
How do we actually know if an AI feature is any good?
One misleading running insight becomes a repeatable AI eval, and a reassuring score hides the failure that changes the release decision.
10 posts tagged “AI and Work”.

One misleading running insight becomes a repeatable AI eval, and a reassuring score hides the failure that changes the release decision.

In one room I sound like a shill and in the other like a sceptic. Both sides of the AI argument arrive with the answer already decided.

What is an AI agent? A plain guide to the model, tools, runtime, permissions and handover that carry work across more than one step.

AI can remove routine tasks while leaving people with harder exceptions and fewer chances to build judgement. The easy work was carrying part of the job.

Detachment is a rational response to corporate work. It also dulls the judgment you were protecting. On caring about the craft rather than the company.

One ordering choice reveals what somebody has been paying attention to. Why leading with reasoning matters more now that execution has become cheap.

A Corolla, a Porsche and a Range Rover. A practical way to choose between AI tools, plus where the analogy breaks and how to test them on your own work.

Every project runs on two layers. One is documented and being automated fast. The other decides whether the project lives, and it was never written down.

Language models are built to return the average. Taste is judgment shaped by evidence only you have, which is why it holds value when tools are shared.

Snap judgments feel like discernment and work like a cage. What writing someone off in five minutes costs you, and why AI removed the usual correction.