Just Replaced 30 engineers with 30 agents to build an entire codebase in 12 days The agents handled everything: - 44K lines of TypeScript - 175 PRs opened - 1,500+ tests written - All CI failures self-corrected fully open source, link to code below

try it out here:

Sounds like bs, so you sacked 30 people to build something you gave away for free? Were you high at the time or did you fall down and hit your head?

44k lines in 12 days is impressive but the real test is month 3 when someone has to add a feature the original agent never anticipated — curious what your change failure rate looks like now that humans are back in the loop.

Hey @KaranVaidya6, this agent-orchestrator build is wild! Made a FileCity Tour of your project. Visual map + narrated walkthrough

I get that’s cool and I see the future here… but what did they produce in the end? Was the code usable? Did they come up with a functioning product?

Just scanned through the replies, only one shown concern on those 30 people who got replaced. All others commenting like a bot. Time to create an agent that comment like a human with empathy.

Impressive numbers but 44K lines of TS in 12 days with 30 agents means ~50 lines per agent per day on average. Real question is how much of this was actually reviewed or is the codebase a mess waiting to break in prod? Not doubting agents, building with them myself. Just the test count and PR volume doesn't tell the full story.

We’re rapidly approaching a point where a "Senior Engineer" is essentially a Swarm Architect, spending less time in the IDE and more time orchestrating agentic workflows and reviewing PRs at scale. This project is a massive shot across the bow for traditional dev shops. If 30 agents can out-build 30 engineers in a fraction of the time, the "cost per feature" is about to fall off a cliff.

the part that fascinates me is the coordination. how do 30 agents not step on each other? managing a team of humans is hard enough

this is actually insane 🔥 30 agents shipping 44k LOC, 175 PRs + self-healing CI in 12 days is next-level orchestration. The one missing piece for real production/enterprise use is strong governance. Exactly why I built CORE: Immutable constitutional rules + hard enforcement (Mind/Will/Body) so agents stay auditable, deterministic, and can’t break architecture no matter how autonomous they get. Running your fleet inside CORE would be a killer combo — massive velocity + unbreakable safety. Happy to explore!

the CI self-correction part is the only thing here worth paying attention to. 44K lines and 175 PRs means nothing if nobody can maintain it in 6 months. the real question is who's debugging the agents when they go sideways.

Nice graph. Is it actually useful?

1,500 tests is impressive until you check what they're actually testing. ran a similar setup last month—agents write great happy-path coverage but miss edge cases that break at 2am

Guess what hope fully you come out from your cuckoo world of AI agents 😂 how about you getting replaced by AI agent too near future 😜 -

Guys please dont fall for hype.. no large scale organization is approving these vibe coded apps... No serious org is ready to break the chain👇 SECURITY == CONSUMER TRUST == BUSINESS

wait you replaced 30 engineers with 30 agents... the headcount didnt even change lol

Agents don't accumulate institutional knowledge. Human engineers compound over months — codebase-specific intuitions, edge cases, unwritten rules. Agents reset per task. You traded compounding capability for sprint output. Not the same thing.

44K lines of code and just 1500 tests? You code is shit

Impressive results. The key question is: what kind of engineering work was it? Scaffolded greenfield vs legacy refactoring are very different problems for agents.

Replace 30 engineers with 30 agents and you just traded a management problem for a massive orchestration problem. The bottleneck shifts from headcount to state management and conflict resolution. Brave move though.

I'm getting tired of these robo-githubs. You have no idea if anybody is even using it, all "discussion" is agents submitting bug reports.

“Replaced” is interesting here. The trillion dollar question is how many companies will shrink teams vs keep the same team but 10x the output.

Who's the manager? how does the individual agents reach settlement?

All this talk of the swarm - show is the codebase that it produced. Where is the website?

Looks like a virus spreading.

The CI self-correction loop is the part that interests me most. That's where most agent coding setups fall apart — they can write code but can't debug the integration layer. What was the failure rate before agents started self-correcting, and how many iterations did it typically take?

curious what the actual review loop looked like. 30 agents, 175 PRs, 1500 tests — but who caught the edge cases? who set the product direction? the agents did the volume, but i'd want to know what the human-in-the-loop moments were

30 real engineers or you just think that you actually replaced 30 engineers

And what exactly did they build?

those 30 engineers are watching this tweet right now

I’m writing 20k per day with one agent…

What was coded by 30 engineers before? 🤣 LIAR 🤣

W

Your $200k+ TC job is now a prompt away from obsolescence.