Haiku 5.5 swapped in for Luna, and it got worse
ThePrimeagen changed one string in a config file to point his automation at Claude Haiku 5.5 instead of GPT-6 Luna, and says it performed significantly worse and significantly slower. The cheapest model is not automatically the cheap option.
The big disagreement
Elementary OS extends AI ban to code, docs and artwork, DHH calls it Luddism
Elementary OS has extended its ban on AI to include code, documentation, issues and artwork, according to The Lunduke Journal, which also notes the distro has banned using AI to write pull request descriptions. DHH's response was one line: hard to know whether to laugh or cry.
It boggles the mind how you can love computers and not be amazed at what we've made them do.
Receipts
ThePrimeagen swapped Luna for Haiku 5.5 and says it got worse
ThePrimeagen ran Haiku 5.5 against GPT-6 Luna on some automation work, both with reasoning set to none, and says Haiku performed significantly worse and was slower on both passes and failures. The only change was one string in a config file, so he rules out it being an artifact of how he was calling the model.
Haiku, literal drop in replacement (it was a one line change of changing a string in a config file), performed significantly worse
How they actually work
Pocock's prompt of the day: ask the agent what it wanted to ask you
Matt Pocock's prompt is "/wait-what questions did you have for me", run after an agent has been working on its own for hours. The point is that agents leave questions scattered through the context window instead of stopping to ask, and this collects them in one pass.