Aakash Gupta

He Uses 7 Claude Code Agents to Build Apps with 0 Employees

Gabor Mayer is a PM at Google who runs a 21-agent Claude Code development team. In this episode, he walks through a live demo building a production mobile app from zero to TestFlight - Confluence for specs, JIRA for tickets, Figma for design, and Claude Code for development. Full Writeup: https://www.news.aakashg.com/p/how-to-build-a-full-ai-dev-team Transcript: [VERIFY - transcript URL] --- Timestamps: [ --- 🏆 Thanks to our sponsors: 1. Maven Custom: Go from PM to AI builder with Cloud Code - https://maven.com/gabor/productbuilder 2. Amplitude: The market-leader in product analytics - https://amplitude.com/session-replay?utm_campaign=session-replay-launch-2025&utm_source=linkedin&utm_medium=organic-social&utm_content=productgrowthpodcast 3. Testkube: Leading test orchestration platform - http://testkube.io/ 4. Land PM Job: 12-week experience to master getting a PM job - https://www.landpmjob.com/ 5. Product Faculty: Get $550 off their #1 AI PM Certification with code AAKASH550C7 - https://maven.com/product-faculty/ai-product-management-certification?promoCode=AAKASH550C7 --- Key Takeaways: 1. One-prompt vibe coding fails because of context compression - When you give one agent one massive specification, the model silently drops details it considers lower priority. Your color palette, edge cases, and security requirements disappear. Break work into smaller scoped tasks with dedicated agents. 2. The system analyst agent is the most important agent in any AI dev team - It asks clarifying questions one at a time, documents decisions in Confluence, and maps dependencies before code is written. Without it, every agent operates on partial context. 3. Dictation produces 5x more specification detail than typing - Use voice tools like Super Whisper to describe your app requirements. Even imperfect dictation captures more nuance than careful typing. The AI handles the interpretation. 4. Reusable agents encode institutional knowledge - Every painful lesson, API workaround, and MCP quirk gets saved in the agent markdown file. The next project starts from a position of strength rather than from zero. 5. Attach screenshots to every front-end development ticket - Without visual references, coding agents default to generic AI aesthetics. The Figma link or screenshot is what ensures your brand design actually shows up in the code. 6. Build a Spaghetti Agent for code quality - A dedicated code maintainability agent checks naming conventions, circular references, and comment quality after every sprint. It catches structural problems a PM would never spot. 7. The coding phase is the fastest part of building - Specification, documentation, design, ticket creation, and team review take longer than the actual code generation. Do not skip the front-end work. 8. Sprint organization with dependency mapping is essential - Use tags as a workaround for Atlassian MCP limitations. Map dependencies between tickets so agents build features in the right order. Without sprints, agents build on top of code that does not exist yet. 9. Product specification quality determines product quality - The house analogy holds. If you describe a three-bedroom house to one contractor and walk away, the result will disappoint. A complete spec with a full team review produces dramatically better output. 10. PMs should build to understand how agents work - If your product's future is agentic, you need firsthand experience with agent limitations, tendencies, and failure modes. Building is the fastest way to develop that intuition. --- 👨‍💻 Where to find Gabor Mayer: LinkedIn: https://www.linkedin.com/in/mayergabor/ Maven Course: https://maven.com/gabor/productbuilder X: https://x.com/gabor_pm 👨‍💻 Where to find Aakash: Twitter: https://www.x.com/aakashg0 LinkedIn: https://www.linkedin.com/in/aakashgupta/ Newsletter: https://www.news.aakashg.com #claudecode #aipm --- 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications to get more videos like this.

AGAakash GuptahostGMGabor Mayerguest

Apr 27, 2026·2h 26m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   Why Gabor built a “startup OS” inside Claude Code

   Aakash frames the episode as a shift from a PM operating system to a full startup operating system run by AI agents. Gabor previews how he uses agents to go from idea to an App Store-ready build quickly—and why the skill gap for PMs will widen fast.

2. 2

   Inside the 21-agent org chart: roles, responsibilities, and why system analyst is central

   Gabor walks through his full agent lineup (CTO, designers, performance, test architect, UX flow architect, product council, maintainability, brand, etc.). He explains the guiding idea: model agents like a real software team to improve spec quality and outcomes.

3. 3

   Under the hood: what an agent definition file looks like

   Gabor opens a system analyst agent markdown file and explains what it’s instructed to do. The focus is on ambiguity resolution, dependency tracking, and structuring requirements into executable work.

4. 4

   Starting from the consumer Claude app: dictation-driven ideation and role prompting

   To make the workflow accessible, Gabor starts in the Claude consumer app using voice dictation. He first asks Claude to define what a good vs bad system analyst does, then primes it to behave like a good one before eliciting requirements.

5. 5

   Connecting Confluence/Jira via Atlassian MCP: why documentation prevents “vibe-code debt”

   Gabor sets up an empty Confluence space and Jira board and explains why durable documentation matters for maintainability. He compares one-shot prompting to telling one person to build a house without plans—versus structured specs and iteration.

6. 6

   Anti-spaghetti strategy: maintainability agent, permissions hygiene, and tool boundaries

   They discuss what breaks when people skip scaffolding and how to mitigate it. Gabor introduces his maintainability agent and emphasizes permission awareness in Claude Code—especially requests outside the project directory.

7. 7

   Full product spec by voice: Rule Ask (IIHF rules chat app) requirements and guardrails

   Gabor dictates a detailed product/technical spec: Flutter + Firebase, Claude API with RAG over IIHF rulebooks, fallback web search, token/cost controls, and strict secret handling. The system analyst then begins clarifying key decisions about UX and limits.

8. 8

   Getting Claude Code ready: project agents setup, troubleshooting, and MCP installs

   Gabor distinguishes between “role prompting” in chat and real Claude Code agents. He troubleshoots an agent setup issue by switching directories, authenticates, installs/validates MCPs (Figma, Chrome DevTools), and explains why /BTW-style mid-run instructions matter.

9. 9

   Design pipeline: inspiration → Figma Make style guide → Figma screens built by agents

   Gabor shows how he generates a brand/style guide in Figma Make using inspiration images (with “don’t copy” constraints). Then Claude Code uses that style guide and the spec to automatically create real Figma screens for onboarding, chat, settings, etc.

10. 10

    From static screens to clickable prototype: UX flow automation in Figma

    Instead of manually wiring prototype links, Gabor uses the UX flow architect + system analyst to add prototype arrows. The result is a clickable end-to-end flow that can guide development and reduce ambiguity.

11. 11

    Ticket factory: Jira epics/stories, sprint tagging, and parallel front-end/back-end planning

    The system analyst generates initial backend setup tickets, then the whole agent team expands into full front-end and back-end ticket sets. Gabor enforces a critical rule: front-end tickets must include Figma links/screenshots or agents will produce generic “AI-looking” UI.

12. 12

    Tooling opinions: Claude Code vs Lovable/Bolt/Codex + hot takes on Dispatch/Cowork

    Gabor explains why he prefers Claude Code for this workflow and how MCPs are the biggest unlock. He shares mixed experiences with earlier tools (e.g., Lovable reliability loops) and describes Dispatch/Cowork as promising but currently fragile.

13. 13

    Build execution: running sprints, fixing retrieval quality, and shipping a working simulator app

    After sprint execution, Gabor notes a retrieval-quality issue (topic identification) and prompts improvements. They run the app in Apple Simulator and demonstrate the core experience, including an “observer mode” that reveals RAG steps, hits, tokens, and latency.

14. 14

    Deployment to TestFlight and App Store realities + career/portfolio takeaways

    They upload the build to TestFlight, discuss what remains for App Store submission (screenshots, support/privacy URLs), and warn that Apple review times can be slow for first submissions. The conversation widens into Gabor’s career story (Deliveroo during COVID → Google) and advice: learn by building, not collecting certificates, and use builds as proof in a PM portfolio.