
Maso222
167 posts



@_ARahim_ @bcherny only boomers fix typos in prompts. llms perfectly understand you even if you mistype.




A DEVELOPER ONCE EXPLAINED A LANGUAGE FROM THE 1980s THAT ALREADY RAN MILLIONS OF TINY WORKERS IN PARALLEL UNDER ONE SUPERVISOR -- THE EXACT THING EVERYONE THINKS "300 AI AGENTS" JUST INVENTED Tim McNamara on what to steal from Erlang -- the language built for systems that can never go down, where thousands of independent processes run at once and a single supervisor watches every one of them. -> The moment it clicks, the whole "Swarm of agents" hype stops looking new. One coordinator that plans the work. A crowd of small workers that each do one thing and share nothing. A supervisor on top that kills and restarts whatever breaks. That's not a 2026 idea -- it's the actor model, and it's older than most engineers reading this. The trick was never raw parallelism. Anyone can spawn a thousand workers. It's the discipline around them: each one isolated, failures contained instead of cascading, and one layer with the authority to restart. "Let it crash" beats "Try to handle everything" because the supervisor already knows what to do. Running many things at once was never the skill -> orchestrating them so the whole thing doesn't collapse is. And now that people are pointing 300 AI agents at a single job and praying, the ones who win will be the teams who learned this pattern from a language that's been doing it for forty years. Everyone's racing to build the swarm. Almost no one is studying the one system that already solved how to keep a swarm from eating itself. Save it. It's the map for everything coming next ↓




Claude bet me they could write a complete short story using only 6 words. ‘For sale: Legacy code. Never reached.’


















