Essays on the intersection of organizational knowledge, AI architecture, and the future of work.
Every association and organisation that uses AI should write down how. With the Café IA method that takes an hour, a circle of chairs and a few good questions. Here is how to run one and turn it into your AI charter, with tips for getting moving.
In Toulouse this morning, Céline Chantry-Daron showed how a gap in the data ends up as a wrong decision on a screen. Medicine makes it easy to see. The same thing happens in any company where a few people build the second brain for everyone else.
At dotAI in Paris last week, Aygalic Jara showed that a language model often knows when it is making things up, and never tells you. Why that is the first thing to learn about AI, and how Praxis teaches a team to check without reading everything twice.
Connectors, projects, skills and routines, demonstrated live on a real business case. The steps, the prompts to copy, and an FAQ that completes the answers we gave during the session.
Jev, a new model from TypeSafe AI, doesn't generate text. It only makes decisions. For a small business that may matter more than the next chatbot. Where it fits, where it doesn't, and how to try it without betting the shop.
Moonshot AI's founder gave his first proper interview in eighteen months and spent it on first principles instead of the business questions everyone wanted answered. Three of his claims are worth stealing.
A principle. Why the person plus the machine is always the best combination - and why bringing down the barriers as low as possible is the only logical move.
Brian Madden has built a public knowledge system that lets AI draw on his thinking in real time. The same methodology sits at the heart of Praxis - and he deserves the credit.
A Lenny's Newsletter article made something click for me about why most AI training fails. The answer was already in how Praxis was built - I just had not said it clearly enough.
Anthropic's Project Glasswing scanned over 1,000 open-source projects and found 6,202 high or critical vulnerabilities - at 90% accuracy. The problem is not the finding. It is how long it takes to fix.
Before the tools matter, the knowledge has to be there. Three things that need to happen in roughly this order in any serious AI adoption.
How we built a self-hosted document intelligence platform that gives litigation teams AI-powered review without a single byte of privileged material ever leaving the building.
A practical walkthrough of MCP - the open standard quietly transforming how AI connects to business tools - and a real engagement where we built it.
Anthropic's landmark study of one million AI conversations reveals which roles are most exposed to AI - and what businesses should do about it.