I Built 52 Startups in 52 Weeks with AI Coding Agents – Here’s What Actually Works

TLDR: Serial builder who shut down AI 51 startups shares brutal truth about AI coding – more output doesn’t mean better outcomes.

Michael Yagudaev drops unfiltered truth about AI coding at Vancouver AI Meetup. This isn’t another “AI will save the world” pitch – it’s field notes from someone who actually ships.
Key Insights:

– Why 52 startups taught him more than any bootcamp
– Claude Code workflow that actually works (with limitations)
– The cognitive cost of AI assistance nobody talks about
– Building in public as competitive advantage
– Quality control when AI writes 125,000 line PRs

Michael built AudiowaveAI after 51 failed experiments. He runs multiple Claude Code sessions simultaneously, cafeinates his Mac overnight for AI tasks, and has changed his entire workflow 4 times in 4 years.

Forget theory—this session is pure tactical gold for builders at the edge of AI, culture, ethics, and community in Vancouver and beyond.

Michael Yagv walks us through:

Building in public → validate faster, ship simpler
Moving from output to outcomes with agentic coding
Claude Code vs. fancy RAG—why simple sometimes wins
The “Software 3.0” mental model (Karpathy-style)
Multi-agent workflows, parallel tickets, and tight feedback loops
Guard-railing code quality with automated tests & screenshots
Cognitive risks of over-relying on LLMs (MIT study)

Tools stack: Linear → Claude Code → Cursor / Windsurf → Cyrus → PostHog

📍 Chapters

Timestamp Chapter Title
00:00 Incoming Transmission & Meetup Intro
00:14 Who Is Michael Yagv? 52 Startups in 52 Weeks
01:21 Stretch Break—Movement Matters
01:50 Why 52 is better than 12 is better than 1: Finding the Winner Fast
02:48 Viral Hit: AudioWave Origin Story
03:55 Output ≠ Outcome—Customers Only Care About Value
05:02 Build Fast, Keep It Simple (Paul Graham Throwback)
06:44 Leverage 101: Scope vs. Tools
09:26 Claude Code over Complicated RAG (Unix-Style Search FTW)
12:00 Evolution of Coding Tools: Copilot → GPT-4 → Agents
14:39 Software 3.0 Framework & 1.0 Code Crazy
16:48 Inspiringly Simple AI Products (v0, remove.bg, linktr.ee)
18:42 Running Multiple Claude Code Sessions in Parallel
21:25 29-Day Countdown: Claude Code Credit Reality Check
22:52 MIT Study: LLMs & Critical-Thinking Decline
25:40 Closing the Feedback Loop (Tests, Screenshots, Evals)
28:06 Michael’s End-to-End Agentic Workflow Demo
31:42 Tool Stack Breakdown & Voice-Prompting Hacks
35:45 Audience Q&A: Agents, PR Merges, Guardrails
45:00 Wrap-Up & Next Steps for Vancouver AI Community