Replit Missions: an agent swarm that works to a budget
Set a goal, success criteria and spending limit; a coordinated swarm of agents plans, builds, tests and iterates until an AI judge signs off or the money runs out.
Why it matters
The first agent product that treats budget as a hard constraint — work stops unresolved when the money runs out — and it rebuilt a $83,700/year observability platform in two days on $3,000 as its proof demo.
What you could build with it
A small SaaS team could hand a Mission the goal of replacing a legacy internal tool — success criteria and a $500 cap set in advance — and get a working replacement or a clean stop with the spend accounted for, instead of an open-ended agent bill.
Does it hold up?
Too early to judge: the only public evidence is Replit's own controlled demo (DataCat in 2 days on $3,000); no independent team has reported running a Mission yet, and the judge's pass criteria are unproven outside Replit.
Built with Replit Missions
- Replit launches Missions to coordinate agents under spending limitsarticle · Coverage of the launch: an AI judge checks results against agreed criteria, and if the budget runs out first, work stops with the outcome unresolved.
- Introducing Replit Missionsarticle · TLDR summary of the launch: Missions built the DataCat monitoring platform in two days within a $3,000 budget, and DataCat itself is open-sourced.
- Replit Agent can now spend your money: 2 ways, 1 riskarticle · A skeptical read of Replit's spending controls from the Nov 12 agent-commerce launch: buying stops at checkout for approval, but ad budgets are soft caps — useful context for trusting a budget-capped swarm.
Learn more
First spotted on article: source.
More AI agents
CLI-Anything: make any app agent-native with a generated CLIFrom HKU's Data Intelligence Lab: a seven-phase pipeline that generates a real CLI harness for any GUI application…agents · JEV 0.78DeepSeek Harness: 1.54M daily users and 35,000 plugins into its everything-is-a-plugin runtimeDeepSeek's open-source agent harness where models, tools, sandboxes, sessions and even the UI are all plugins — now…agents · JEV 0.76Dify 1.20.0: OpenTelemetry trace export and malware scanning for LLM appsThe open-source LLM app platform's 1.20.0 release adds trace export to any OpenTelemetry collector, keeps Langfuse…agents · JEV 0.76AG-UI: the open protocol connecting AI agents to user interfacesEvent-based protocol that streams agent activity — text, tool calls, state — to any frontend over 16 standard event…agents · JEV 0.74PaperOrchestra: five agents that turn lab notes into a submission-ready paperGoogle Research's multi-agent framework turns an idea file and an experiment log into a full LaTeX manuscript…agents · JEV 0.74Superpowers: the skills framework that makes coding agents work like careful developersA free Claude Code plugin of 15 skills that forces a disciplined workflow — brainstorm, plan, test-first, review…agents · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.