What happened
Replit announced Missions on November 17, 2026. The company describes it as a way for teams to delegate complex outcomes—such as finding enterprise customers, improving product performance or replacing legacy software—without building and maintaining the infrastructure for coordinating an agent swarm. A team defines what it wants; Missions asks clarifying questions during setup, then uses the answers to shape how it prioritizes work and measures results. For a goal of winning ten enterprise customers, for example, the team must decide whether paid pilots count or whether success requires signed annual contracts above a minimum value.
The agreement for each mission sets a spending limit, actions requiring permission and acceptable tradeoffs. Replit’s example permits a $1,000 cap while requiring approval before agents contact prospects, share a proposal, change customer-facing sales materials or make product commitments. Agents can work in parallel—for example, researching accounts while building tailored demos—and use responses from approved outreach to adjust subsequent work. Teams choose whether to receive blocker, progress and status updates immediately, daily or weekly.
Replit says an independent AI judge checks work against the agreed success criteria and returns feedback for agents to use in refinement. A mission is marked complete when those criteria are met; if the budget runs out first, work stops unresolved. Users can inspect completed work, active tasks and next steps, including items awaiting approval.
To demonstrate Missions, Replit set it the goal of building DataCat, an application-monitoring platform intended to run on Replit’s own infrastructure. The company says the swarm built dashboards, alerts, application tracing, an agent for collecting server monitoring data and tracing libraries in two days within $3,000. In a deliberately bugged database test, DataCat alerted in 38 seconds against a 60-second target and traced the failure to the query and its logs. A three-hour load test stored 26 million activity records and 4.4 million log entries with no data lost; all 36 data-correctness checks and 15 alert-reliability checks passed.
Replit also reports comparisons with an existing monitoring platform and a separate build made from the same initial prompt using Claude Code’s /goal. In matched Node.js tests, DataCat handled 39% more requests with 51% less monitoring-related processor overhead. Against the Claude Code build, DataCat accepted all tested data types, passed 51 of 56 API-compatibility checks versus 45 of 56, and received traces at 49,000 per second on Postgres versus 9,800. Replit says it is open-sourcing a self-hostable version of DataCat. Missions can be started from Replit’s sidebar by selecting “New” and then “Mission”; the announcement gives no general pricing for Missions.
What the archive adds
New Runtime’s earlier coverage reported that Replit Cloud Agents for company codebases became available to Pro and Enterprise customers (November 2, 2026 note). Missions is a new event: it shifts the emphasis from agents working in company codebases to coordinating a goal-driven swarm under explicit limits and completion criteria. The earlier coverage establishes Replit’s ongoing expansion of agent work for business teams, but does not establish that the products share the same architecture.
Trend signal
Missions supports the pattern that agent systems are being organized around verified completed work rather than model calls: the mission’s success criteria determine when it is done, and the budget can stop it before completion. See Agent economics moves to completed work. Replit’s account is a vendor launch and demonstration, so it shows a product design choice, not evidence that this approach reliably reduces costs across customer deployments.
Why now
Replit says Missions addresses complex work that teams lack the time or capacity to take on, and presents DataCat as an internal effort to reduce monitoring costs and gain control. The announcement does not explain a broader market trigger or why the product is launching on this date.
Connections to other events
Replit’s earlier coverage of Cloud Agents for existing company codebases described wider access for Pro and Enterprise customers. Missions extends the company’s agent story from working with existing codebases toward coordinating multi-step outcomes. The two announcements concern distinct product capabilities; Replit does not state that Cloud Agents are a prerequisite for Missions.
Why it matters
For teams, the product’s consequential design is not simply that many agents can work in parallel. Missions puts a budget, permission boundaries and a definition of success around delegated work, while leaving people responsible for setting the goal and approving actions. That creates a more legible unit for evaluating autonomy: the team can ask whether a bounded mission met its agreed outcome before its budget expired.
The mechanism could make ambitious projects approachable for small teams, but the DataCat results should not be treated as a forecast of customer outcomes. They are Replit’s own tests, and the announcement says the comparisons do not establish performance against the original platform’s hosted service or on Replit’s production systems. Builders considering this model should examine the quality of the success criteria and approval design as closely as the agents’ output; a poorly scoped mission can still spend its budget without resolving the goal.
Limits and uncertainty
The announcement does not disclose Missions pricing, customer availability tiers or usage limits. Replit’s DataCat measurements are company-reported controlled tests, not independently verified results; the company explicitly says they do not establish performance against the original monitoring platform’s hosted service or on Replit production systems. Its comparison with the Claude Code build used one run per setup on identical cloud machines and scored shared product areas only. The announcement also does not report customer deployments, success rates across missions, or how often the judge’s completion decisions are overturned by human review.
Skill to build now
Before delegating a consequential task, write a short mission contract: define observable completion criteria, set a spending ceiling, list actions requiring human approval and state acceptable tradeoffs. Replit’s examples make the boundaries concrete—approval before external outreach or customer-facing commitments, for instance. Use this checklist on one bounded internal task this week; if the team cannot agree on what counts as done, the mission is not ready to run autonomously.
Reader takeaway
Missions packages agent swarms as bounded work: a stated outcome, spending cap, approval gates and a judge tied to agreed criteria. Replit’s DataCat demonstration is compelling but self-reported. The question for builders is whether these controls produce repeatable, inspectable results beyond a polished launch example.
Open question
How often do real Missions meet their agreed success criteria before exhausting their budgets, and how will Replit report those outcomes?