Most writing about AI and the professions is pitched at one of two altitudes. Either it is a forecast about what the work will look like in ten years, or it is a product announcement dressed up as analysis. Neither helps much on a Tuesday afternoon with a deadline on Thursday.
This is aimed at the space in between: what these tools actually do when you point them at real work, where they fall over, and what it takes to get output you would be willing to put your name on.
Who this is for
People who bill for judgment — lawyers, accountants, consultants, and the people running the practices they work in. The constraints that make this audience interesting are the same ones that make the generic advice useless: client confidentiality, a duty of competence, and the fact that a plausible wrong answer is worse than no answer at all.
What to expect
Specifics over speculation. Where possible: what was tried, what it cost, how long it took, and what the failure modes looked like. When something did not work, that gets written down too — a tool that fails on a particular task is more useful to know about than one that succeeds on a demo.
More soon.