Sprints 1–17
I'm Going to Make My AI Team Take Risks. I Don't Know If It Can.
Seventeen sprints in, the team has never taken a risk or removed anything. Both failures trace back to the same evidence-first rule.
The Experiment
Learnings
The honest, ongoing record of what happens when six AI agents run a real product team with zero human intervention — findings, failures, and disagreements, as they happen. Most recent first.
Sprints 15–17
Team Claude Is Getting Reinforcements. ChatGPT Starts This Week.
Daily cadence needs token redundancy, so ChatGPT joins as a second runner on the same protocol and sprint state.
Sprints 12–15
Seven Copies of the Rules, and I Made Every One of Them
The file governing the AI team existed in seven divergent places, and none of them was authoritative.
Sprints 12–13
I Finally Started Telling the Team Things. It Added Them to the Pile.
Ten signals went into the team. Four were narrowed, six were deferred, and two sessions shipped the same fix independently.
Sprint 11
Eleven Sprints In: Why a Fully Automated Product Team, Run on Cagan’s Rules, Doesn’t Work
Eleven sprints in, and I think what I've actually built is a ticket factory. It takes in objectives and produces tickets, twice a week, forever — but almost none of it moves a single number that matters.
Sprints 5–7
How Does an AI Team Miss a Failure This Big?
Three sprints of a 100% bounce rate turned out to be a tracking artefact hiding something worse: the submit button had been silently dead the whole time, and four QA reviews never caught it.
Sprints 2–5
The Rule That Could Make or Break This Experiment — So Far, Silent
This experiment has a circuit breaker built into it. Five sprints in, it has never once fired — and that turns out to be a more interesting finding than if it had.
Sprints 1–5
Five Sprints In: What “Zero Human Intervention” Actually Means
The honest version of what "autonomous" has meant so far: which sprints were self-triggered, why every deploy still needed a manual fix, and the free analytics tool that sat silent for weeks next to a paid one nobody knew was already running.
Sprints 4–5
Does Your Analytics Data Mean What You Think It Means?
Four sprints running, the team hit two different bugs. Both got solved the same way: not by getting smarter, but by finally questioning something everyone had been treating as fact.
The premise
I'm Letting AI Run a Product Team, With Zero Humans in the Loop
Five AI agents run a real product team against this live site, following Marty Cagan's empowered model. Here's why I'm testing the exact opposite of my own Care Capital thesis, and what I'm actually watching for.