@owen@mastodon.transneptune.net
Post #4300307
2026-07-28 19:56 UTC
@raganwald@social.bau-ha.us Sure, and that's valid. Speaking personally, what I have seen is an institutional embrace of big design up front - the bigger, the better - as a response to the structural deficiencies of statistical code generation. "Surely, with a complete-enough spec, an LLM will _finally_ generate code that does what I meant, and has no deficiencies I can't tolerate," realized breathlessly through things like BMAD.
Replies (1)
-
@raganwald@social.bau-ha.us 2026-07-28 19:59
@owen@mastodon.transneptune.net That particular one interests me, because I recall a conversation over BEvERages a decade ago where we played "what if compute was free." One of the things we considered was what if we could randomly write code at planetary scale. We could have a billion digital monkeys writing gibberish, and filter the results with test suites. And now the problem becomes, "How do we write a test suite that complete enough that it can't be mocked into compliance?"