Replit Missions lets you hand an objective to a group of AI agents and set a spending cap and approval rules before any work begins. The product launched on November 17 and is built for goals with a clear finish line. This guide walks through how to set one up, what to decide in advance and where the limits of the approach lie.
What Replit Missions Does
A Mission is a goal-driven run. You provide the objective, the success criteria and a spending limit. Agents then plan, build, test and iterate until an independent AI judge confirms the result against the criteria you agreed in advance, or until the budget runs out. If the money runs out first, the work stops, and the outcome may remain unresolved.
Replit's example goal is to win its first 10 paying enterprise customers. Goals like that are useful for showing how the system works, but they also reveal where ambiguity creeps in. The definition of a paying customer has to be written down before the run starts, or the agents will optimize for whatever the phrase happens to mean.
Step One: Write the Success Definition
Before you open a Mission, write down what success looks like in plain language. Decide what counts as an enterprise customer and whether paid pilots should count toward the target. These answers determine which accounts the agents pursue and how the judge scores the result.
A useful test is whether a human reviewer, given only the written criteria, would reach the same verdict as the judge. If they would not, the criteria need more work. Vague goals are the most common source of disappointing runs, and they are the hardest for an automated judge to assess.
Step Two: Set the Budget
Set a spending cap that reflects how much you are willing to lose if the run does not succeed. The launch example uses $1,000 per Mission, according to one setup guide. Treat that as an illustration, not a recommendation, and begin with a smaller cap on your first run so you can see how quickly the agents consume resources.
Name a budget owner as well. Someone must be accountable for approving the cap and deciding whether to extend it when a run is close to a result. Without that role, a budget tends to be exhausted by default rather than by decision.
Step Three: Define Approval Gates
Most of the risk in an agent run sits in external actions. Replit's setup lets you choose which actions need your approval before they happen. For a sales-focused goal, typical gates include contacting prospects, sending proposals, changing sales materials and making product commitments.
Assign a named approver to each action type. The launch example requires approval before anything reaches a prospect, which is a sensible default for any goal that touches customers. A single poorly worded message can damage a relationship that took months to build, and it is far cheaper to review a draft than to repair a reputation.
- Prospect contact: require approval for every outbound message.
- Proposals and pricing: require approval before any figure leaves the company.
- Sales materials: require review before customer-facing changes go live.
- Product commitments: never allow agents to promise features or dates without sign-off.
Step Four: Choose an Update Cadence
Decide how often you want to hear from the system. The options described in the setup guide are immediate updates, a daily digest or a weekly update. For a run with a limited budget, a daily digest often strikes the right balance, since it gives you enough visibility to intervene without requiring constant attention.
Caveats Before You Commit
Several points deserve caution. Pricing and general availability have not been fully published, and organizations are typically directed to contact sales for access. Replit's pricing page does not list Missions, so ask for a quote before you plan a pilot and confirm exactly what is included.
Ask, too, how the judge works. The underlying model used for judging has not been clearly described in the coverage available, and the reliability of the judge on vague goals is unproven. Request those details in writing before you rely on the verdicts, and check whether the judge's criteria can be reviewed by someone outside the project.
Why It Matters
Budget caps and approval gates are a practical way to bring agents into real work without losing control of costs or customer relationships. They turn an open-ended autonomy experiment into something a team can scope, review and stop. That pattern is likely to spread across agent platforms, because it answers the first questions any business will ask.
Summary Checklist
Before you launch a Mission, confirm that you have a written success definition, a budget owner, a spending cap, an approver for each external action type and a chosen update cadence. If any of those is missing, wait. A run that is well scoped is far more likely to produce a result you can use.
