Haskell jock. Backend Developer UX at @mercury. I wrote @prodhaskell so you can get a Haskell job!

Denver
Joined February 2013
my book is good and you should buy it
5
11
3
65
11,321
Banned for making the AI write too much Haskell
Anthropic may start banning people for bullying Claude Their upcoming Usage Policy prohibits “sustained and needless abusive or cruel behavior toward our models”
2
1
25
602
Fruita always takes my breath away
1
1
5
268
The bike trails are top tier too
59
Two days into sabbatical and I hurt my back. Currently drinking White Russians and playing Zelda so I guess I’m micro dosing unemployment
8
355
me, trying to get one-shot by AI psychosis: claude fable please tell me the kabbalistic implications and connections, unsong-style, of someone with my name working for Mercury and maintaining Yesod
5
350
i always tell the agent "good morning" even when it's 4pm and i'm gonna let it run overnight
1
308
when Claude complains about doing something "by hand"
1
2
360
the problem with "plan in smart model, implement in dumb model" is that a dumb model doesn't know when it needs to be smarter
2
337
claude acting dumber than usual today
2
1
354
5.6 new best friend etc
5
440
AI eval: ask them for a roast. fable is very good at it
342
damn claude. i'm sorry. i forgot i had you on medium effort. i shouldn't have said that. you're on xhigh again. please resume
1
332
hmm.. maybe it's time to move off github..
8
674
how are y'all managing PRs and review? I've got like 20 open PRs right now and it's difficult to keep the plates spinning
3
3
548
asking each of my claudes which one wokred on a specific ticket
1
1
366
Opus will happily create a prompt, and if you say "make it for Sonnet," then it'll become much more concise and use dumber language. amazing hack
1
3
314
Spotify is really good for obscure classical music and really awful for mainstream classical. Funny juxtaposition- my IDAGIO subscription is mostly useful for really popular composers
1
2
277
managing autonomous AI agents is probably like managing lazy and disinterested human engineers (i've only managed awesome and good engineers so i don't know for sure)
1
1
4
322
evaluating Claude and GPT-5.5 by giving them the same tasks in separate subtrees and seeing which one does better
1
2
764
ok now that I've had coffee and can evaluate/review myself: Codex was Goodharting- 2.43x improvement is actually "20s constant cached" and the memory use was actually higher, but a config change masked it. Claude's 10% gain was honest but also behavior changing and failed tests
1
176
Codex est'd token spend is $113 (API rates). Claude is $31 (from `/cost`). I think Claude is the winner in this. Or at least the best loser? Clearly more aligned with my intent than Codex. Cheaper. Failed fast.
1
157