@sinclairinat0r I have a lot of respect for Anders given what he has achieved but also how clearly he talks about the subject with such pragmatism. Young developers should learn from his attitude.
@m_zokov Searching through documents on the fly wastes tokens and takes time.
RAG is an optimisation pipeline that happens prior to a user searching for stuff (text, semantic or ideally both). Once you have a shortlist of matches you can then throw those at the LLM for a deep dive.
Only recently, I started asking Opus 5.5. to review the specs it created. It seems to take at least 5 iterations to stop finding issues with it's own creation and corrections. Each review is with a fresh context. What am I doing wrong?
@0xrwu Are you running multiple agents at the same time? How are you running those valuable-but-bulky tests in parallel with running out of cpu and or memory? Genuinely interested.
@LondonJazzBass@dhh I totally agree with that now but at the current rate of model improvement this may not be the case in a year or so.
If the software remains performant and quick to change, who will care what the code looks like?