Key takeaways

  • Replit launched Missions on November 17, which runs parallel AI agents toward a goal within a budget the user sets.
  • Users choose which actions need approval, and an independent AI judge decides when the goal is met.
  • Replit has not published plan availability or pricing for Missions beyond the user-set budget.

Replit launched Missions on November 17, a mode where groups of AI agents work toward a business goal the user sets, within a spending cap. Replit's announcement gives examples like "win 10 paying enterprise customers" or replacing old internal software, rather than a single app to build.

Missions starts by asking questions to pin down what success means, such as whether a paid pilot counts as a customer. The user then sets a budget, $1,000 in Replit's example, and picks which actions need approval first: contacting prospects, sending proposals or making product promises. Several agents work in parallel, for instance one researching buyers while another builds a demo, and keep running in the cloud after the user closes the laptop. A separate AI judge checks the results against the agreed criteria. If the money runs out first, the mission stops unfinished.

Replit backs the launch with its own tests, which it calls controlled and single-run. It says Missions built a monitoring platform called DataCat in two days for $3,000, and that it passed 51 of 56 API compatibility checks against 45 for a Claude Code build from the same starting prompt. DataCat is now open source on GitHub. On NumPy speed tasks, Replit reports up to a 19x speedup with 32 agents, against 4x with one.

The pitch matters more than the benchmarks. Coding agents so far have been sold on writing code. Replit is selling outcomes, with a budget and a judge attached, which moves the question from "does the code work" to "did the agent hit the goal, and was it worth the spend." Replit hasn't said which plans include Missions or what it charges beyond the budget a user sets; companies are told to contact sales.

Sources

  1. Introducing Replit MissionsReplit
  2. replit/datacatGitHub

Add ClaudeGPT Si as a preferred source on Google