Lol Claude, no.
“Excellent! All tests are passing now. There are some SQL errors in the logs, but they don’t affect the test results. Let me mark the todo as complete:”
-
-
Alex Standiford
Starting to form around a team of AI agents.
– Copilot is a junior dev, writing tests and bugfixes.
– Codex for reviewing code.
– Claude is senior dev. It writes new features, and more complex things in close coordination with me direct in my IDE.
-
Alex Standiford
It is INTOXICATING to watch my little setup find the highest priority function that needs unit tests, submit an issue on GitHub, then watch Copilot submit a PR with complete tests for me to review. This is supposed to free me up, but I’m just marveling at the whole thing.
-
Alex Standiford
I’ve finally managed to wire up a nice n8n -> copilot flow that automatically picks the highest priority method in Siren’s codebase to improve, and submit a pull request to improve it. Starting small with doc blocks, but I can see unit test PRs coming soon, as well.
-
Alex Standiford
Settling into a pretty effective flow – use ChatGPT for most research and generally as a “cheap AI” to gather the context, and prep everything to be built by the “expensive, but more capable AI”, like Claude.
-
Alex Standiford
I have chat pre-prompted to always push back on me and challenge my assumptions. Today it gave me this lovely one-liner:
“But “easy” is doing a lot of work in that sentence, so I’m going to poke at it.”
Fediverse reactions
-
-
Alex Standiford
Well, that was neat. I have a ClickUp integration that I use for an internal task system at Novatorius, but wanted something similar for Trello.
Claude Code just churned it out flawlessly, given the ClickUp example.
