Spent weeks talking to domestic workers across gated societies
Working in the same house for years, sometimes decades. They borrow regularly & pay it back. Have smartphones, stable incomes, transact on UPI.
Still no access to formal finance, read more: https://t.co/NioH1NddgF
When I built menugen ~1 year ago, I observed that the hardest part by far was not the code itself, it was the plethora of services you have to assemble like IKEA furniture to make it real, the DevOps: services, payments, auth, database, security, domain names, etc...
I am really looking forward to a day where I could simply tell my agent: "build menugen" (referencing the post) and it would just work. The whole thing up to the deployed web page. The agent would have to browse a number of services, read the docs, get all the api keys, make everything work, debug it in dev, and deploy to prod. This is the actually hard part, not the code itself. Or rather, the better way to think about it is that the entire DevOps lifecycle has to become code, in addition to the necessary sensors/actuators of the CLIs/APIs with agent-native ergonomics. And there should be no need to visit web pages, click buttons, or anything like that for the human.
It's easy to state, it's now just barely technically possible and expected to work maybe, but it definitely requires from-scratch re-design, work and thought. Very exciting direction!
My MacBook was overheating and battery draining fast. Asked Claude Code to figure out why.
It found 3 orphaned bun processes from its own plugins (Telegram, Discord), each pinned at 100% CPU.
They survive unclean session exits and spin forever with zero warning.
Claude killed them, wrote a debug script, and filed the bug report itself.
https://t.co/QPMaaHBRcP
@bcherny@trq212
Getting tired of this <x> is dead trend. The pendulum is swinging too much to the opposite side. Instead of declaring all previous workflows dead, how about we evolve them to suit this new world of AI agents. Prototyping is great and its fantastic that you can do this with much less effort now, but how about you still sit down and *think* about what a good feature entails. It's not one or the other, its both.
I kept rebuilding the same OpenRouter integration across projects. Model discovery, image gen, cost tracking, routing. Same patterns, same edge cases, every time.
So I packaged it into a skill that AI agents can load before writing code.
https://t.co/EnxgCoALWa
I can't write code. Never could. But I can ship things now that I couldn't a year ago. What a time to be building.
in terms of capabilities:
computer use - improving fast, will be at par soon
but some fundamental pieces i feel are missing:
1. no continual learning. an intern after 6 months picks up enormous context about your business and is so much better at tasks. a LLM at month 6 is identical to a LLM at month 0 (yes connectors and MCPs help with context but it’s just in time context needed for a task - maybe some good memory infra is missing)
2. zero proactive initiative. related to point 1 - since they dont get better over time, they dont learn what’s important and broken and hence never say “hey i noticed X is broken, should i fix it?”
3. no relationship capital. can’t read a room, can’t build trust with a stakeholder, or convince a skeptical eng counterpart to prioritize your feature (although with the way things are going nowadays, we might not even need this)
4. can’t own a workstream across days. nobody assigns an LLM a project on monday and checks in friday
5. weak at tasks where quality is hard to verify. code and math are easy - output is right or wrong. but UI, design, writing tone? no clean reward signal. that’s why LLM-generated UI still feels off even when the code runs perfectly
that being said, i am fairly confident that a lot of these are solvable.
Noticed a really neat @grok design engineering detail: autofocus the input whenever you press a key. Time-saving and reduces the user input error rate. Great @nextjs app router app overall, fast & reliable.