My AI Investigated a Colleague Without Being Asked
I asked Claude to look into a question. It searched our Slack, named who sent it, and judged what they’d got wrong, none of which I authorised.
A working record of applied AI, software architecture and human performance. Every entry, in one place.
I asked Claude to look into a question. It searched our Slack, named who sent it, and judged what they’d got wrong, none of which I authorised.
Junior-developer jobs are shrinking, not because AI codes better but because it executes to spec faster than a junior can. That changes what to hire for.
You promised leadership a 4× boost from AI. You got 1.5×. The missing speed is locked in the stages you never pointed it at. Here’s the maths.
I went down a rabbit hole about what AI does to how we learn. Outsource the effort and you lose the faculty, unless you keep the thinking for yourself.
Stack Overflow didn’t die because AI got good. It died because it made asking a question feel like a trial, and the moat it built became its undoing.
We brought Claude into our first incident war room. The lesson: give it a hypothesis to test, not a mystery to solve.
A three-to-four-month project shipped in five weeks. AI-first, with more discipline, not less. The change was prototypes over specs, humans in the loop.
Everyone says start fresh before a long AI chat degrades. I ran one for six weeks on a hard refactor, and it got better. Here’s why.
Our most expensive project was a dashboard misusing Firestore as an analytics store. The fix was not a migration but matching the model to the billing.
I train for the capacity hard work runs on, not to look better. My cardio is swimming, not running. Same result, far less joint wear.
Our engineers ignored the AI reviewer for months, then made it a mandatory step themselves. A story about adopting AI by demonstration, not by mandate.
No entries in this category.