > — Yasin, The LeanScale Podcast Ep. 74 (44:23) > "The power user section is like, okay, let's get out the construction equipment — let's get out the bulldozer, let's get out the chainsaw." > > — Yasin, The LeanScale Podcast Ep. 74 (46:08) > "Just in the last couple weeks our team has used 1.86 billion tokens, because we now have our entire team using these agent platforms." > > — Yasin, The LeanScale Podcast Ep. 74 (47:01) ## Practical advice by role ### Founders - Get an agent platform running this week — it's three steps, five minutes, and $20/month: VS Code, the official Claude Code extension, and a Claude subscription. - Treat your team's context as infrastructure: stand up a shared file ecosystem (GitHub can replace Google Drive) of playbooks, SOPs, frameworks, and skills that every function can reuse. - Point at the endgame deliberately — connect MCP servers (Salesforce, HubSpot, Teamwork) so the agent becomes the interface to your stack instead of fifteen SaaS UIs. ### RevOps Leaders - Build a purpose-built forecast-analyzer agent with an explicit SOP and your red-flag criteria (skipped stages, stale activity, repeatedly pushed close dates, single-threaded deals) so you can audit a commit before the board sees it. - Architect prompts as systems: context files, sub-agents, a to-do list, token-efficiency instructions, file outputs, and a 'think like a strategist' framing. - Manage the token window as a resource — watch usage, and compact to hand off a clean context before token-heavy jobs. - Codify recurring outputs (SOP-makers, wireframe-to-PowerPoint) as skills so a small team can run sophisticated, on-brand workflows. ### Sales Leaders - After a tough call, feed the transcript plus your playbook and objection-handling docs to an agent to get a call analysis, a recovery email, and honest CRM notes — with live market research on why the ask missed. - For coaching decisions, give the agent a rep profile, your sales framework, and the last several call transcripts, and ask for the pattern (not per-call summaries), a coachability verdict, and a 30-day plan. - Save the outputs you like as SOPs so future analyses and emails come out in the format you want. ### Marketing Leaders - Turn a batch of prospect calls (wins, losses, stalls) plus your current messaging into a messaging verdict with transcript citations — what to keep and amplify versus what you're saying wrong. - Chain the workflow: messaging analysis → on-brand blog post → wireframes you approve in chat → an editable, branded slide deck; use a brand-tone file and a wireframe-to-PowerPoint skill to keep it on-brand. - Save wireframes to a file before compacting context so a long, token-heavy deck build doesn't lose them. ### Customer Success - When a champion leaves, give the agent the full account history and your champion-departure playbook and have it build a stakeholder map of the three to four people to engage next. - Use sub-agents to research the new leadership and mine past calls for who else attended, their concerns, and signals of internal politics or budget pressure. - Ask for an honest turn-risk read and a week-by-week action plan against the renewal date so you know fast whether the account is salvageable. ### Revenue Executives - Stop equating 'AI' with the browser chatbot — the professional leverage lives on the agent platform, and moving GTM work onto it is an operating decision, not a tooling preference. - Judge adoption on real usage (LeanScale tracks a live token leaderboard) and make the agent platform the shared operating surface across every revenue function. ## AI takeaways **Thesis:** The leverage in AI-for-GTM isn't the model — it's the agent platform plus your own context, wielded by an operator who prompts like a strategist. The browser chatbot most people call 'AI' is the wrong product for real work; agent platforms that run sub-agents, write files, connect to your stack via MCP, and manage their own context are already doing usable work across every go-to-market function. - **Agent platform, not chatbot** — Each lab ships a model, a consumer app, and an agent platform. The consumer app is for general Q&A; the agent platform (e.g., Claude Code in VS Code) is where professional GTM work lives — with to-do lists, sub-agents, file I/O, permission/plan modes, queued messages, and compaction. - **Context is the moat** — Feeding the agent your playbooks, frameworks, personas, and call transcripts is what turns raw intelligence into department-specific output. Your proprietary context — not the model — is the durable advantage. - **Prompt like a strategist** — Architect prompts as systems: context, sub-agents, a to-do list, token efficiency, file outputs, and a 'think like a strategist / thought partner' framing that elevates output from summary to decision-grade. - **Sub-agents + to-do lists = autonomous work** — A main agent orchestrates sub-agents that research in parallel and holds intent via an internal to-do list, working 10–40 minutes (sometimes hours) on a well-specified job. - **Token literacy** — The context window is finite working memory; hit the limit and the agent forgets. Compaction summarizes and hands off to a fresh window — bigger windows aren't always better because more context dilutes attention. - **Skills are downloadable capabilities** — A skill is a folder of files that teaches an agent a task (PowerPoint, SOP, PDF, wireframes). Download them or build your own to let a thin team run sophisticated, on-brand workflows. - **MCP endgame** — Connect Salesforce, HubSpot, and Teamwork via MCP and the agent becomes the interface to your stack — the goal of never logging into a SaaS UI again — with a shared file repo (GitHub over Google Drive) as the team brain. **Agent & automation ideas** - Sales-call recovery agent: ingest a call transcript plus playbook and objection docs; launch a sub-agent to research recent news about the account; output a call analysis, a recovery email, and CRM notes, and write them to Salesforce via MCP. - Rep-coachability agent: analyze the last N call transcripts for patterns against your sales framework, cross-reference prior coaching notes, and return a coachability verdict plus a 30-day plan or the hard-conversation outline. - Messaging-analysis agent: turn a batch of win/loss/stall prospect calls plus current messaging into a keep/fix verdict with transcript citations, then chain into a blog post, wireframes, and a branded PowerPoint via skills. - Champion-churn agent: from account history and a departure playbook, spin up sub-agents to research new leadership, mine past calls for stakeholders and politics, and produce a stakeholder map with a turn-risk read against the renewal date. - Forecast-analyzer agent: a purpose-built agent with an SOP that cross-references pipeline, forecast, and activity data against red-flag rules to return a realistic commit range and what to tell the board. - SOP/skill builders: agents that turn a good output (a great recovery email, a wireframe layout) into a reusable SOP or skill so the workflow compounds over time. ## Operations takeaways ### Revenue operations - **Operate on the agent platform.** The professional surface for RevOps is the agent platform, not the browser chatbot — that's where you can run sub-agents, write files, and connect to the CRM via MCP. - **Purpose-built forecast agent.** Encode your pipeline red-flag criteria into a forecast-analyzer agent with its own SOP so you can audit a commit before the board and defend a realistic range. - **Prompt architecture.** Build prompts as systems — context, sub-agents, to-do list, token efficiency, file outputs, strategist framing — so agents produce decision-grade work, not summaries. - **Token discipline.** Treat the context window as a managed resource: monitor usage and compact to hand off a clean window before token-heavy jobs. - **Context library as infrastructure.** Stand up a shared, reusable file ecosystem of playbooks, SOPs, frameworks, and skills (GitHub replacing Google Drive) that every function draws on. - **MCP over SaaS UIs.** Connect Salesforce, HubSpot, and Teamwork via MCP so the agent becomes the interface to the stack instead of a browser tab per tool. ### Pipeline & marketing ops - **Red-flag audit.** Skipped stages, no activity in two weeks, close dates pushed more than once, no logged customer meeting in 30 days, and single-threaded deals are the flags an agent can systematically check against the whole pipeline. - **Commit vs. reality.** The agent turns a $2.3M submitted commit into a realistic $1.2M–$1.75M range and names the rep carrying dead deals as committed — evidence, not gut, for the forecast call. - **Pattern detection.** Ask the agent for patterns by stage, deal size, or time-in-stage that correlate with weak deals so you can update SOPs and your understanding of the business. - **Board-ready output.** The forecast agent drafts what to tell the board and the immediate actions required this month, not just a number. ### Customer operations - **Reconstruct the relationship graph.** When a champion leaves, an agent rebuilds the stakeholder map from account history and past calls — who else attended, their concerns, and internal politics — instead of starting blind. - **Research the new leadership.** Sub-agents investigate the replacement's role and priorities and the account's financial health so CS can engage the right three to four people. - **Honest turn-risk read.** Ask the agent to say plainly whether the account is likely to churn and whether it's a lost cause, given the weeks remaining to renewal. - **Week-by-week action plan.** The output includes a recommended action plan mapped to each week before renewal, turning silent risk into a concrete play. ## Metrics mentioned | Value | Metric | Context | | --- | --- | --- | | $20/month | Setup cost | The base Claude subscription runs everything demoed; upgrade prompts appear only after the plan's tokens are exhausted. | | 3 steps / ~5 minutes | Setup effort | Download VS Code, install the Claude Code extension, log in with a Claude subscription. | | 200,000 tokens | Claude context window | Claude 4.5's working memory; a focused window that Yasin calls a good sweet spot for attention to detail. | | 1,000,000 tokens | Gemini context window | Gemini 3's larger window — more ingestion capacity, but more context means less attention to detail. | | ~3–4 characters | What a token is | The atomic unit of how AI thinks; 'hello how are you' is about six tokens. | | 10–40 min (up to 4 hours) | Autonomous agent runtime | How long agents work asynchronously on a task; Yasin has had one run for four hours. | | 1.86 billion tokens | LeanScale team token usage | Consumed in the last couple of weeks across the team, tracked on a live leaderboard that refreshes every ~30 seconds. | | $2.3M commit → $1.2M–$1.75M realistic | RevOps forecast audit | The forecast-analyzer agent's honest range after flagging one rep's dead deals carried as committed. | | 73% used / 27% remaining | Context window indicator | The live usage readout an agent platform shows, prompting a compact before a token-heavy job. | ## Entities mentioned - **LeanScale** (company) — Yasin narrates as LeanScale's Head of AI; the demo uses LeanScale playbooks, branding, and its team-wide token leaderboard (1.86B tokens in a couple weeks). · https://leanscale-knowledge-hub.netlify.app/company/leanscale/ - **Anthropic** (company) — Maker of Claude and the official Claude Code extension used throughout the demo; cited as one of the three big AI players alongside OpenAI and Google (transcribed as 'Entrop'). · https://leanscale-knowledge-hub.netlify.app/company/anthropic/ - **OpenAI** (company) — Named as one of the three major AI companies; ships GPT-5.2 as its model and a consumer app. · https://leanscale-knowledge-hub.netlify.app/company/openai/ - **Google** (company) — Named as one of the three major AI companies (ships Gemini 3); still used for everyday tasks like searching 'VS Code download.' · https://leanscale-knowledge-hub.netlify.app/company/google/ - **Notion** (company) — Example prospect account for the sales-rep use case — the agent analyzes a tough Notion call, drafts a recovery email, and researches recent Notion news to explain the pricing pushback. · https://leanscale-knowledge-hub.netlify.app/company/notion/ - **Amplitude** (company) — Example account for the customer-success use case — the champion (Sarah) leaves six weeks before renewal, and the agent builds a stakeholder map and turn-risk read for Amplitude. · https://leanscale-knowledge-hub.netlify.app/company/amplitude/ - **Yasin** (person, narrator) — Former Head of AI at LeanScale; host of the AI-agent build-along episodes. · https://leanscale-knowledge-hub.netlify.app/guest/yasin/ - **Claude Code** (tool, AI Dev Tool) — The agent platform used for the entire demo, run as a VS Code extension; the official Anthropic extension you install in step two of the three-step setup. - **Visual Studio Code** (tool, Code Editor / IDE) — The editor that hosts the Claude Code extension; step one of the setup is downloading it, and its file-tree workspace becomes the shared context store. - **Wispr Flow** (tool, AI Dictation / Voice) — Voice-to-text app (transcribed as 'Whisper Flow') Yasin uses to talk to the agents; it cleans up dictation and formats text per app (email, Slack). - **Claude** (tool, AI Assistant) — The model/subscription powering the agent platform — Claude Opus 4.5 with a 200,000-token window; the $20/month plan runs everything demoed. - **ChatGPT** (tool, AI Assistant) — OpenAI's consumer app, the archetype of the browser chatbot most people mistake for 'AI'; GPT-5.2 is its underlying model. - **Gemini** (tool, AI Assistant) — Google's model (Gemini 3) cited for its 1,000,000-token window and 'nano banana' image generation — used to contrast bigger-vs-more-focused context windows. - **Salesforce** (tool, CRM) — CRM the sales-rep agent is told to update via MCP; in the demo the MCP isn't connected, but Yasin explains how to connect it so agents write to opportunities directly. - **Model Context Protocol (MCP)** (tool, AI Integration Protocol) — Model Context Protocol — how the agent platform connects to Salesforce, HubSpot, and Teamwork so you can act on your stack without opening the SaaS UI. - **HubSpot** (tool, CRM) — Named alongside Salesforce as an MCP the agent platform can connect to as part of the no-more-SaaS-UI endgame. - **Teamwork** (tool, Project Management) — LeanScale's project-management tool; Yasin's example of asking the agent, via MCP, to summarize project progress instead of logging into the app. - **GitHub** (tool, Developer / Code Hosting) — Starting to replace Google Drive at LeanScale as the shared repository for skills, files, and context that the whole team draws on. - **Google Drive** (tool, File Storage / Collaboration) — The team's legacy file store that the agent-platform file ecosystem (on GitHub) is beginning to replace. - **Microsoft PowerPoint** (tool, Presentation Software) — The marketing agent builds a branded PowerPoint from approved wireframes using a PowerPoint-maker skill; every element comes out as real, editable text rather than an image. - **Google Slides** (tool, Presentation Software) — Where the finished, on-brand deck lands — Yasin drags the PowerPoint in and gets a fully editable Google Slides version. - **Slack** (tool, Team Messaging) — One of the apps Wispr Flow detects and formats dictation for (asking whether to format a message as casual or professional), showing the tool works beyond the agent platform. ## FAQ **Q: What is an AI agent, and how is an agent platform different from a chatbot?** A: An AI agent is software that can plan and carry out multi-step work autonomously — reading files, launching sub-agents, researching, and producing outputs. An agent platform (like Claude Code running in VS Code) is a different product from the consumer chatbot most people call 'AI': only the agent platform can maintain a persistent to-do list, spin up sub-agents, read and write files on your machine, switch between plan and permission modes, queue messages, and compact its own context. The browser chatbot is built for general Q&A; the agent platform is built to get professional work done. **Q: How do you set up an AI agent platform, and what does it cost?** A: It's three steps and about five minutes. Download VS Code, install the official Claude Code extension from Anthropic (the building-blocks icon in VS Code), and log in with a $20-a-month Claude subscription. Restart VS Code, click the orange icon, authorize, and the agent is enabled. Everything demoed in the episode runs on the base $20 plan until you exhaust its tokens, at which point you're offered an upgrade. **Q: Which AI companies and models does the episode reference?** A: It names the three big players — OpenAI, Google, and Anthropic — and notes each ships three products: a model, a consumer app, and an agent platform. The models cited are OpenAI's GPT-5.2, Google's Gemini 3 (with a 1,000,000-token context window), and Anthropic's Claude Opus 4.5 (with a 200,000-token window). The demo itself runs on Claude Code. **Q: What is a token and a context window in AI?** A: A token is the atomic unit of how AI processes text — roughly three to four characters (the phrase 'hello how are you' is about six tokens). The context window is the model's finite working memory, holding all of your inputs and its outputs; for example, Claude 4.5 has a 200,000-token window and Gemini 3 has a 1,000,000-token window. Once the window fills up, the agent starts forgetting earlier context, which is why agent platforms show live usage and let you 'compact' — summarize the current context and hand it off to a fresh agent with a clean window. **Q: How can RevOps use an AI agent to check a forecast?** A: Give a purpose-built forecast-analyzer agent your pipeline, forecast export, and rep activity data, along with your red-flag criteria — deals that skipped stages or moved too fast, deals with no activity in two weeks, close dates pushed more than once, no logged customer meeting in 30 days, and single-threaded deals. The agent cross-references every deal against those flags and returns a realistic commit range (in the demo, $1.2M–$1.75M against a $2.3M submitted number), identifies the rep carrying dead deals, surfaces risk patterns, and drafts what to tell the board. **Q: What are AI agent 'skills'?** A: A skill is a folder of files that teaches an AI agent how to perform a specific task — making a PowerPoint, an SOP, a PDF, or wireframes. The major AI labs adopted a shared skills standard, so you can download skills from the internet or build your own, then point the agent at the skill's file path to execute it. Yasin compares it to Neo instantly learning kung fu in the Matrix: the skill gives the agent a new capability on demand. **Q: How do AI agents connect to tools like Salesforce and HubSpot?** A: Through MCP (Model Context Protocol), which lets the agent platform connect to external systems such as Salesforce, HubSpot, and Teamwork. Once connected, you can tell the agent to update a Salesforce opportunity, summarize a Teamwork project, or pull status from a tool without ever opening its interface — the basis for Yasin's goal of not logging into a SaaS UI again. **Q: Can AI agents really do work across every go-to-market function?** A: Yes — the episode demonstrates it live across five functions running concurrently: a sales rep recovering a tough call (analysis, recovery email, CRM notes), a sales manager judging a rep's coachability from patterns across five calls, marketing turning six prospect calls into a messaging verdict, a blog post, and a branded deck, customer success building a stakeholder map after a champion leaves, and RevOps pressure-testing a $2.3M forecast. The agents launch sub-agents, research the market, and write real files, though they still require an operator to supply context and architect the prompt. ## Timeline - **00:00** — What is an AI agent? Five GTM use cases - **00:54** — Sales rep: recover a tough Notion call - **03:28** — Sales manager: is the rep coachable? - **05:28** — Marketing: is our messaging landing? - **07:08** — Customer success: champion churn before renewal - **09:50** — RevOps: pressure-test a $2.3M forecast - **12:35** — The AI landscape: model vs. consumer app vs. agent platform - **14:25** — Features only agent platforms have - **14:54** — The three-step setup for $20 a month - **17:01** — Permission modes, queued messages, and Whisper Flow - **20:24** — Reviewing the sales-rep output - **24:28** — The sales-manager coachability assessment - **26:52** — Marketing output, then a blog post - **29:07** — Slide decks: skills and wireframes - **32:15** — Token windows and context compaction - **37:55** — The CS stakeholder map and RevOps forecast check - **42:24** — The finished, on-brand deck - **43:55** — The endgame: MCP and no more SaaS UIs - **44:50** — 2026: power-user vs. mainstream agent platforms - **46:32** — Adoption and the token leaderboard ## Related episodes - **Ep. 95: Why AI Means More RevOps Hires, Not Fewer** (Jimmy O'Halloran) — The strategic complement to this build-along: AI as a productivity multiplier for operators, not a replacement — the thesis behind arming GTM teams with agents. · https://leanscale-knowledge-hub.netlify.app/podcast/jimmy-ohalloran-new-relic-revops-consumption-revenue/ - **Ep. 85: Why AI + GTM Engineers Can't Replace RevOps** (Tessa Whittaker) — Agents can execute a strategist's judgment but not replace it — the human-in-the-loop counterpart to prompting 'like a strategist.' · https://leanscale-knowledge-hub.netlify.app/podcast/tessa-whittaker-ai-gtm-engineers-revops/ - **Ep. 88: Why AI Won't Close Your Biggest Deals** (Michael Kiernan) — A CRO's view of where AI fits versus where humans win — pairs with the practical limits and oversight this episode assumes. · https://leanscale-knowledge-hub.netlify.app/podcast/michael-kiernan-nextdoor-ai-wont-close-deals/ - **Ep. 6: Why Your Forecast Is Broken** (LeanScale) — Foundational forecasting episode that the RevOps forecast-analyzer agent operationalizes with pipeline red-flag rules. · https://leanscale-knowledge-hub.netlify.app/podcast/why-your-forecast-is-broken/ - **Ep. 15: Where Should RevOps Report?** (LeanScale) — Org-design context for centralizing AI and RevOps as the operating surface for the whole revenue engine. · https://leanscale-knowledge-hub.netlify.app/podcast/cameron-legge-where-revops-report/ - **Ep. 91: Why Outcome-Based Pricing Is a Trap for Most AI Companies** (Roee Hartuv) — The economics side of AI-in-GTM — how AI companies package and price — complements the tooling side shown here. · https://leanscale-knowledge-hub.netlify.app/podcast/roee-hartuv-outcome-based-pricing-trap/ ## Full transcript _Machine-transcribed and not diarized; speaker attribution is inferred._ _Transcript only, as a separate file: https://leanscale-knowledge-hub.netlify.app/podcast/ai-agents-for-every-go-to-market-role/transcript.md_ ### 00:00 — What is an AI agent? Five GTM use cases **[0:00]** There's a lot of hype around AI and go-to-market. What does it even mean to be using AI agents? What the heck even is an AI agent? Well, in this video, we're actually going to be going ahead and using AI agents in every function of go-to-market. In this video, we're going to be talking about how to be able to use AI agents as a sales rep. What do you do when a call goes sideways? How do you actually use an AI agent in order to help you with that? Using AI agents as a sales manager. Is a rep coachable or should you cut your losses? How to use AI agents inside of marketing, taking prospect **[0:27]** insights, turning those into a deck in your brand voice as well as your format, and then talking about using AI agents in customer success. Let's say, for example, you're in CS and you have a champion who leaves. Well, who do you talk to now and is the account still worth it? And last but not least, how to use AI agents in RevOps. Let's say we're going to go over a scenario. Is my $2.3 million forecast real or am I lying to the board? We're going to go ahead and go over each and every single one of these functions and how to use AI agents in every single one. And while those AI ### 00:54 — Sales rep: recover a tough Notion call **[0:54]** agents are working, because yes, they are actually going to be working live inside of this video, we're actually going to be covering the landscape of AI agents in 2026 as it stands right now. And so that way you can start to understand how to use AI agents today. So let's go ahead and kick off some of these AI agents. Let's go over the first use case. Let's talk about how do you use AI agents as a sales rep. So let's say, for example, I just got off of a tough call with one of my accounts. Let's say the account is Notion. So I'm going to go ahead and talk to my AI agent and give **[1:21]** it all of the context so that way it can tell me how I can turn the call around. So I'm going to go ahead and fire up WhisperFlow over here, which is an AI in order to talk to AI. And I'm going to say, hey, so I just got off of a tough call with Notion. Here's the transcript of that call. And then what I'm going to do is I'm going to go ahead and feed it the transcript. And don't worry, inside of this video, we're going to walk over what this platform is, how do we go over that, et cetera. But I want to kick off these agents so that way you can understand how all of this works. And then we'll **[1:47]** go ahead and continue talking to the agent. And that call didn't go as well as I wanted it. They pushed too hard on pricing, and I really fumbled the timeline objection. So here's what I need you to do. First, I want you to go ahead and pull in my sales playbook as well as my objection handling tactics from the agent knowledge folder. And then what I'm going to do is I'm going to go ahead and give it that agent knowledge folder so that way it has that context right now. And then I'm going to go ahead and continue talking to it. So I'm going to go ahead and say, and then, Mr. Agent, **[2:13]** what I want you to do is I actually want you to go ahead and launch a couple sub-agents in order to research what Notion has announced recently. Has there been any news in the last 30 days that might explain why they might have went ahead and pushed back in the sales call? And then while that sub-agent is running, what I want you to do is I want you to go ahead and analyze the transcript. What should I have said to them? Where did I lose them? What should I have said differently? As well as, once you have that context, I want you to make three files for me. I want you to make a call analysis **[2:40]** file in terms of telling me what went wrong and what should I have done instead of that. I want you to make a recovery email for me that's going to address their real concerns. And then I also want you to go ahead and draft for me what the notes are going to be so I can put them into Salesforce, what is going to be the honest version so that way I can communicate to the rest of the team. And then we're going to go ahead and continue talking to it. And then, hey, Mr. Agent, once you have all of that context, what I want you to do is I want you to track your progress with a to-do list **[3:04]** so that way I can see what you're working on. And then once all of that is done, I want you to go ahead and connect to Salesforce via MCP. And I want you to update the opportunity record with those notes from the CRM summary. And I want you to think through this like a strategist. Okay, so what you're seeing here is quite a few things. And again, don't worry, we're going to break it down inside of this video. But I want to walk you through how we've gone ahead and architected this prompt as a sales rep. I'm going to go ahead and launch that right now. Next up, ### 03:28 — Sales manager: is the rep coachable? **[3:28]** while that's running, we're going to go ahead and actually launch what that would look like for a sales manager. So you'll see I have some demo files here on the left side in terms of the sales manager. Let's go ahead and build the prompt for what the sales manager would say. Hey, so I need your help figuring out what's going on with one of my reps. His name is Michael and here's his rep profile. And then what I'm going to do is I'm going to go ahead and feed it the rep profile right over here. So I have agent knowledge, I have the rep profile, I'm going to copy path, **[3:53]** put that right there. And then we'll go ahead and continue talking to the agent. And then I'm going to go ahead and give you our sales framework so that way you can understand what really good sales calls look like for us. And then while that's loading right there, I'm going to go ahead and feed it what our sales framework looks like, boom, just like that. Okay. And don't worry, we're going to open up each of these files later on in this video and talk about how we've gone ahead and architected this AI system right now. And then we'll go ahead and talk about it and say, Hey, so, and then I'm **[4:15]** also feed you Michael's last five transcripts. They're going to be located inside of this folder that I'm about to give you right now as well. And then from there, we're going to go ahead and give them the transcripts for the calls. And then we're going to continue talking to it. And I need you to analyze all five of these calls. And I don't want you to summarize each one separately. I want you to actually look at the pattern for each of them. And then actually what I want you to do is I actually want you to spin up a sub agent for yourself in order to actually think through each of these **[4:42]** transcripts in parallel. So that way you can pull the key moments in which Michael was actually struggling because that'll be faster for you to do that rather than reading them separately. And I'm going to go ahead and put that in the prompt. So you see how I'm basically architecting this prompt together. And then agent, what I need you to do is I want you to actually look at what are his consistent weaknesses, not the one-off mistakes, the actual pattern. Is he getting better? Is he stuck? What are his worst moments over the last two weeks? And I want you to really rate him **[5:05]** on each of these dimensions using our sales framework. But I want you to be honest with me. I've been coaching Michael for the last three months now. And I need to know, should I keep investing in him or should we actually have a different conversation? And oh yeah, in order to be token efficient, I don't want you to just dump all of the transcript on me. I actually want you to go ahead and pull relevant quotes that are going to show the pattern. And I want you to track your progress using a to-do list so I can see what you're working on. And then I want you to save all ### 05:28 — Marketing: is our messaging landing? **[5:28]** of that in the root folder as a file called Michael assessment. And let me know, is he coachable? Give me a 30 day plan. And if he's not, then tell me what that real conversation needs to be and map out the outline for that conversation. Think this through with me like a thought partner. And so that's the use case for a sales manager. Let's go ahead and get this agent chugging. So I'm going to go ahead and send that along its way. Okay, cool. So we now got our two agents who are working in the background. Let's go ahead and talk about the marketing use case. So let's say, for example, **[5:55]** marketing is trying to decide whether or not they should change their messaging. And so you'll see on the demo files here on the left side, what I have is I have an agent knowledge folder. And inside of this, we have our competitor positioning, our current messaging, as well as our target personas. And so I'm just going to go over, go ahead and go over to AI. And I'm going to say, hey, I so I've got six prospect calls over the last month, they're a mix of wins and losses, stall deals, and I want to know what's actually landing with our messaging. So what I'm going to **[6:20]** give you is I'm going to give you a folder that's going to have the transcript of each of those six prospect calls, as well as another folder that's going to have all of our current messaging. And what I want you to do is I want you to actually create from a document inside of the same folder. So you're going to go ahead and write the document, put it in there. And you're going to go ahead and tell me what's working with our messaging and what's not. And then after this, we're going to go ahead and write a blog post. And then after that, we're going to go ahead and make wireframes **[6:45]** and a slide deck based off of that new messaging. You don't have to worry about those later phases right now. Let's go ahead and first focus on the messaging piece of it. And then from there, all I got to do is I go ahead and go over to agent knowledge, I right click, copy path, put that there. And then I go ahead and give it the prospect calls as well. Boom, shocka locka. And then I go ahead and send that agent on its way. All right, we're not done yet. Now we got to go ahead and go over to the next function. So we've just launched the marketing agent that's ### 07:08 — Customer success: champion churn before renewal **[7:08]** working on behind the scenes. You'll see, boom, boom, boom, it's working behind the scenes. We got our sales manager agent going right now, as well as our sales rep agent. We're now going to go ahead and go over to the customer success use case. And so again, I'm just going to launch Whisper Flow over here. I'm going to go ahead and talk to my agent. Hey, so I just found out that my champion at Amplitude just left Sarah and she was my main point of contact over the last 18 months. And she recently got recruited by a competitor and there was no handoff. And so I just found out because **[7:35]** her email bounced. And so what I want you to do is I'm going to go ahead and give you the account history. And so that way you know everything about the relationship. And then I'm going to go ahead and give it the actual account history from our CRM. So I'm going to go ahead and go over to our demo folder here. You'll see that I have right here, the account history. And then we'll go ahead and right click on that. And this folder has all of the account history. So you'll see we have pass calls. We have a Salesforce export right over here. I'm going to go ahead and give it that context. **[7:59]** And then from there, I'm just going ahead and continue talking to it. And here is our current playbook for when champions leave. And so you'll see what I've gone ahead and done is prior to this entire demo, we went ahead and created a playbook on whenever a champion leaves, we have the context of what that looks like. And this is what that looks like right over here. So you'll see we just have a bunch of context for the AI in order to work through. As well as here is the current status of this account inside of our CRM. And then I'm going to go ahead and give it the Salesforce **[8:32]** export as well from there. And then right over here, we're going to save. And then also I have for you inside of that folder that I just gave you all of the pass calls with this account that are located inside of that folder as well. And so what I've got is I've got six weeks until the renewal and I have no idea what I'm dealing with right now. So here's what I need you to do. I need you to actually spin up a bunch of sub agents in order to research what is going on with amplitude's current leadership. What is Sarah's old role as well as what was their background and what she **[8:59]** was doing inside of amplitude. And then who is the person who replaced them and what they actually care about inside of that new role. And then while that sub agent is running, I want you to actually go through all of those pass calls and I want you to pull out who else was on the calls besides Sarah. What was the concerns that they raised as well as were there any signals that were located inside of the transcript about the internal politics and what might be going on with budget pressure at amplitude. And then I go ahead and send that. And then last but not least, I go ahead and **[9:26]** tell it, hey, I want you at the end of this to actually make me a stakeholder map. Who are the three to four people who I need to actually engage with. I want you to actually create a to-do list for yourself so that way you can stay on track over a long period of time. And then I also want you to go ahead and be token efficient. So don't just dump all of the transcripts on me. What I want you to do is I want you to actually go in and make a file for yourself inside of the folder where all of these documents are located. Title that the stakeholder map so that way I can look ### 09:50 — RevOps: pressure-test a $2.3M forecast **[9:50]** at it and then go ahead and be honest with me. If this account is likely to turn, go ahead and let me know about that as well because I only have six weeks and I want to know if this is actually a lost cause. I then go ahead and send that. And then just like that, we now have our CS agent, which is off to the races. Okay. So now we got our other agents who are working. Let's go ahead and launch the RevOps agent. So I'm going to go ahead and talk to my agent over here. Hey, Mr. Agent, I need a sanity check. So I got this forecast call with the board meeting on Thursday **[10:17]** and I want to submit a $2.3 million commit for Q1. And I need to know if that number is actually real. And so what I'm going to do is I'm going to go ahead and feed you what are the red flags that we typically watch out for whenever we are auditing pipeline. I'm also going to go ahead and feed you what our current pipeline looks like, as well as a Q1 forecast export, as well as what our actual reps have been doing and what that activity data looks like. And then what I need you to do is I want you to cross reference each of these based off of flags that the calls and the deals are not **[10:46]** going to go through. So I want you to look for deals that have skipped stages or moved too fast. I want you to look for deals with no activity in the last two weeks. I want you to look for deals where the close date was pushed more than once. I also want you to look for deals where the rep hasn't logged actual activity in a customer meeting in the last 30 days as well as single threaded deals with only one contact. And again, just for the video, I have known as a RevOps manager that if I look at any of those flags, that's going to tell me that the forecast isn't as accurate as **[11:14]** it actually needs to be. And then I'm going to go ahead and talk back to the agent. And then I want you to go ahead and actually put that in a report for me inside of the folder that all of these files are in. And then I also want you to make a massive comprehensive to do this for yourself to make sure that you are thinking through every single aspect of my prompt. And then at the end, I want you to give me an honest take. What's the realistic number that I can report back to the board? What is actually compared to what I've submitted? And then are there any patterns that you're **[11:39]** noticing in terms of certain stages, deal size or time and stage that correlate with weak deals that I can then go ahead and update our SOPs and our understanding of our business as well as what's going to tell us around if deals are going to close or not. And then tell me at the end of the day, what should I go ahead and tell the board on Thursday? And then I go ahead and send that. And then all I got to do from there is I go ahead and go over to my file. And you'll see I have the pipeline data right over here. I go ahead and give it the folder right there. And then I go ahead and **[12:08]** give it our agent knowledge right over here. And then for this specific one, I actually have a forecast analyzer agent that works really well for this as well, which is something that I've gone ahead and created before. And this agent is purpose built. You'll see there's an entire SOP for this that it walks its way through actually analyzing the forecast. So what I'm going to do is I'm going to go ahead and actually copy path and give that to this agent as well. So that way understands what it needs to do. And I'm going to go ahead and send that RevOps agent on its way. ### 12:35 — The AI landscape: model vs. consumer app vs. agent platform **[12:35]** All right. So while those agents are actually running, let's go ahead and actually zoom out and talk about the landscape of the market. And let's talk about what that platform was that I was actually inside of. What are agent platforms? What are the capabilities and the features that agents have in GoToMarket that are not going to be something that you can use in a normal chat or a normal AI interface? And so really what's going on is that there are three big players in the AI space right now. Obviously you're familiar with it, OpenAI, Google, and Entrop. Now there are three **[13:01]** different products that each of these companies have. The first product is going to be the model. And the model is the baseline infrastructure, right? So you think of Gemini 3, OpenAI just launched their GPT 5.2. And then Claude has Opus 4.5 that just got released a couple weeks ago. After that, the second product is the consumer app. Now the consumer app is what typically people think of when they think of AI. And that was the case anywhere between 2021 all the way, probably until halfway through 2025, in which all of the way that we would interact with AI was actually through the **[13:32]** browser. Now what's happening is that we're going to talk about the agent platform setup that has really unleashed this entire dam of what's happening with agents going forward. But really that's the first thing that's a light bulb moment for anybody that I talk to when it comes to the AI landscape is that when you think of AI and you think of the consumer app and you think of the browser, you want to understand that that is primarily for the general use case and kind of like what general population might use for Q&A, kind of like the therapist, the health advice, **[13:59]** all of that stuff. Okay, so that's going to be the consumer app. And again, that's what most people think is actual AI. However, what's going on now is that there's this entire different product, which is what you saw me inside of, which is the agent platform. And the way that I describe it is that the agent platform is for professionals, it's actually to get work done. And now there's two different parts of the agent platform, there's the mainstream users, as well as the power users, and we'll go over the division between both of them. Because as of right now, the week that I'm ### 14:25 — Features only agent platforms have **[14:25]** recording this video, there's an entirely new product that has been released in order to use agent platforms for the mainstream. And we'll talk about that more in just a second as well. Now, the main thing that you need to know about the difference between consumer apps and the agent platform is that there are actually a number of features that you saw me demo as we were going ahead and launching those agents that are not available inside of the consumer app. So the ability to actually make a comprehensive to do list for the agent that allows it to be able to work for 10, 20, 30, 40 minutes, I've actually had agents that have worked for four hours ### 14:54 — The three-step setup for $20 a month **[14:54]** asynchronously on a task, the ability to actually launch sub agents is a feature of an agent platform, the ability to be able to feed it context and then have it actually write files for you on your local computer. All of that is a feature set that only an agent platform has that the consumer app does not. Now, how do you actually download these agent platforms? How do you get started? It'll take you five minutes and three steps. I'm going to walk you through it right now. So what you do, we still use Google for some stuff. We go to VS code download. Okay, so you just go over to Google **[15:25]** and type in VS code download, you go ahead and click on that, and you go ahead and download whatever version works best for you. Okay, once that downloads, you're going to go ahead and put that into your application folder. And then you're going to go ahead and open up VS code. What you're going to do is when you have VS code opened up, you're going to go ahead and click on the icon at the top that looks like building blocks. And you're going to go ahead and download the Claude code extension. Literally, that's it. That's step number two, you go ahead and you download the **[15:48]** Claude code extension is the official one from anthropic. Okay, once you go ahead and download that, you're going to have a icon that shows up on the top right that looks like an orange cloud icon. Okay, what you then do is you go ahead and click Claude AI subscription, it's going to then actually match up to your Claude subscription, you only really need the $20 a month plan in order to be successful with this. And then once you run out of tokens on the $20 a month plan, I'll give you the option or upgrade. But literally, you can run this entire system and everything that I just **[16:12]** showed you for $20 a month. How insane is that? And it's literally three steps. Download VS code, download the extension for Claude code inside of VS code. And then you go ahead and log in with your Claude subscription. And then that's it. And then from there, you just go ahead and restart VS code, you hit the orange icon. And then what you do is you go ahead and click on Claude subscription. From there, it's going to go ahead and open up into another tab, it'll go ahead and ask you in order to authorize. Once you go ahead and authorize, you go back over to VS code, and then **[16:36]** boom, just like that, you're going to have the AI agent enabled. And then from there, you can have all of your files. And you can see how this really starts to become an ecosystem of files that your entire team can then share. And for us at Lean Scale, it's actually starting to replace Google Drive for us. And so we'll have another video later on talking about exactly how we've created this ecosystem and how we're actually using GitHub as our replacement for Google Drive in order to share all of these skills, these files, this context, and then using all this in order to ### 17:01 — Permission modes, queued messages, and Whisper Flow **[17:01]** connect to the MCPs as well. Let's talk about some of the big features that you saw me demo as we were going ahead and launching the agent. So you'll see that I am right now inside of VS code with the Claude code extension, which is the agent platform. And let me go ahead and open up another agent here. So you can see now, I know that this does look like the consumer app in which it has kind of a chat interface. But there's a couple big things that are different here that the agent platform has that the consumer app does not. The first one is going to be the ability to actually **[17:27]** have permission mode. So you'll see as I'm switching through this, on the bottom left, I can actually put the agent in a different mode. Okay. And the reason why this is important is because you can control a little bit of the mindset that the agent is in. So for example, when it's in plan mode, what you're going to notice is that it's actually going to step back and think through things. And then what it does is when it's in plan mode, it actually makes a plan file behind the scenes and ask you, hey, let's go back and forth. Let's optimize this scoping document. And then it'll ask you for approval. Does this plan file look good? And **[17:54]** then it'll actually go ahead and put itself into bypass permissions mode or ask before edits and actually execute the plan. Again, this is a feature set that the consumer app does not have, only the agent platforms have. The other piece that you're going to go ahead and notice about the agent platform as I go ahead and put it into bypass permissions mode is going to be if I say, hi, how are you? Okay. I can then go ahead and while it's working, say another thing. I can go ahead and say another thing. I can go ahead and say another thing, right? Right here, you can see, **[18:20]** you can queue up messages. And what it'll do is as it's working, it'll take into consideration each of the messages that you're going ahead and putting in there. Now, with that being said, let's actually go ahead and talk about the agents that we've launched. So that way I can really start to demo for you some of the prompts that we were going ahead and putting together. So the first thing is that you saw me that I was actually talking to the AI agent. Now, the way I'm doing that is using another application. That application is called Whisper Flow. This is what it looks **[18:43]** like. Highly recommended because all you got to do is you just click one button. And then what it does is actually cleans up the transcript that you're going ahead and giving into the computer. Now, the cool thing is that this Whisper Flow technology works outside of just an agent platform. And so you can actually use it in Slack and emails, and then it actually detects what app you're inside of. And then if you talk, it'll actually format that text inside of an email. It'll go ahead and ask you, do you want to format this as a Slack message, casual, professional? It's a really **[19:09]** cool AI app that's completely separate than the agent platform, but very, very useful when you are using agent platforms in order to talk to them. And you'll see, just like that, it actually went ahead and cleaned up the transcript for what I was just saying. Now, in terms of the actual prompt, right? So you'll see, this was what we did when we were talking about the sales rep prompt. And a couple of things that you'll notice on here that I was able to do. First off is that with the agent platform, you're able to have all of your files here on the left side. And so for the sake **[19:34]** of this video and for the sake of this demo, I went ahead and actually set up each of these folders based off of what was in there, right? And so you can see for the agent knowledge for the sales rep, we have this B2B playbook that I can open up right here. So you see, this is our B2B sales playbook. We have our objection handling documentation, which I was able to go ahead and feed to it. We also have our product overview. So what you're noticing is that with the agent platform, I'm able to really just create files. And the cool thing is you can actually use the AI agents in order to **[20:00]** create those files using your voice. You just say, Hey, I want to create this file about our product overview. Here's a couple of things about our product. Here's our website. Go ahead and make from your product overview. Super simple, right? And so what you have is you have this agent knowledge database right over here. And then sure, there's like the basic ABC use case about literally just asking it, Hey, can you tell me like what's going on with our objection handling, like asking questions about that? But that's like ABCs. What you saw me use was operationalizing ### 20:24 — Reviewing the sales-rep output **[20:24]** that into a prompt that the agent was actually able to do work for me. And so what I was able to feed it was not only the transcripts, you'll see I had the notion called transcript right over here. And then you'll see that this agent actually finished and I'll walk you through exactly what it did. And then we can actually go through its entire thought process. You'll see, I hear you, tough calls happen and it's smart to debrief properly. And then what it did is it read the files. It went ahead and made a to-do list for itself because remember we're talking about right **[20:47]** over here. We talk about what are some of the features that agent platforms have that consumer apps are not is the ability to make an internal to-do list for itself so it can maintain intent over a long period of time. And then you'll see what it did is it went ahead and actually launched a sub agent. And now this is the other cool thing that agent platforms are able to do. So what happens is that you have the main agent and then what it does is it launches all of these sub agents that are actually going out and doing different tasks and then reporting back to the **[21:12]** main agent. And so in this case, all because in my prompt I told it to actually launch a sub agent, what it did is it went ahead and launched the sub agent and the sub agent was researching whether or not Notion had announced any layoffs, any restructuring, and then returning that as a summary. So you'll see it went ahead and launched the sub agent. You'll see that running the research agent in the background, let me go ahead and analyze this call against your playbook. This is where it gets real. So then it said updated to-dos. Yeah, AI is still a little bit hypey on there, right? **[21:39]** And then you see it gives us the call analysis. It went ahead and compared all of that. And then it went ahead and wrote all of that, creating the call analysis files, boom, boom, boom, boom, boom. And then it says, I don't see the Salesforce MCP. So for the sake of this video, I don't have the Salesforce MCP connected. However, you can go ahead and connect your Salesforce MCP to the agent platform. And then it'll actually go into your Salesforce and update things from there. Super easy. You just launch another agent and you tell it, hook up to my Salesforce MCP, and it'll walk you **[22:06]** through all of the steps in order to do so. And then last but not least, you'll see, remember, this is the sales rep use case. What it did is it said, hey, everything is ready. Here's a summary. And then we have all of the documents that it created. And so right now we just go ahead and go over to the left side. We go over to Notion call analysis. I'm going to go ahead and open that up. I'll go ahead and open up this table of contents on the right side. And so you can see it gave me executive summary. Hey, the call derailed in the first 25 seconds when you abandoned discovery **[22:32]** in favor of a feature demo. Boom. What went wrong? And I was going ahead and actually codifying for me every single part of the transcript and parlaying that with our playbooks and letting the rep know exactly where they lost the call. And again, you'll see that the power of this is because we're able to feed the LLM context and actually make it specific to us. This is the big difference between just using raw intelligence of AI inside of a browser and then actually feeding it the context that's going to allow it to be personal for each and every single department. Then here's a cool **[23:03]** part. It actually went ahead and appended all of the research from the sub agent inside of this as well. So you'll see it says market context that you did not have research shows that why your 60 to 75 K ask did not hit. And then it went ahead and actually he told us what is the news that is going on with Notion inside of the market as to what's going on in the broader ecosystem to let us know why the call was not successful. From there you'll see again I'm not going to go through this entire document but all of that is there for the sales rep in order to go through. From there we actually **[23:33]** then also have a recovery email. So you'll see this is the recovery email for Notion. You'll see it goes ahead and puts that for the stakeholders and then it breaks down what that email might look like. Now here's a cool thing. I could actually come back to this and say right over here hey so I really love the way that you wrote that email. I want to actually make an SOP so that way we can actually start to take all of my calls in the future and write the email in the same way. So what I want you to do is I want you to go ahead and make an SOP for yourself using this SOP on **[24:02]** how to make SOPs and I want you to go ahead and save that inside of the folder as well. Now what you just noticed is now we're able to compound on all of that learning and we can go ahead and say okay cool we're now going to save this as a separate file and really start to use this as an SOP going forward. And then we can go ahead and launch that right over there and the agent can get busy actually making for me a custom SOP based off of the input. Now I could have actually put into the prompt hey I love this in particular about the email I want you to really ### 24:28 — The sales-manager coachability assessment **[24:28]** emphasize it in the SOP and it will actually go ahead and save that as an SOP and then guess what in the future I can just right click copy path and tell it hey here's the SOP on how to make emails for me the way that I like it go ahead and format an email for me on that. Okay so that's the sales rep use case. Now let's go ahead and actually talk about the sales manager use case. So if you recall we had one of our reps Michael who wasn't performing and we went ahead and gave it a bunch of context about what that would look like and so you'll see again it made a to-do list **[24:55]** for itself it's reading Michael's rep profile it's reading our sales framework it's analyzing the five calls and then it's identifying what the weakness patterns are it's rating Michael on the framework dimensions determining the coachability and it's going ahead and saving Michael's assessment inside of the root folder. So you'll see it went ahead and worked probably for about five ten minutes and this is where you got to have some realistic expectations because if you expected a human to do this it would take them some time right and so the agent does work **[25:21]** autonomously behind the scenes for that period of time and then it went ahead and gave us a summary here at the end and then we can actually go ahead and open up the entire assessment right over here but real quick let me open the summary you'll see Michael is conditionally coachable but you need to change your approach here's what I found across all five calls the pattern that won't break the pitch first instinct that you flagged in November is still there how does it know that because remember we went ahead and gave it some information about my coaching sessions with Michael **[25:51]** and so it's going ahead and breaking down in every call he heard Sarah say our CRM is a mess and interrupted her to pitch he lasted eight seconds on velocity before launching into product and so you'll see it's going ahead three months of feedback hasn't fixed this that's the signal it went ahead and gave me his aggregate scores it went ahead and gave me the exception on the meridian call apparently he did pretty well and then it went ahead and gave us its recommendation as well as what's your read on this do you want me to go ahead and dig any deeper now I could literally **[26:19]** just go back and forth with it here and say you know what no actually you know I think you might have over indexed on xyz this is where I want you to change things boom and now I can actually have a conversation because it has all of that context I went ahead and gave it and then if we go ahead and open up Michael's performance assessment boom ten year eight months as an AE year-to-day quota 62 percent executive summary and you'll see this entire breakdown in terms of an analysis for this sales rep has been generated custom based off of our own context how insane is that so that's the ### 26:52 — Marketing output, then a blog post **[26:52]** sales manager use case let's now go ahead and go over to our third use case which was the marketing use case so if you recall what we did for this one is we said we had six prospect calls over the last month and we had documentation that we fed it about our current messaging and we went ahead and fed it all of that right and now you'll see what it did is it went ahead and created a messaging analysis file inside of that same folder and what it's doing is it's giving us a summary of the analysis that it found so I'm going to open up that document right over here and you'll see **[27:22]** messaging analysis what's landing and what's not and you'll see what it says is based on the six prospect calls two wins one losses one stalled and two active here's the executive summary here's what's working so your current messaging is implementation in weeks not months the verdict is to keep and amplify it you'll see from there you'll see usage based pricing focused on product explainable ai is giving us hey this is strong this is the sleeper hit right here right this is telling us hey your current messaging is one platform versus five tools and it pulled that **[27:50]** from our current messaging file over here and it's letting us know this is partially working but what you're saying it is you're saying it wrong the one platform positioning sounds like you're trying to do everything but prospects actually respond to the opposite a focus tool that does one thing well and it's going ahead and giving you the citations from the transcripts about what the prospects have said around that messaging piece how wild is that from there it's going to tell us what's not working as well so you'll see it goes ahead and give us a breakdown on here and then it gives us recommendations around what the messaging changes need to be **[28:19]** now here's the cool thing i can go back to that agent that we have over here okay let me go ahead and open up that agent which was this one and i'm going to go ahead and tell it hey i love this i actually want to go ahead and turn this into a blog post what i'm going to go ahead and do is i'm going to go ahead and give you a little bit more information so you can understand what our brand tone is so i'm going to go ahead and feed you the file right over here and i want you to go ahead and write for me a blog post in our brand tone and go ahead and do that right now and save that **[28:45]** inside of the blog post folder as well for me now of course i could have gone ahead and given it some more context here as well in terms of the length i could have given us some more files but this is just a demo i want to show you what's happening here right and of course if i go ahead and open up this file you'll see this is something that we already used a different agent in order to prepare this and it goes ahead and gives us what is the brand tone how we should actually write what are the voice characteristics do's and don't examples and now i can literally just with the ### 29:07 — Slide decks: skills and wireframes **[29:07]** click of a button feed that into the agent and the agent is now literally able to take all of that context into consideration and write for me that blog post and now of course if we continue this demo what we can also do is we can tell it to go ahead and make an entire deck a slide deck for us based off of that blog post and it's literally just as simple as i come over here and i just go ahead and copy not only the sop that we have on how to make powerpoints but you'll see if i have this folder right here called skills and sop's there is a powerpoint maker agent all i do is **[29:35]** i right click that copy path and i put that right over here now these agents are known as skills we probably have a completely separate video on this but entropic and all of the ai companies right now have adopted the standard card skills and so you can actually go onto the internet and download agent skills and it's really cool because it's kind of like neo in the matrix where he was like i don't know kung fu and then he's like nope boom now you know kung fu that's what skills are to ai agents and those are literally live right now you can go onto the internet say **[30:00]** hey i want to have a powerpoint maker skill i want to have an sop maker skill i want to have a pdf maker skill and inside of these skills inside of these folders is all of the files needed for the ai to know how to do that task and all i got to do is i right click copy path here make this into a slide deck for me and it'll literally go out there and make it into a slide deck and just to show you what i mean by this just because i think this is a really cool use case i'm actually going to first before i tell it to make slides i'm going to tell it to make some wire frames for me inside **[30:28]** of the chat so i can approve what the layout of those slides is going to be so i'm going to talk to it hey so i actually want you to go ahead and create for me in the chat right now some wire frames because we're going to actually make this into a slide deck but before we make it into a slide deck i want to see what your creative vision is to pick some of the hardest hitting points from this blog post that you just wrote and put inside of the chat for me what the wire frames for that would actually look like and you'll notice what i did ahead of time is i actually had an sop that was **[30:55]** pre-created because we've done these wire frames a bunch of times here at lead scale i already have an sop a skill if you want to call it that in order for the llm for the ai agent to know how to actually create wireframes i'm going to go ahead and send it that right there right and then once those wire frames are showing up inside of the chat boom it's going to read the sop it's going to create the wire frames inside of the chat and then i can actually go in and approve those wire frames look at this look at this live right now okay wire frames what your prospects actually want i'm **[31:21]** pulling in high impact moments here's the creative version and then i can look at these slides as if they are visuals on a slide right now and i can approve them right so why tech corp signed three weeks over here why meridian walked you guys had a good product it wasn't about the product robert the cro lost deal and so this is something that i can now literally create right here look at this right here ai credibility crisis revenue intelligence means nothing mike why should i trust your ai sarah right so it's pulling from the transcripts it's going ahead and making those **[31:51]** wire frames now if i wanted to go and say right here hey you know i love these wire frames but i want to go ahead and tweak wire frame number slide number seven and you know instead of saying this do this instead show me what that would look like it would actually show what those wire frames look like right over here and then guess what what i can now do is once i approve these wire frames i can go ahead and actually give it the sop on how to make a powerpoint i'm going to do that live with you right now so that we can really understand how amazing and powerful agent ### 32:15 — Token windows and context compaction **[32:15]** platforms are but before i do that i want to introduce you to another very important feature that is all about agent platforms which is token management and so if you don't know what tokens are let me explain that very quickly now this is a huge piece of using ai agencies understanding that every ai model has a token window what does it mean a token window and what the heck is a token in the first place so a token is the atomic unit of how ai thinks and so one token is approximately three to four characters and so when you type in hello how are you that's about six tokens now **[32:47]** the reason why that's important is because every single llm every single ai has a token window you can kind of think about it like it's working memory all the stuff that it can remember at any given moment now clod latest model 4.5 has a 200 000 token window gemini 3 has a 1 million token window now that does not mean that more is better because the more token window that you have the less it's able to pay attention to the details but the more is able to ingest information and summarize it as opposed to a 200 000 token window which is like what we're using today allows it to focus on the details and it's a really good sweet spot however the reason **[33:21]** why this is important is because all of your inputs to the ai are included in that 200 000 token window and so what that means is that when you have let's say your user message okay let's just hypothetically say that was 10 000 tokens right all of the prompts that we went ahead and gave it but then that also includes out of that 200 000 right because we only have 200 000 right over here out of that 200 000 the output message is also included so let's say for example that cost an additional 10 000 tokens okay and when i say cost it doesn't mean that actually **[33:48]** charged you those tokens because we can get into the building side of it completely separately but what you means is that it took away from your 200 000 token window in terms of that working memory and so what you now have is you have 20 000 tokens that have been used up there well what you do then is you then send it another message okay and then you went ahead and sent it it went ahead and outputted another message from there and then you sent another message and you go back and forth let's say you got over here and you're like okay this is now going to be like 190 000 tokens **[34:12]** that are used well i only have 10 000 tokens left well what's going to happen is that when you get to that 200 000 limit it's going to start to forget what happened before that and so that's the reason why when you're going back and forth through chat and it's like okay going going going going going you're like whoa wait a second it just completely lost track of what we were doing it's because it went ahead and hit its token limit now why is that important when we are using agent platforms is because in the agent platform itself it tells us when we are going to be running into **[34:38]** that token window and this is a huge difference between again that consumer app facing side of things as well as the power user agent side of things is because if you look at the bottom of the screen right over here it says that we've used 73 of our context window and it says we have 27 of context remaining now inside of the agent platform what we can do is we can actually go ahead and compact what does compact means it means that if i go ahead and click this it's going to create a summary for itself in order to hand off to the next agent and that next agent is going to **[35:06]** come up with its own fresh 200 000 token window now the reason why that's important is because right now i'm going to go ahead and tell it to make a powerpoint slide for me now a powerpoint slide in an entire deck is going to take probably a bunch of tokens and i'm probably not going to be able to do it in 27 of the token window so what i'm going to do is i'm actually going to go ahead and click in order to auto-compact and the thing is that this agent is almost like when somebody ends their shift and they have to kind of hand off the notes to the next employee who's going to start that's **[35:33]** exactly what's happening right now this agent we're saying hey you're done you had 27 of your shift left i got a really big project i want it i want you to be fresh so let's go ahead and compact create a summary and i want you to go ahead and hand that off to the next agent and you're going to see in just a second it's going to go ahead and give us that summary and now we're going to go ahead and start with a new token window and so just like that you'll see what it's gone ahead and done is it's gone ahead and actually created a summary for the next agent and you'll see that **[35:57]** that indicator actually disappeared and so this is what i think a lot of people don't really understand about using ai is that as you're starting to use it for a work context these are some of the intricacies that you have to keep in mind and it really is a new paradigm of working it's a completely different way in order to interact with work in which you're using agents but it requires for you to understand the limitations of ai as well as the agents and so you'll see right over here went ahead and gave us a summary it's going ahead and handing that summary off to the next agent from here we now have a new token window that i'm now going to **[36:26]** go ahead and tell it to okay so here's a funny uh bloopers for you guys so what happened was we told the last agent to go off of off of duty right we told him hey you're done but in the summary okay it didn't have the wireframe so what i should have done is i should have actually told the last agent to save those wireframes inside of a document and then the next agent could actually pull from that document and so we're going to have to redo this over here i'm going to go ahead and say hey so just just so we don't lose these wireframes i actually want you to go ahead and save these **[36:53]** wireframes inside of a new document inside of the wireframes folder so that way if you do end up running out of context we don't lose all of this valuable wire framework that we can just build into a slide deck and so just like that you'll see that we have right now and it gave us the wireframes prospect messaging deck if i go ahead and open up that file right here you'll see that we have all of our wireframes and so now i know i'm not going to lose that information now what i can do is i can go ahead and say cool i'm going to give it this skill right here i also **[37:22]** have another skill for powerpoint that i'm going to feed it right over here and i'm going to say cool so here are two context skills inside of the wireframe to powerpoint skill is actually our lean scale branding and our colors and so i want you to actually go ahead and create these wireframes into slides and you can save those slides inside of my marketing deck folder and then i want you to go ahead and get cranking on that now and then just like that we go ahead and send the agent on its merry way and it is going to go ahead and start cranking on making those wireframes into ### 37:55 — The CS stakeholder map and RevOps forecast check **[37:55]** a powerpoint and i'll show you what the output of that looks like in just a second but wait there's more we're not done yet remember we were actually talking to the cs the csm right the csm was talking about how they had a champion that had recently left and so a couple things that you'll notice about this prompt was that i actually told it to go ahead and launch some explore agents for itself i gave it a lot of context i also told it to make a to-do list for itself right and so you'll see what it did is cool i read the files it went ahead and read the files around the playbook **[38:25]** it went ahead and launched a sub agent for itself in order to research amplitudes leadership and then it went ahead and read all of the other files excellent i have the four transcripts good the research agent is actually running the searches while that works let me go ahead and synthesize what i've learned from the transcripts and create the stakeholder map so went ahead and gone and created those has a stakeholder map boom boom boom once the sub agent went ahead and finished it says the research agent is making good progress but still working okay the research agent is **[38:54]** wrapping up it's cool and then boom okay previous research boom the research agent is complete and now is appending all of the research agent's work into that document and it went ahead and gave me a summary here at the bottom and we can go ahead and open it up so you go over to customer success and you'll see that we have the stakeholder map right over here i'm going to go ahead and open that up inside of a document and just like that you will see that we have the amplitude stakeholder map executive summary the priority contacts action required who the contact is why he matters **[39:26]** sarah's replacement okay still unknown because there wasn't any information about that but then we have the cfo what he cares about the evidence from the calls what the risks are without sarah no one is actually managing his skepticism the coo it went ahead and mapped all of that internal politics summary the red flag assessment based off of the playbook and then it went ahead and gave us a recommended action plan based off of every single week and then the honest assessment and said hey is the account still salvageable and then it went ahead and gave us all of that information **[39:57]** as well and then external research findings boom could not verify based off of the linkedin search however we can go ahead and give you some additional context on here as well and it went ahead and pulled their financial health based off of the company as well strategic direction documented wins and it gave us the sources on there right how cool is that that a csm was able to go ahead and pull that out right there and then the other thing is we could come back and say hey you know what i didn't like this i like this let's have a back and forth conversation launch another couple sub **[40:23]** agents in order to search their financial reporting scotch another agent in order to search their website we can put that in there say hey append that into the document all of that we can do by the power of our voice and an agent platform and then of course last but not least our revops agent so if you remember our revops agent and the scenario that we had was we have a meeting with the board on thursday and we say that our forecast is 2.3 million dollars in commit for q1 however we're not sure based off of the deals that are in pipeline whether or not that forecast is accurate **[40:52]** so again you'll see on here it went ahead and made a massive to-do list for itself it was able to work asynchronously for 10 20 minutes it went through all of the information it says now i have all of the data boom boom boom you can see how it's thinking through each and every single part of every to-do just like that and then you'll see that it went ahead and wrote a q1 forecast reality check and it went ahead and gave us a summary right here in terms of the bottom line i can go ahead and go over to our revops folder i'll go ahead and go over to the pipeline data **[41:20]** and then it says q1 forecast reality check i can open up that document right over here and you'll see this is the submitted commit it gave us an executive summary the hard truth you cannot commit 2.3 million with confidence your realistic number is somewhere between 1.2 and 1.75 the gap is primarily driven by one rep ryan whose pipeline is almost entirely dead deals being carried as committed and went ahead and gave us all of the high risk deals it went ahead and broke down what all that looks like the stage the last activity all of that you'll see on the right side this is the table of contents right over here the pattern analysis it gave us the **[41:55]** realistic numbers what to tell the board on thursday check that option one option two immediate actions required this month and then for your sop's and then it gave us a table summary of everything from there as well how cool is that right now let's go actually go back to our slide deck and see what's going on with that right here check this out we have our prospect messaging deck so i'm going to go ahead and open up our marketing folder we're going to go over here to the deck and right here we'll see that we have our powerpoint i'm going to go ahead and reveal in ### 42:24 — The finished, on-brand deck **[42:24]** finder from there we're going to go ahead and pull that up right over here open up the deck boom look at that and it's in lean scales branding and what i'm going to do is i actually want to show this off for you because what i want you to check out is how accurate this is compared to the wireframes this is how it gets really crazy and then guess what you can just throw this into google slides and then here's a really cool part every single one of these let me actually just show you this every single one of this is not nano banana Gemini 3 images this is as if a real human went **[42:55]** in and put that text in there i can edit this text i can edit this box i can edit any piece of this imagine how much time this is now literally just saved me look at this boom boom boom boom and then this is all based off of one-to-one what was inside of the wireframes my screen's a little bit small here but you'll be able to go ahead and see this is that you'll see on the left side this was the wireframe this is the slide right over here you'll see why tech corp signed why tech corp signed boom speed focus pricing boom let's go over the ai problem we got the ai problem right over here **[43:25]** check this out look at how accurate this is okay and then all i do is i literally just drag that over here and then it'll open up boom and then guess what i now have a google slides version and how insane is that that literally every single piece of this is something that i can click and edit as if a real human went in and put that and it's all according to the brain tone imagine how powerful that is and so really what i want to do inside of this video is i really hope that we were able to put the light bulb moment for you to understand the difference between the consumer app ### 43:55 — The endgame: MCP and no more SaaS UIs **[43:55]** which is again what every everybody thinks what ai is or they think that ai is going to be that little chat bot that lives inside of every single one of the individual tools that we use no no no the future is going to be utilizing agent platforms and then guess what everything that we discussed here today okay was everything that was just living in terms of the internal use case we haven't even yet talked about how you can now connect these agent platforms to the mcp's and so from my agent platform right over here i can connect to the sales force mcp i can connect to the hubspot mcp **[44:23]** i have a joke that i make here at lean scale where i say i do not want to log into any single sass ui ever again everything that we do going forward is happening within the agent platform and i literally just tell the agent hey go into our teamwork go into our project management software go ahead and give me the summary of what's going on with this project tell me what the progress of this other task is and we can all do that through the agent platform utilizing our own files as well as using the ai using the model in this agentic way and just to give you a breakdown ### 44:50 — 2026: power-user vs. mainstream agent platforms **[44:50]** of what the landscape is currently looking like is that up until middle of 2025 a lot of these agent platforms didn't really even exist okay this is how new and really how cutting edge all of this stuff is and so we thought hey cool that's ai right that's where everybody's living right now but the truth is that now you've gone ahead and explored what the agent platforms look like now within that what we've gone ahead and demoed here today is what is called clod code and then right here you're going to see that there's kind of two different versions of this there is the vs **[45:16]** code extension which is what you saw us use and what i went ahead and helped you set up and then there's also the terminal cli version but again this is what i call kind of like the power user version okay so up until probably last six months there was the only the consumer app and then there was the power user version okay now the thing is that these agent platforms took off so much over the last couple months being able to do all the stuff that i just showed you is that now clod is actually coming out with a mainstream version of the power user version here and like i said that **[45:42]** actually just came out just a couple days ago and so it is still getting traction it's primarily for max users right now it is a research preview and the way that i describe it is that it won't have all of the feature set that we went ahead and talked about inside of this video and so you can kind of think about it's like the consumer app is in order to get work done it's like using a drill that you might buy at walmart it's great in order to fix things at your house every once in a while right and then the power user section is going to be like okay let's get out the construction **[46:08]** equipment we're talking about let's get out the bulldozer let's get out the chainsaw it's really going ahead and allowing you to get a lot of work done and then the mainstream user version of this is like yeah we have a drill it's a little bit more hardcore than the drill that you get at walmart but it's not going to be the same as being able to do a lot of the functionality that we went ahead and talked about now you might be wondering well okay yeah that's all great what's the actual enablement piece of this how is the adoption going on at your organization right now well ### 46:32 — Adoption and the token leaderboard **[46:32]** we actually have a leaderboard that we have gone ahead and created and remember tokens we talked about ai tokens well this is actually a live leaderboard right now that what happens that the agent platforms are going ahead and putting all the information about their token usage onto this leaderboard and you'll see this we can actually see the usage of the tokens this month and this refreshes for us every single month over the last 30 days and it refreshes literally every 30 seconds based off of the token usage and so just in the last couple weeks our team has used 1.86 **[47:01]** billion tokens because we have now our entire team using these agent platforms and you see down here we got the power builders who are not incorporated in this but you'll see this myself and we have our head of engineering jake who's using tokens there as well so you'll see this is where you're now able to go to the entire team and then see the usage of each and every single team member based off of the tokens that they're using inside of the agent platform and so that ladies and gentlemen is in the course of however long this video has been you've now seen me pretty much do every **[47:27]** function in go-to-market or a major task in every single function in go-to-market using ai agents in the course of what a couple minutes in order to go ahead and execute all of that and so that is the power of agent platforms and really this is all on the cutting edge ai is changing so fast and so if you have any issues setting any of this any of this up in order to actually understand what this looks like reach out to us we can go ahead and answer any questions and then like i said we're gonna have a series of these videos talking through not only the use cases but then **[47:51]** also how we're now redoing all of our playbooks for go-to-market in this new ai first approach and so we can't wait in order to share that with you so with that being said this is Yasin from LeanScale checking out catch you in the next one --- _LeanScale Podcast Knowledge Hub. Free to quote and cite with attribution to The LeanScale Podcast (https://www.leanscale.team)._