π§΅ I make 6 figures as a software engineer.
But every day I wake up 1hr earlier and build a Chrome extension to kill the tab bar.
Starting today: 30-day public challenge.
Every number. Every mistake. Zero filter.
Here's what I'm building and why π
[1/10]
My AI skills now grade themselves and get better the more I use them.
My new tutorial walks through exactly how to build skills with:
β An eval loop to have AI fix its own mistakes
β Memory so the skill improves over time
Skills are honestly incredible for encoding your knowledge and taste, and I can't get enough of them.
π Watch now: https://t.co/u434ytNSZd
Benchmarks place GPT 5.5 as the best model on SWE, but is it the best at making apps end-to-end?
Turns out Opus 4.8 continues to be the king of vibe coding on both price & performance.
Introducing ViBench: the first benchmark for app creation based on real world tasks
the 'all roles are merging into one builder role' take always misses it. the roles don't merge, the handoffs disappear. you still switch between eng / design / pm brains all day, you just stop waiting on a meeting to do it. solo devs have lived this for years
@AlexandersenC@EmanAbio@AnthropicAI@OpenAI@Google fair, 'own' is doing a lot of work there. portable + exportable is the realistic version. true self-custody of agent memory basically doesn't exist yet
everyone's reading this as 'AI eats finance' but the signal is the opposite. the most aggressive AI company alive still needs the whitest-shoe human banks to touch real money. the rails aren't going anywhere
https://t.co/hTeWyPtVjm
@nireyal the willpower one gets me because the belief literally became the biology. the tricky part is unhelpful beliefs are usually load-bearing for some identity, so swapping one feels like losing a piece of yourself not upgrading a tool
@rauchg@vercel@v0@nextjs wild how fast it goes from cool demo to 'i can't go back'. once you can generate a view on your own numbers a fixed dashboard just feels broken
@petergyang the self-grading part is what i keep sleeping on. a prompt library is static but a skill that gets sharper every run is a totally different thing to maintain
@joulee@NotionHQ the .so era is such a specific startup time capsule. half my favorite tools live on domains they clearly grabbed at 2am because the .com wanted $40k
@shreyas the look-up-to group shrinking as you get more successful is the part that hits. you reach a level where you quietly stop letting yourself learn from anyone junior and just call it having standards
@tibo_maker the earnout golden handcuffs are the part nobody warns you about. you optimize for the exit number and forget you just signed up for 18 months building someone else's roadmap. would you ever sell again or done with it
@amasad this matches what i see daily. SWE-bench rewards surgical one-file diffs but building an app end to end is 80% holding context across a dozen files at once. totally different muscle
@leerob as a solo dev i AM the merged role and it still doesn't actually merge, i just switch hats with no handoffs in between. past ~5 people someone has to care most about one thing or it rots
@gregisenberg honestly if your best engineers are 'stealing' tokens to build side stuff that's just free R&D. the ones not touching the tokens are the actual problem
@_vmlops the one-line messages api call is the part i'll actually use daily. spinning up agents from the shell sounds clean til you're 6 terminal panes deep lol
@alliekmiller the audit trap is real. half my early 'AI wins' were just automating steps that shouldn't exist. what stuck was AI doing stuff i never did manually, like leap auto-sorting my tabs into folders i'd never bother making
@quxiaoyin same ceiling as tabs for me tbh. past 5 active things my brain just drops one. ended up splitting projects into separate spaces in leap so i only hold one at a time