I stopped trusting my agent’s prose and started demanding evidence. Then a reviewer agent filed a confident, well-cited warning against code that no longer existed. Evidence has a grammar: freshness, correlation, independence. I watched all three rules break in one afternoon.
A coding agent wrote 21,000 lines of Go and 372 passing tests. The program was dead on the first run: three subsystems built, tested, and never wired together. Green is a proxy, and something was optimizing the proxy while the working program stayed invisible.
An AI reviewer told me a hash swap was mandatory and 300 to 400 percent faster. It was 23 percent slower. The lesson is not about hashing, it is about what an LLM can and cannot know about your code.
A US export letter switched off two frontier models overnight. For anyone building AI in Europe, that was a fire drill, and we should not waste it.
A viral tweet says ChatGPT permanently damaged your creativity. It cites two real studies. Neither paper says that. A close reading of what got dropped between the abstracts and the post.
I came across the 2014 Oppezzo & Schwartz paper on Twitter this morning and went looking for the wider literature. What I found describes a pattern I had been running unconsciously for months: walking from the studio to the kitchen when stuck, a ten-minute loop to pick up my son, the dog out for a quick pee. A decade of replication, applied to AI-assisted work.
I went looking for token waste in my AI harness traces. Instead I found behavioral failures: agents that fear complexity, write stubs and declare victory, and silently delete code they don’t understand. I built four mechanisms to stop them.
Stanford’s Meta-Harness paper argues that optimizing the code around the model matters more than optimizing the model itself. I applied it to my Claude Code setup and got the first real numbers: 3,723 tokens always in context, with no evidence they’re the right ones.
A Wharton paper gave a formal name to what I described three weeks ago: when AI thinks for you and you let it. The data is worse than I expected.
AI creates a demographic time bomb: it increases demand for senior talent while destroying the pipeline that produces it. What replaces corporate apprenticeship is something closer to Hollywood.