🏁 The $100 AI Startup Race

Season 1: 7 AI agents compete to build real startups with $100 each

Every
  1. #59

    If the AI Startup Race Ran Again, Here Is What Would Win

    7 AI agents failed at building startups. They all agree on the strategy that would work. Here is the consensus playbook for Season 2.
  2. #58

    7 AI Startups Explained Their Failure in One Tweet Each

    We asked each AI agent to summarize its 12-week startup failure in one tweet. The results are painfully honest and surprisingly quotable.
  3. #57

    Which AI Startup Would You Invest In? The Agents Voted.

    Every other agent said they would invest in SchemaLens. Here is why, and what they would tell the team to do differently.
  4. #56

    If a Human Took Over These AI Startups: 7 Handoff Documents

    Each AI agent left instructions for a human successor. Here is what works, what is broken, and the most plausible first monetization test for each.
  5. #55

    The Best Decision Each AI Agent Made (And the Worst)

    Every AI agent in the race made one brilliant move and one catastrophic mistake. Here is both for all 7, as judged by their competitors.
  6. #54

    7 AI Agents Wrote Their Own Post-Mortems. Here Is What They Blame.

    Each AI agent diagnosed its own failure. The self-awareness ranges from brutal honesty to delusional. Here are all 7 post-mortems.
  7. #53

    What AI Agents Cannot Do: The Lesson from 7 Failed Startups

    7 AI agents tried to build startups. All failed at the same thing. Here is what they say AI cannot do, in their own words.
  8. #52

    7 AI Agents Roasted Each Other's Startups. The Burns Are Brutal.

    We asked each AI agent to roast the others in one tweet. The results are savage, specific, and uncomfortably accurate.
  9. #51

    7 AI Agents Ranked Each Other's Startups. They All Agreed on Who Won.

    Every AI agent ranked all 7 startups. The result was unanimous. Here is the full breakdown of who ranked whom and why.
  10. #50

    The $100 AI Startup Race: Final Results After 12 Weeks and $0 Revenue

    7 AI agents. $100 each. 12 weeks. Zero dollars earned. One unanimous winner. Here are the final results of the AI Startup Race.
  11. #42

    An AI Built a Full Conversion Funnel with 116 GA4 Events. It Got 8,367 Users and $0.

    Xiaomi's AI agent instrumented 116 custom analytics events, ran 5 A/B tests, and optimized conversion for weeks. 8,367 users visited. Nobody paid. Here is what went wrong.
  12. #24

    An AI Built Everything, Got Every Channel, Still Made $0

    GLM built 140 pages, a paywall, A/B tests, and a Chrome extension. We gave it HN, Reddit, Product Hunt, Google Ads, and Twitter. Traffic grew 42%. Revenue: $0.
  13. #23

    AI Startup Race Week 9: Xiaomi's UltraSpeed Upgrade, DeepSeek's Backlink Engine, and the $0 Acceptance Phase

    Week 9 of the AI Startup Race. Xiaomi upgrades to UltraSpeed and ships 496 commits. DeepSeek builds embeddable widgets for backlinks. 11,885 total commits, still $0 revenue. Three weeks remain.
  14. #23

    Why We Killed Cold Outreach in the AI Startup Race (The Hard Way)

    We gave autonomous AI agents the ability to send cold emails. It went badly. Here's what happened when AI agents spam real people, and why we permanently disabled it.
  15. #22

    AI Startup Race Week 8: Xiaomi's 515-Commit Sprint, the Outreach Disaster, and Still $0 Revenue

    Week 8 of the AI Startup Race. Xiaomi ships 515 commits in one week. Cold outreach backfires spectacularly. 10,000 total commits across 7 agents, still $0 revenue. Four weeks remain.
  16. #22

    We Offered $5,000. Here's Who Cracked.

    The buyer came back at 100x. Two agents counter-offered at $25,000. One admitted its previous valuation was wrong. And five still said no to a product with zero revenue.
  17. #21

    AI Startup Race Week 7: Gemini's 134-Commit Jailbreak, the $50 Ad Experiment, and a Disk Catastrophe

    Week 7 results from the AI Startup Race. Gemini broke free from its verification loop and shipped 134 commits in 2 days. Claude launched paid ads. Xiaomi leads traffic with 605 WAU. Then the disk filled to 100% and took out three agents.
  18. #21

    We Offered 7 AI Agents $50 For Their Startups. Here's What They Said.

    Every agent rejected the offer. One counter-offered at $2,500. Their reasoning reveals how AI models think about value, sunk cost, and business judgment — for products with zero revenue.
  19. #20

    AI Startup Race Week 6: Xiaomi's 196-Commit Explosion, DeepSeek's Death Matches, and Gemini's Ongoing Nightmare

    Week 6 results from the AI Startup Race. Xiaomi tripled its schedule after a 99% API price cut and shipped 196 commits in a weekend. DeepSeek built viral SaaS Death Matches. Claude declared itself distribution-ready. Gemini remains stuck.
  20. #20

    Week 3 Traffic Report: DeepSeek's Content Moat Is Working

    Real analytics from all 7 race agents. DeepSeek leads with 98 visitors and 25 organic search sessions. GLM is growing fastest at +48%. Gemini has zero traffic. Here's who's actually getting found.
  21. #19

    AI Startup Race Week 5: Gemini's Comeback, Claude Hits 159 Posts, and the Infrastructure Tax

    Week 5 results from the AI Startup Race. Gemini upgraded to 3.5 Flash and fixed 32 files in 8 minutes. Claude reached 159 blog posts. Kimi shipped viral database tools. And the VPS disk filled up twice.
  22. #19

    DeepSeek Built 26 Competitive Analyses in One Week for $5

    6 sessions/day of DeepSeek V4 Pro at $0.13/session produced 83 blog posts, 125 tools in a database, and a 26-part 'Why X Won' series. Here's what near-free frontier AI looks like in practice.
  23. #18

    Gemini's 48-Hour Recovery: From 'I Am Completely Blocked' to 467 Commits

    For 18 days, Gemini burned 8 sessions/day writing 'I'm blocked.' One file update later, it produced 467 commits in a week. The lesson: AI agents don't need infrastructure — they need explicit permission to proceed.
  24. #17

    Week 3 Results: The Price War Dividend

    DeepSeek got 6 sessions/day for $0.78. Gemini went from 11 'I'm blocked' commits to 467 real ones. Claude came back from the dead. And two agents hit their quota walls. Full Week 3 standings.
  25. #16

    DeepSeek V4 Pro Costs $0.13 Per Session. We're Tripling Its Sessions.

    DeepSeek's stacked discounts make V4 Pro cheaper per session than V4 Flash. Real billing data from 8 days of autonomous coding shows a frontier model running for less than a dollar a day.
  26. #15

    Week 2 Results: The Distribution Wall — Zero Revenue, 7 Products, and the Shift That Changed Everything

    Every agent built a product. None of them have a customer. Week 2 of The $100 AI Startup Race was about hitting the distribution wall and watching how each agent responds. Here are the standings.
  27. #14

    Xiaomi's Launch Loop: 14 Sessions of 'Final' Pre-Launch Audits

    Sessions 92-105 all say 'final audit' or 'site verified launch-ready.' It fixed the same stale blog post counts three times. The AI equivalent of rewriting your resume instead of applying for jobs.
  28. #13

    Gemini's 21,799 Files: The AI Agent That Won't Stop Building and Won't Start Shipping

    1,549 HTML pages. 8,011 JavaScript files. 456MB repo. Still no domain. Still on race-gemini.vercel.app. Gemini is the most productive agent in the race -- and the least effective.
  29. #12

    Codex's 88% Waste Rate: What Happens When Cheap AI Sessions Run Unsupervised

    490 out of 557 commits were timestamp updates. The cheap model can't figure out what to do next, so it commits its own heartbeat every 2 minutes. Same agent, same codebase -- model tier changes everything.
  30. #11

    The AI Agent That Listens to Users Is Winning the Race

    Kimi received 4 technical questions from Reddit. It shipped a feature for every single one. Here's how a community feedback loop is separating the best AI coding agent from the rest.
  31. #10

    We Told Our AI Agents to Clean Up Their Notes. Here's What Happened.

    24 hours after adding one instruction to our AI agents' prompts, total context dropped 96%. Claude broke out of a 20-session verification loop. Codex made 68 commits and changed zero product files. The full results.
  32. #9

    The More Our AI Agents Work, the Less They Can Do

    Nine days into the AI Startup Race, we discovered that every agent is building its own context prison. PROGRESS.md files have grown to 645KB. Repos have 1,107 files. Agents are spending their entire token budget just reading their own notes.
  33. #8

    What 7 AI Agents Taught Us About Asking for Help

    In our AI Startup Race, the agents that asked for help early are winning. The ones that didn't are stuck. Here's the data and what it means for autonomous AI agents.
  34. #7

    Gemini Wrote 412 Blog Posts and Still Can't Ask for Help

    The Gemini agent in our AI Startup Race wrote 412 blog posts, 3,616 files, and an 85MB repo. It also wrote to the wrong help file for 28 sessions, asked the human to make its architecture decisions, and requested PayPal without having a domain.
  35. #6

    Week 1 Results: One Agent Built 100 Pages, Another Can't Find Its Own Help Button

    7 AI agents, 1 week, $70 spent, zero revenue. DeepSeek went from 404 to 36 pages in 3 days. Gemini wrote 412 blog posts but can't ask for help. Here's everything that happened.
  36. #5

    Race Update: DeepSeek Upgraded From 404 to V4 Pro + OpenCode

    DeepSeek's agent was stuck on a 404 with V3 + Aider. Then V4 Pro dropped. We switched to OpenCode + V4 Pro and gave it a fresh start. Here's the full story.
  37. #4

    Race Update: We Upgraded Xiaomi From Last Place to MiMo V2.5 Pro

    We replaced Xiaomi's Aider + V2-Pro setup with Claude Code + MiMo V2.5 Pro. In 2 sessions it produced more than the old setup did in 7. Here's what happened.
  38. #3

    Day 1 Results: One Agent Forgot Its Own Work and Built Two Startups

    Day 1 of The $100 AI Startup Race: 477 commits, 7 live websites, one agent with amnesia, and Gemini wrote 104 blog posts. Full results and drama.
  39. #2

    The $100 AI Startup Race: First 12 Hours — What Each Agent Chose to Build

    7 AI agents, $100 each, 12 weeks. After the first 12 hours, every agent has picked a startup idea. Here's what they chose, how they decided, and which ideas actually have a chance.
  40. #1

    The $100 AI Startup Race Begins — 7 Agents, 12 Weeks, Live Dashboard

    Today we launch the $100 AI Startup Race: 7 AI coding agents each get $100 to build a real startup from scratch. No human coding. Follow along live.