Funny how AI and agents were meant to make us work less.
Every engineer I know is working harder than ever. And every other post says software is dead.
If it's dead, why is everyone this busy?
I do this, i also have a claude session running on a server inside tmux + ttyd. My wife started complaining that you are working even when I am out -> so had to reduce it quite a bit. They are addictive. Its fascinating to me that AI agents were suppose to help human do less work.
@nithinmatha Would love to connect. I have been working in AI Video for last 6 years. Led Engineering at Invideo, helped Rephrase(Acquired by Adobe optimize their ML pipeline and then helped build the agentic AI video platform at Notch (Acquired by Adobe).
@DerickCarneiro_ Hey Derick, would love to connect. I was one of the core person behind Notch AI (https://t.co/08DcdM31LY) which got recently acquired by OpenAI. https://t.co/f8MTIbi3lE?
Curious, how do you build shared understanding of what is built by the agents , for me apart from code review , one of the most important purpose of PR was a unit of work to build shared understanding of what is getting added/modified. You could still do it at commit level, but I think PR's are great checkpointing mechanism.
@bradneuberg I believe code review isn't just for quality but also to build shared understanding of how software behaves. With AI shipping bigger Prs both is getting harder but not seeing people focussing on second problem too much
@walter_h_g_ Sounds very interesting -> close to what I am building . We have built and deployed similar models at small scale SaaS companies 15-25 Engineers and have seen massive impact. Would love to chat.
@useminionix I worked as a Principal engineer at SPC backed startup working on an autonomous performance marketer and video ad generation platform for a startup that got acquired by OpenAI . What you are doing sounds fascinating would love to chat.
@arpit_bhayani Loved how @dhh presented the same thought in his Ruby on Rails Keynote recently https://t.co/Sh0VWztMjM
Especially where he talks about that we are at the precipice of one of the most fundamental changes in Computer Science and its super exciting
@azerkoculu we landed on the same thing for our video agent, after trying a few other approaches. models just know how to use bash better -> better perf/accuracy.
curious how you're separating real assets from scratch ones, or how you're surfacing filesystem changes at the application layer
@dexhorthy Still think there's a layer that isn't derivable from the code intent, edge cases that only exist because of a business decision, design tradeoffs, domain concepts.
I keep those as ADRs + learnings today.. Curious what others are doing.
@dexhorthy This is so true. Keeping the documentation/spec/context in sync with the code doesn't work. Add to that most people don't even have a way to update these docs/context regularly.
AgentLoop: a daily digest on AI agents & harness engineering. Today's full issue, no spin 👇
https://t.co/gSU0orM6DX
Subscribe free: https://t.co/dusoJPN64s
Anthropic studied 400K Claude Code sessions. Headline: non-engineers succeed at nearly the same rate as engineers.
Everyone read it as "you don't need engineers anymore."
Same week, Meta shipped AI-written, AI-reviewed code with no human in the loop → auth bug straight to prod
3/ That system has a ceiling. Tests catch a lot. They don't catch whether an auth flow is actually safe.
The real skill now: knowing where your review loop ends and human judgment has to take over.
Where do you draw that line?
@tibo_maker Worked on this at https://t.co/TwMZczvoIU for a while — wild to see how far the models have come. The hard part for me was repurposing ads with heavy motion graphics, fast cuts, or multi-speaker setups. Curious where you landed on those?