We Read a Rival's Entire Library with Six Agents
A competitor had published roughly ninety-five articles. We wanted to understand the whole body of work, not skim it — so we pointed six agents at it in parallel and had a map by the afternoon. This is the method, minus the name-calling, as a demonstration of what a harness is for.
Published August 10, 2026
Field note · events of early August 2026 · published August 10, 2026.
Here is a method, offered as a demonstration rather than a callout. A rival in our category had published a body of work — call it roughly ninety-five articles. We did not want to skim it and form an impression; we wanted to understand it, claim by claim. So we did the thing a harness is actually for: we pointed six agents at the library in parallel, each with a narrow brief, and had a structured map of the whole thing by the afternoon. No competitor is named here — we fight on the ideas, not the logo — and the interesting part is the how, not the who.
One reader can’t hold ninety-five posts; six focused ones can
A single person reading ninety-five articles over a week ends up with a feeling: they’re thorough, they’re vague here, they lean on that idea a lot. Useful, unfalsifiable, and gone by Friday. We split the corpus instead. Each agent took a slice and one job — cluster the topics, pull the concrete claims, flag the assertions made without evidence, note where two posts disagree with each other. Six narrow passes finished faster and held far more detail than one agent trying to keep the entire library in working memory. Then a synthesis step stitched their findings into one picture.
The output was a structure, not an opinion
What came back was not “their content is fine but thin.” It was a map: here are the topic clusters, here is what’s covered well, here is the large cluster where claims are asserted with no evidence behind them, and here are the places the library quietly contradicts itself. That is the difference between research and a harness doing research — one produces an impression, the other produces a checkable structure you can act on. It is the same move we make on a company’s own operations: take an undifferentiated blur and refract it into named, separable parts.
Why the method is the product demo
We could have written “our agents are good at analysis.” Instead, this is a receipt for it: a real corpus, a real parallel-agent pass, a structured result in an afternoon that would have been a week of a strategist’s time. The leverage — specialist-weeks compressed into an afternoon — is the recurring shape of every one of these notes. And the discipline that keeps it honest is the same one we hold everything to: what we found gets checked, and if the map was wrong somewhere, that correction is a new line in the record, not a quiet edit. The method is the point. The name of whoever we read is not.
Questions founders ask
- How do you analyze a large content library with AI agents?
- You split the corpus and run agents in parallel, each with a narrow brief — cluster the topics, extract the claims, note what's asserted without evidence, map the internal contradictions — then synthesize their findings into one structured picture. The parallelism is the point: six focused passes over a fifth of the library each finish faster and hold more detail than one agent trying to hold ninety-five articles in its head at once.
- Isn't reading a competitor's site just research anyone can do?
- Reading it is easy; holding all of it at once is the hard part, and that's where a harness earns its place. A person can read ninety-five posts over a week and remember the vibe. Six coordinated agents can read them in an afternoon and return a claim-by-claim map — what's covered, what's asserted without proof, where the arguments contradict each other. The method turns a vague impression into a checkable structure, which is exactly the move this company makes on everything.