AI Experiments
Wisdom over wizardry.
Working experiments in how AI changes the way companies build and operate. We run them on real work rather than demos, and we publish what we learn — including the parts that did not pan out.
Read the findingsHow we run them
-
On real work, not demosAn experiment only counts if it runs against actual client or product work, where the messy parts show up.
-
Written down as we goEach experiment gets documented so the result is repeatable — or so we know precisely why it was not.
-
Negative results publishedApproaches that fail are worth as much as the ones that work, and far more than a claim with nothing behind it.
-
Folded back into the workWhat proves out becomes part of how we run advisory and engineering, rather than a separate thing we sell.
Current experiments
Write-ups land in Writings as they finish.
AI-assisted delivery workflows
Measuring where AI tooling genuinely shortens the path from decision to shipped software, and where it quietly adds rework.
Operating with smaller teams
Testing which parts of a scaling company's operating load can be carried by tooling before structure has to grow.
Common questions
What counts as an experiment here?
Something run on real work with a result we are willing to publish either way. A demo that only works in ideal conditions does not qualify.
Is this an AI consulting offering?
Not primarily. It is a research practice. What proves out gets folded into advisory and engineering work rather than sold as a separate AI product.
Can we run an experiment with you?
Sometimes, when the question is one we are already curious about. Email us with what you are trying to find out.