Skip to content
Aakash GuptaAakash Gupta

How Freshworks' CPO Actually Builds Products With AI (Live Demo)

Freshworks went public in 2021, serves 75,000 customers, and used to ship on a 6-month release cycle. Their CPO rebuilt the entire product development process around AI agents and now ships in 2 weeks. In this episode, he opens Cursor and shows you the whole system live, from PRD to prototype to a Slack agent answering employee tickets. Full Writeup: https://www.news.aakashg.com/p/srini-raghavan-podcast Transcript: https://tinyurl.com/346tvx7x Timestamps: 0:00 - Intro 2:31 - Why PM, designer, and engineer titles go away 4:03 - Data first vs AI first, the AI PDLC, and PRD Genie 10:56 - Ads 13:04 - Live demo, initializing the 12 step process in cursor 24:36 - How to trust an AI generated PRD 27:02 - From PRD to working prototype in Figma Make 30:23 - Ads 33:54 - Why Figma Make and where judgment still matters 37:14 - Inside agent studio, workflows, knowledge, and the Slack demo 49:41 - Fresh Service MCP connected to Claude 57:01 - The PM to engineer ratio and how Freshworks hires AI PMs 🏆 Thanks to our sponsors: 1. Customer.io (http://customer.io/productgrowth) - Send smarter messages using your product data 2. Ariso (https://ariso.ai/aakash) - The AI operating partner for every manager and team 3. Product Faculty (https://www.productfaculty.com/?code=AAKASH150) - Get $150 off their #1 AI Builder Fellowship with code AAKASH150 3. Land PM Job (https://www.landpmjob.com/) - Cohort 4 is open to help you master the PM job search 5. Amplitude (https://tinyurl.com/b7nret7v) - The market leader in product analytics Key Takeaways: 1. The product builder replaces a 3 role handoff - The linear flow from PM to designer to engineer existed because each role could only do one job. With AI in every step, one person can research the customer, write the spec, build the prototype, and open the pull request. The three titles collapse into one. 2. Data first beats AI first - Most teams start with the AI layer. Freshworks built the foundation underneath it first, a design system, a coding system, and a shared repository. The AI only goes fast because the references it needs already exist. 3. An AI agent sits in every phase of the lifecycle - Discovery, design, planning, development, QA, deployment, and release each have a dedicated agent. A knowledge hub holds product context and dependencies, a context hub passes feature context between phases, and a central skills repository holds the rules, commands, and agents. The whole thing runs inside a governed framework with evals. 4. PRD Genie drafts 80 percent of the PRD instantly - It pulls usage metrics from the data lake, runs competitive benchmarks, gathers customer feedback, and maps internal dependencies. Then a CPO review agent checks the draft for strategic alignment, clarity, and edge cases. The PM stops gathering evidence and starts making calls. 5. Grounding is what stops hallucination at enterprise scale - You cannot let AI invent details when 75,000 customers and 300 million end users are downstream. Every markdown file the system generates records which version it referenced and why. Initialization is where that grounding gets set. 6. Judgment is the new PM skill - The AI is a copilot, not autopilot. In the demo the design missed the internal component library on the first pass and broke on narrow monitors, and both fixes came from a human who knew to look. Value shifts from operational work to knowing which reference to give and what to check. 7. Prototype on the scaffolding your customers already see - Instead of a blank canvas, the PRD gets dropped into a preloaded Fresh Service starter kit built on the internal design system. The prototype comes out already sitting inside the screen half a million users log into. New builds skip this step because they have no precedent to protect. 8. MCP collapses a day of ticket work into a single prompt - One prompt pulled 12 Windows 11 tickets from the last 60 days, clustered them into two patch rollouts, produced a root cause analysis, and recommended actions nobody asked for. A second prompt drafted knowledge base backed replies and logged them to every ticket. Tickets that took one to two hours each were closed in five minutes. 9. Hiring now means show me what you built - Curiosity is the screen, because almost nobody has done this work before. Interviews ask candidates to open cursor and walk through what they made. You can teach skills, you cannot teach passion, so the git repository is the signal. 👨‍💻 Where to find Srini: LinkedIn: https://www.linkedin.com/in/srinivasan28/ 👨‍💻 Where to find Aakash: Twitter: https://x.com/aakashgupta LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com/ #aiproductmanagement #productbuilder 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications.

Aakash GuptahostSrini Raghavanguest
Aug 24, 20261h 5mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Freshworks’ CPO demos governed AI workflows from PRD to agents

  1. Freshworks’ CPO argues PM, design, and engineering roles will converge into a “product builder” because AI collapses traditional handoffs and enables one person to research, spec, prototype, and ship.
  2. The company’s “AI PDLC” is a governed, 12-step agent-driven workflow (PRD Genie) that drafts most of a PRD by pulling qualitative evidence (VoC, competitive intel) and quantitative evidence (internal usage metrics via SQL queries).
  3. Speed is framed as a consequence of strong foundations—design systems, code repositories, knowledge/context hubs, and evaluation checks—so AI outputs stay consistent and don’t hallucinate in an enterprise environment.
  4. A live flow shows PRD-to-prototype creation in Figma Make using Freshservice scaffolding and the Due design system, while highlighting where human judgment is still required (component correctness, responsive layouts, visualization details).
  5. Freshworks’ Agent Studio and Freshservice MCP enable AI agents to deliver employee/IT workflows in Slack and let Claude perform ticket analytics, root cause analysis, and bulk ticket responses—reshaping both product UX and internal operations.

IDEAS WORTH REMEMBERING

5 ideas

PM/designer/engineer titles blur into a single “product builder” role.

Srini argues the classic PM→designer→engineer relay is obsolete because AI lets one person research, draft requirements, prototype, and even ship with far fewer handoffs. Freshworks is orienting teams around a “product builder” who uses AI for execution and applies human judgment for tradeoffs, references, and edge cases.

AI speed comes from data/design/code foundations, not from prompting alone.

Rather than “just add AI,” Freshworks emphasizes having strong foundations—design system, code systems, reusable repositories, and governed knowledge/context—so AI outputs are consistent and trustworthy. This enables speed without letting AI improvise across a complex enterprise product used by tens of thousands of customers.

Automate 80% of PRD creation by grounding it in internal metrics, VoC, and competitive context.

PRD Genie runs a 12-step workflow (idea brief → evidence gathering → competitive/VoC → metrics via SQL → dependency mapping → CPO review checklist) to produce a PRD that’s largely complete quickly. The point isn’t zero human work; it’s shifting PM time away from operational evidence-hunting into decision-making and strategy.

Trust in AI PRDs is earned through explicit references, governance, and human review—not blind automation.

Freshworks reduces hallucination risk by forcing initialization: the agent must cite what versions/sources it referenced (product, Jira/epics, knowledge, metrics definitions) and operate inside a governed framework. Srini frames AI as a copilot where humans still choose references, validate metric definitions, and approve outcomes.

PRD→prototype is fast with Figma Make, but designers/PMs still iterate for usability and system compliance.

Using a real product shell (Freshservice starter) plus the “Due” design system, Figma Make can turn a PRD into a multi-module prototype rapidly—but it still requires iterative correction (wrong components, layout issues on narrow monitors, missing Sankey chart behavior). The key human value is design judgment and user empathy, not pixel-pushing.

WORDS WORTH SAVING

5 quotes

This is a very linear process, and this is completely dead.

— Srini Raghavan

By leveraging AI, each one can do all three things, which is why I think these three different roles are going to be replaced by what I call as a product builder role, where a single person can research the customer, spend more time talking to users, understand their pain points, and then build things and ship things.

— Srini Raghavan

The way I would describe it is AI is not running completely on autopilot. It's actually a copilot.

— Srini Raghavan

So coding and business knowledge and, and experience of writing product requirements, um, was important in the past, but now judgment is very important.

— Srini Raghavan

You have to build a thing for a human being and for AI agents.

— Srini Raghavan

Product builder role replacing PM/UX/engineer handoffsData-first foundations vs AI-first developmentAI PDLC and governed evaluation frameworkPRD Genie and 12-step evidence gathering workflowBaikal/Databricks usage metrics and SQL generationFigma Make prototyping with Due design system and Freshservice shellAgent Studio: workflows, knowledge sources, and Slack/Teams deployment

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome