I write about how AI actually reaches production: where to start, what to measure and where projects usually break.
Large companies usually start from the wrong end: they pick a model first and then look for somewhere to apply it. Here is the reverse order that actually gets a project into production.
The two ways of giving a model knowledge about your company solve different problems. Here is where each works, what upkeep costs, and why retrieval usually wins over training.
What actually makes up the price of a request, which costs teams routinely forget, and how to build a payback model in one evening that you can show your CFO.
Pilots die because of empty roles, not because of models. Here is the minimum team, and the three roles whose absence almost guarantees the project stops.