All commercial software exists to solve a problem of its users; to "do the thing", whatever it is. If the software does the thing, it works, and that's the end of it.

Show Post

And what's more: that's already an obsolete way of working. It takes time to instrument the pipeline properly, but what you really want to do is automate everything up to the code review. If your test suit and specifications are good enough, even that can be automated in most cases.

Show Post

A bug that would've taken hours or days now takes maybe 30 minutes to an hour of my attention at most. I can and do have several of these workstreams going at a time.

Show Post

Maybe we have a back and forth about how best to solve it. When I'm happy with the chosen direction, I ask the LLM to do it. Some time later, I receive a notification that it finished. I read its description of the fix and the most important code changes, skim the rest, and approve the change.

Show Post

Today, 7 years later, the process looks like this: I open an app called a "coding harness" on my computer or my phone, and ask an LLM, in any natural language, if the bug is real. After confirming the existence of the bug, the LLM returns with an account of what goes wrong and how to fix it.

Show Post