A repo called reverse engineering is over is making the rounds with a big claim: LLMs have ended reverse engineering as a high barrier skill. The case study behind it: Claude Opus 4.6 on a Max 5x plan, plain web chat,…
Came back to a session after a few minutes and it had moved on without me. In the transcript there is a question with three options, and directly under it a line reading . It then picked one of the options on its own…
Updated the Codex CLI and now nothing it tries to run actually runs. Chat still works. It answers normally. But the moment it wants to touch the shell I get back: Other times it surfaces as . It is not selective. fails.…
I had a large body of text written with a model and wanted to know how obvious that was, so I measured rather than guessed. Ran the whole published vocabulary list people watch for against it. Clean: zero instances of…
Went shopping when my usage stopped fitting. Most comparisons are about the model, and after trying several the model was the smallest difference. What actually differs What it decides to read. The biggest gap by far.…
Persistent memory sounded like the feature that would make an assistant genuinely useful and for about six months it made mine worse. Everything I told it stayed, including things that stopped being true, and I ended up…
Every session I end up saying the same things. We do not use that logging library. Tests go in this directory, not next to the source. Do not add a comment above every line. I say it, the session gets it right, and the…
There is a clear pattern in recent launches: harnesses for multi agent coding workflows, small crews of specialised agents, approval first setups for one person teams, long horizon terminal agents. My experience with…
A product appeared this week specifically about keeping AI context portable between tools, which tells me I am not the only one with this problem. Every assistant I use has accumulated a setup, instructions about how I…
Long refactoring session, one conversation, big monorepo. First hour is excellent. Somewhere around the point where the context indicator says 60 70k it starts: reintroducing a bug we fixed 40 minutes earlier referring…
I work on a backend split across a lot of services. What I need from a coding assistant is mostly "read the existing docs and code and tell me where the thing I want already lives", far more than "write this function".…
Career changer, two months into a junior role, and I am closing tickets at a rate nobody has complained about. The problem is that in review I cannot explain why the code is shaped the way it is, and last week I could…
Asked it to fix an off by one in a pagination helper. It fixed the off by one. It also renamed four variables, extracted two functions, turned a for loop into a reduce, added JSDoc to everything, and changed the error…
Team of four, we merge roughly 15 PRs a week and maybe 70% of the code in them is model written. Test suite is decent, 1,100 tests, about 6 minutes in CI, covers the money paths well. Two camps internally. One says…
Five developers, everyone using agents, and our PR count went from about 12 a week to just over 40. Nothing else changed, so review is now the constraint and two of the five are visibly burning out on 900 line diffs…
Six years of application development, mostly TypeScript and Python, and I had planned to spend the autumn properly learning a lower level language because I have always felt the gap. Now I am not sure that is where the…
Inherited a service with no tests. Asked for coverage, got 40 Vitest files, 380 tests, all green, 84% coverage. Felt great for about a day. Then I deliberately broke a validation rule: changed a to a in the middle of…
60k lines of TypeScript, five years old, and we want to move from a hand rolled DI container to plain constructor injection. Touches roughly 180 files. Nothing about it is hard, it's just enormous and boring: exactly…
Been using an agent in the editor for about four weeks. Asked it to add a field to a form and wire it to the API. It read 14 files, ran the test suite twice, grepped for four different things and made six edits: roughly…