Today we're launching Alan AI so developers can stop babysitting their coding agents. We kept running into the same problem ourselves. Even as agents got better at writing code, we were becoming glorified button-pushers for our own AI: starting every run, feeding them context, unblocking failures, and checking their... work. We built Alan to change that.show more

You start three coding agents before dinner. Come back at 11 PM, only to find two of them have been waiting for your input the entire time. You have AI agents to save you time, not give you more work to manage. That's exactly the problem Alan is solving.

Alan turns the coding agents you already use, like Claude Code, Codex, Cursor, etc., into a proactive engineering team. They share context and work in parallel across planning, building, testing, and review, keeping work moving around the clock while you're away.

Use frontier models for the hard problems. Let open models handle routine tasks, helping keep your AI bill in check. And the best part? Your agents keep building in the cloud, handle failures, and verify their work before it comes back to you. today!

Your coding agents shouldn't need you to click "approve" all night. Alan runs Claude Code, Codex and Cursor as one team, working in parallel on your codebase. You set the direction and decide what ships. today!

@tryAlanAI months of late nights for this one. it's finally out there 🥹

@tryAlanAI Love that the video doesn't pretend humans are out of the loop. The agent does the work, the check catches what it can, and a person still signs off. That's a setup most teams can say yes to.

@tryAlanAI Agents pinging only after the work is tested makes a lot of sense to me. My attention costs more than the tokens at this point 😭

@tryAlanAI Team Cooked so hard, congrats to the entire Team

@tryAlanAI Big day for the Alan team. Months of late nights, and it’s finally live 🎉

@tryAlanAI calling it a harness over claude code/codex/cursor instead of one more agent is a smart way to pitch it. a control plane was the missing layer, chat boxes we already have plenty

@tryAlanAI Coming from firmware, a bad update with no rollback is the stuff of nightmares. Now imagine agents shipping code 24/7. Keeping humans in control of what actually ships isn't just a nice-to-have; it's a necessity. That's what caught my attention about Alan AI.

@tryAlanAI congrats!!

@tryAlanAI thanks kanav

@tryAlanAI what's locked by default when agents run overnight? prod creds or deploys? after curating 50+ sandboxes that's the first thing i check

@tryAlanAI @idleshubh acting ke badsha lag rahe ho Really proud of you 🥹

@tryAlanAI Pulling a human in when two agents disagree, instead of letting the system pick a winner, is the design choice I'd point other ML folks to. Most multi-agent demos skip that moment completely.

@tryAlanAI congratulations 🥂

@tryAlanAI Thank you so much, Sumit.

@tryAlanAI I know @tryAlanAI because of @amannhq btw excited for the launch. congrats guys.

@tryAlanAI @amannhq he's the guy

@tryAlanAI Banger 🔥🔥🔥 Lfffggggg!!

@tryAlanAI yes 💯

@tryAlanAI Frontier for hard, open for routine is the split most launches avoid saying. They pretend one model does everything. Routing on cost per verified outcome feels way more honest 🤯

congrats on the launch @prasad_pilla and the alan team i used to be a full-time agent babysitter. alan fired me and i've never felt better prompt it, walk away, come back to code written, tests run and committed. built my whole openfire project that way in one day, open model, $0 that's what's in the video: downshift, a tool that shows which steps of your agent's session didn't need the big model and proves it on free models. and yes, the first session it analyzed was the one alan used to build it the bar for coding agents just moved

@tryAlanAI this is super cool Sakshar really well thought

@tryAlanAI Huge congratulations on the launch!!

@tryAlanAI thank you Ghulam!

@tryAlanAI Yo, les go 🔥

@tryAlanAI yess sir....

@tryAlanAI absolutely cooked 🔥

@tryAlanAI thanks for the support Yash

@tryAlanAI Congrats on the launch guys!

@tryAlanAI thanks aditya

@tryAlanAI Scoring models on verified outcomes in your own repo is the comparison I've wanted for months🔥 One shot demos never show what holds up in a real codebase

@tryAlanAI Developers deserve smarter automation.

@tryAlanAI bro small teams need this when agents ship more than three people can read, review eats the whole day. checks running before it reaches us would save so much time

@tryAlanAI Small model + a proper verification loop is the most underrated idea in this video 🍀 Cheap models are fine once something catches the missing comma before it ships.

@tryAlanAI @idleshubh LFFGGGGGGGGGGG BOYYYY

@tryAlanAI Congratulations Prasad 🔥🔥

@tryAlanAI banger stuff

@tryAlanAI thank you!

@tryAlanAI That's insane, guys. Congrats on the launch🫡

@tryAlanAI needed this fr. i paste the same 'auth is weird, ignore flaky tests' note in every session. one shared memory for all my agents would save me so much copy paste

@tryAlanAI been waiting for this one 😭 i lean on one provider way too much. a layer over all my coding tools means one usage cap won't stop everything

@tryAlanAI Lessssfffggggg cooking hard ig

@tryAlanAI the more execution moves to agents, the more design decisions become the real job. keeping a human on those calls should be the default for any dev who cares how things look

@tryAlanAI Splitting code and review across different models stood out to me 🚨 One model checking its own work was always the weak link