Plain-language articles from the lab — what we're measuring, what surprised us, and what it means for organizations of humans and agents. For the strict versions, see Research.
A hallucination in front of a client is a model failure. A team quietly switching the automation back off is not — and it's the more expensive one. Why the human side of agent reliability is measurable, and what we found when we measured ours.
Why trust in an agent is a product, not a sum: a perfectly aligned agent that fails randomly is untrustworthy, and a perfectly reliable one pointed the wrong way is worse. The one formula we run the whole lab on.
Everyone claims to be "AI-first". Here's what a graded scale looks like when you refuse to hand out participation trophies — including why we score ourselves L0.9.
We expected the raw corpus to cost more than our distillate — it didn't. What a lab does next is the whole game.
Not a tool stack — an operating model: humans and agents working, learning, and evolving together under working institutions. Measurably. The SEO-companion to our glossary.
Social sharing: each note carries its own OG card. Newsletter mechanism — TBD (Substack recommended, decision pending).
