Hello
This is where I will write. A short note to open the page, partly so it is not empty, and partly because I have wanted somewhere to think out loud for a while.
I build AI products inside a large enterprise software company. The work is more interesting than the job title suggests, because the hard decisions are almost never about models. They are about what problem you are actually solving. Deciding that "find the right colleague" is a graph problem and not a search problem is worth more than any amount of prompt tuning, and it is the kind of thing that has no obvious home in a ticket, a spec or a sprint review.
So most of what goes here will be decisions. A choice I made, the alternatives I turned down, and what happened. Some of it will be about agentic AI and evaluation and retrieval, because that is what I spend my days on. Some of it will be about the less glamorous parts of enterprise product work, like getting ten large customers to agree on what a word means, or persuading a team to ship a smaller thing sooner.
I will try to be honest about the ones that did not work. A portfolio of successes is not very useful to anyone, including me.
If something here is wrong, or you disagree, tell me. That is the point of writing in public rather than in a notebook.