Taste is the last human moat.
Models generate plausible. Practitioners learn to tell plausible from correct. We teach the distinction the way masters taught apprentices: with eyes, hands, and repetition.
We build AI systems with your team, or teach you to build your own.
Meet the founders
Over a decade I shipped more than sixty AI systems for banks, insurers, public-sector teams, and a couple of unicorns that no longer exist. The pattern was always the same. The models worked. The humans around them did not know how to scope, measure, verify, or refuse.
Every failure I've seen close-up was a human failure. Somebody automated a mess. Somebody shipped plausible when plausible was not enough. Somebody did oversight cosplay while the model silently broke trust with its real users.
Cohorte started in 2022 with two people in a room in Paris. Nearly four years later, the firm has trained over two hundred practitioners and ships a weekly newsletter to fifteen thousand readers.
We teach the boring work. Scoping. Measurement. Verification. Governance. The courage to say "this cannot be automated yet." Taste is the last human moat. We make it teachable.
Reliability is all three: the answer is right, it holds up in the real world, and you can afford to run it.
I've sat in too many rooms where AI was announced before it was understood. Teams pushed to adopt before they're equipped to operate. Leadership committed to outcomes nobody can measure. The gap between the slide and the system is where the program quietly dies.
Partnerships at Cohorte bring practitioner discipline inside your organization, so leadership, operators and engineers stop talking past each other and work to the same standard.
Models generate plausible. Practitioners learn to tell plausible from correct. We teach the distinction the way masters taught apprentices: with eyes, hands, and repetition.
It makes great operators faster and bad operators louder. The gap between the two widens every quarter. Our job is to train the first kind.
The prompt is never the bottleneck. Every failed AI program we've watched began by automating organisational folklore. You cannot automate a mess.
The proclamation is the joke. A company becomes AI-first when it has shipped the systems, the discipline and the standards.
Every Cohorte graduate leaves with a written playbook: scope, measurement, verification, governance. Their team can run it without them.
Last updated 2 October 2026.
Four people run Cohorte. Practitioners from production AI roles teach and mentor alongside them.

Brand voice and curriculum lead. Has built AI systems for banks, insurers, ed-tech and consulting teams. Writes The AI OS weekly. Reads every application personally.
Founder
Builds Cohorte's enterprise relationships. Previously at EY. Turns "we want AI training" conversations into shipped engagements with organizations.
Co-founder
Hosts and curates The Engine Room. Books guest operators, runs the Friday Digest, moderates the Helpdesk. Onboards every new member personally.
Operator
Runs newsletter operations, content distribution and social media, turning our founder's frameworks into reach across channels.
OperatorPractitioners from production AI roles who teach between shifts. Cohort alumni who mentor incoming students. Plus guests from our network (Daniel Hulme of WPP, Thomas Scialom of Meta Superintelligence Labs, and others) who lead live sessions inside The Engine Room.