Replit launches Missions to coordinate AI agents on complex business goals

Replit has launched Missions, a service that coordinates swarms of agents to pursue complex business goals, from winning enterprise customers to replacing legacy software. Users define the outcome, budget and boundaries, while the agents plan, build, test and iterate toward it.
- Replit Missions coordinates agent swarms around user-defined goals.
- Users control budgets, boundaries and approval requirements.
- Replit used Missions to build the DataCat monitoring platform in two days.
- Replit says DataCat passed 51 of 51 data-correctness and alert-reliability checks.
- A self-hostable version of DataCat is being open-sourced.
Users set the goal, budget and permissions
Missions is designed for work that teams do not have the time or capacity to handle. A user could define a mission such as winning the first 10 paying enterprise customers without deciding in advance which industries to target, which buyers to contact or which customer problem to prioritize. During setup, Missions asks how success should be measured, including whether paid pilots count or whether the goal requires signed annual contracts above a minimum value.
The service also uses a Missions agreement to establish a spending limit and define actions that require approval. Replit gives the example of a $1,000 cap that requires permission before agents contact prospects, share a proposal, change customer-facing sales materials or make product commitments. Those restrictions remain in place as the mission evolves.
Agent swarms work in parallel and adapt
Agents can research accounts, build tailored demonstrations and prepare outreach for review. They can work in parallel, then use responses and shared findings to guide subsequent work. For example, a mission might focus on an industry showing stronger interest, revise a demonstration around a recurring buyer question or change messaging that receives few responses.
Replit says agent swarms can also help with optimization by exploring several implementations at once. In 13 NumPy benchmark tasks, a single Auto (Max) agent achieved speedups of up to 4×, compared with 17× for a 16-agent swarm and 19× for a 32-agent swarm. With Auto (Power), the corresponding figures were 2×, 7× and 9%. Replit says the tests showed how parallel exploration could help find faster implementations, while also validating correctness.
Users can choose whether to receive blockers, progress updates and general status immediately, in a daily digest or weekly. An independent AI judge checks results against the agreed success criteria, and the mission is marked complete when those criteria are met. If the budget runs out first, the work stops unresolved.
Replit used Missions to build DataCat
Replit says it used Missions to build DataCat, an application monitoring platform intended to run on and be customized for its own infrastructure. In two days and within a $3,000 budget, the agents built dashboards, alerts, application tracing, an agent for collecting monitoring data and tracing libraries. Replit says DataCat raised an alert for a deliberately introduced database bug in 38 seconds, beating a 60-second target, and traced the failure to the exact database query and its logs.
During a three-hour load test, DataCat stored 26 million application activity records and 4.4 million log entries with no data lost. Replit says all 36 data-correctness checks and 15 alert-reliability checks passed. The company also reports 51 of 51 checks passed in a broader summary, 39% more requests handled with 51% less monitoring-related processor overhead in matched Node.js tests, and 49,000 traces per second with DataCat using Postgres, compared with 9,800 for a Claude Code build.
Replit cautions that the controlled tests do not establish performance against the original platform’s hosted service or on Replit’s production systems. The company says it is open-sourcing a self-hostable version of DataCat, giving users a concrete example of the kind of work Missions is intended to support.
Replit: Introducing Replit Missions


