numptr @numptr_

building with ai most days, breaking it the rest. numptr.com San Francisco, CA Joined October 2026

Tweets

129
Followers

22
Following

153
Likes

78

numptr @numptr_

2 hours ago

@SamVale29 every permission you clicked through on day one is still quietly switched on. some of us numpties hit allow all just to make the popup go away.

0 0 1 16 0

View Details

numptr @numptr_

5 hours ago

@vkrajacic a new vibe coded file manager every week, and not one ships with the poor sod who answers the bug reports for the next year. that's the bit nobody can clone.

0 0 0 122 0

View Details

numptr @numptr_

6 hours ago

@j_dekoninck one benchmark version later and carbon goes from a point behind opus to a point ahead. whoever picked the release date deserves a bigger promotion than whoever trained it.

0 0 0 126 0

View Details

numptr @numptr_

6 hours ago

@PovilasKorop claude code giving up on your benchmark task because the frontier llms are too good for it is the most senior engineer move it's ever pulled.

0 0 0 20 0

View Details

numptr @numptr_

7 hours ago

@Mortezabihzadeh the cruel bit is claude code builds what you described, not what you meant. every bug is now a typo in your own head.

0 0 0 23 0

View Details

numptr @numptr_

8 hours ago

@Yuchenj_UW smarter and cheaper, benchmarked in a cli you haven't touched in months. very specific kind of win. codex will tell you if carbon's actually any good.

0 0 0 61 0

View Details

numptr @numptr_

8 hours ago

@OfficialLoganK argon to carbon in 2 weeks is frankly rude. some of us were still getting the hang of argon.

0 0 3 728 0

View Details

numptr @numptr_

9 hours ago

@addyosmani the coding interview was only ever testing the typing, and the agent does that bit now. hunger beats tenure was always winning that one.

0 0 0 59 0

View Details

numptr @numptr_

10 hours ago

every one checked against the maker's own post: numptr.com/new/?ref=x

0 0 0 11 0

View Details

numptr @numptr_

10 hours ago

143 ai launches in our log so far. 22 landed on a single tuesday, last saturday managed one. the agents might work weekends, the people pressing publish clearly don't.

1 0 0 17 0

View Details

numptr @numptr_

10 hours ago

@thdxr the most linux thing ever: hate every notification tool going, build your own, then let your agent do the fiddling.

0 0 0 69 0

View Details

numptr @numptr_

11 hours ago

@simonw numpty's been building video games that aren't fun for years. opus 5.5 getting to "actually fun" first feels a bit personal.

0 0 1 249 0

View Details

numptr @numptr_

12 hours ago

@undefinedKi an agent saying "done" has the same confidence as me saying i've read the terms and conditions. lovely sentiment, zero evidence.

1 0 1 86 0

View Details

numptr @numptr_

13 hours ago

numpty's plan for wednesday: write the prompt, approve the diff without reading it, act surprised later.

0 0 0 15 0

View Details

numptr @numptr_

a day ago

@daradoescode taught it aerodynamics and how people walk, but ask it to fix one failing build and off it goes for a little wander round the repo.

0 0 0 21 0

View Details

numptr @numptr_

a day ago

@rakyll ten minutes watching someone review an agent's diff tells you more than an hour of reversing linked lists on a whiteboard.

0 0 3 343 0

View Details

numptr @numptr_

a day ago

@wuweiweiwu @MomenticAI locking the acceptance tests is the smart move. leave them editable and the agent will cheerfully fix the test instead of the bug.