Replit Missions explained: what it changes for small teams
What Replit launched on November 17, 2026
Replit launched Replit Missions on November 17, 2026: you describe a goal, and a swarm of agents plans, builds, tests and iterates toward it within a budget and boundaries you set. Compared with Replit Agent 4 and Claude Code, Missions is the only option we reviewed that writes a per-goal budget, a list of approval points and an independent judge into the work, but Replit has not published Missions pricing, so act this month only if you can get a quote and name an approver.
In plain terms, a mission is a goal plus a Mission Contract. The contract sets the spending limit, the actions that need your permission, and the success criteria that an independent judge checks. Replit's launch post frames the target work as complex projects teams have not had capacity for, such as winning more enterprise customers, improving product performance or replacing legacy software (Replit launch post).
The example in the post is "win our first 10 paying enterprise customers." Setup asks what counts as an enterprise customer and whether paid pilots count toward the ten, and those answers decide which accounts the mission targets. Missions example spending cap: $1,000 per mission according to Replit (November 2026 launch).
The previous version is Replit Agent 4, introduced on March 11, 2026 (Agent 4 launch post). Agent 4 splits a request into background tasks that run at the same time, and approved tasks merge into the project. We found no page saying Agent 4 has been renamed, merged into Missions or discontinued, and the Replit pages we opened do not explain how the two relate.
For a small or mid-size service business, the change is in the kind of work you can hand off. Agent 4 is built around building and changing an app and its supporting artifacts in one project. Missions is also aimed at go-to-market work, such as researching target accounts, building demos and preparing outreach, with approval before anything reaches a prospect. Once an agent is drafting messages for prospects, the approval rules become the control you have to own.
Key Takeaways
TL;DR: Missions is Replit's goal-based agent service, launched November 17, 2026. It is worth a pilot only after you get a quote, because the vendor has not published its price.
Missions adds a per-goal budget, approval points and an independent AI judge. The Agent 4 pages we opened describe parallel tasks and billing alerts, not per-goal budgets.
Replit's DataCat results report a build in two days within a $3,000 budget. These are vendor-run controlled tests, not independent results.
Gartner forecasts, according to W.Media, that more than 40% of agentic AI projects will be canceled by the end of 2027, citing cost, unclear value and weak risk controls.
For a small B2B company, the useful question is which outcome you can hand off with a cap, an approver and a success test, not which agent is best.
Keep outbound contact, proposals and product commitments behind a named human approver, whichever platform you choose.
How we evaluated these tools
This is an informed buyer's guide built from public vendor pages and independent coverage. It is not hands-on testing. We weighted seven criteria by what a small B2B buyer has to decide before a pilot. The weights are our judgment, and the evidence column shows which pages support each row. We do not add the rows into one total score, because several cells are unpublished, and an unpublished cell should not be counted as a zero.
| Criterion | Weight | Score scale | Evidence used |
|---|---|---|---|
| Goal scope fit (software, sales or operations) | 20% | 0-5 per option | Launch post and Agent 4 post |
| Budget control (per-goal cap, stop rule) | 20% | 0-5 per option | Launch post, billing docs |
| Approval gates for outbound and commitments | 15% | 0-5 per option | Launch post, Agent 4 post |
| Verification (independent check of results) | 15% | 0-5 per option | Launch post, pages opened for Claude Code |
| Published price and access path | 15% | 0-5 per option | Replit and Anthropic pricing pages |
| Parallel capacity | 10% | 0-5 per option | Plan docs, launch post |
| Data control (self-hosting, tenancy) | 10% | 0-5 per option | Launch post, pricing page |
The three options are Missions, Agent 4 (the previous Replit version) and Claude Code. We picked Claude Code as the rival because Replit's own build test compared DataCat against a build made from the same prompt in Claude Code, so buyers will see it in the launch material. Market context comes from the 2025 Stack Overflow developer survey and from Gartner's forecast, as reported by W.Media.
Feature matrix: Missions, Agent 4 and Claude Code
| Feature | Replit Missions | Replit Agent 4 | Claude Code |
|---|---|---|---|
| Launch date | November 17, 2026 (launch post) | March 11, 2026 (Agent 4 post) | Not stated on the pages we opened |
| Unit of work | A goal plus a Mission Contract | A request split into background tasks | Coding help inside a paid plan; no goal contract described |
| Spending control | Cap per mission, $1,000 in the example | Billing alerts and budgets in settings; no per-goal cap described | Usage limits every five hours and weekly (pricing); no dollar cap stated |
| Approval points | Set in the contract: prospect contact, proposals, sales materials, product commitments | Approve each task before it merges; confirmation before paid work | Asks permission before file edits and commands; no business-level approval list for outbound actions |
| Result checking | Independent AI judge against your success criteria | Review results before merge | Not described on the pages we opened |
| Parallel capacity | Agent swarms; the vendor has not published a task limit | 1 task on Core, up to 10 on Pro | Not described on the pricing page |
| Self-hosting | Not stated for Missions; only DataCat is open-sourced as self-hostable | Not stated | Not stated on the pricing page |
| Published price | Quote-based; organizations contact sales | Not priced separately on the pages we opened | Pro from $17 per month, billed annually |
Read the matrix as what each vendor states, not as a ranking. A blank-looking cell means the page we opened was silent, which is a gap in the public record and not proof that a feature is missing. The judge's underlying model is not named, and the launch post does not describe how the judge is evaluated, so buyers should ask for that in writing before a pilot.
Vendor profiles
Replit Missions
Best fit. Teams with one goal that has a measurable finish line, a budget owner, and a named person who approves any contact with customers. The launch example fits that shape: research, demo building and outreach drafts, with approval before anything reaches a prospect (Replit launch post).
Limitations. The vendor has not published Missions pricing, general availability dates or regions. Work stops when the budget runs out, "with the outcome still unresolved." The evidence is Replit's own. In the DataCat test, a deliberately introduced database bug raised an alert in 38 seconds against a 60-second goal, and the NumPy benchmark reached 4× with one agent and 19× with 32 agents on Auto (Max). The post says the controlled tests "do not establish performance against the original platform's hosted service or on Replit's production systems."
Implementation. Click New in the sidebar, choose Mission and describe the goal. Answer the setup questions that define success, then write the Mission Contract with the cap, the approval list and any acceptable tradeoffs, such as paid pilots or annual-contract discounts. Choose the update cadence: immediately, daily digest or weekly. Prerequisites are a written success definition, a budget owner, and an approver for each type of external action. Organizations contact sales to bring Missions in.
Evidence. Replit launch post.
Replit Agent 4
Best fit. Teams already building in Replit that want parallel build work inside one project, including UI variants on a design canvas. The Agent 4 post says parallel task execution is available to Pro and Enterprise users, with Core users included temporarily (Agent 4 launch post).
Limitations. Parallel work is capped by plan, with one task on Core and up to ten on Pro in Replit's plan documentation. The Agent 4 post does not state caps on agents or sub-agents, and it presents "10x faster" as a goal rather than a measured result. The pages we opened do not describe an independent judge.
Implementation. Work inside the Replit project, review background tasks and approve each one before it merges. Replit's billing docs say it asks for confirmation before a paid action starts, and that you can set alerts and budgets on the Billing page in Settings (Replit AI billing docs).
Evidence. Agent 4 launch post and Replit plan docs.
Claude Code
Best fit. Engineering teams that want coding help inside a paid Anthropic plan. The pricing page marks Claude Code as included in Pro, Max and Team, and as not included in Free (Anthropic pricing).
Limitations. Claude Code asks permission before file edits and commands, but it has no per-goal dollar budget, no business-level approval list for outbound actions (prospect contact, proposals) and no independent judge. Usage limits apply every five hours and every week on the plans the page lists, and the page says how fast you reach a limit "depends on the model you use, how long your chats are, and how large your files are." The Replit comparison in the launch post was run by Replit, so treat it as the vendor's result.
Implementation. Buy a paid seat. Team standard seats are priced per seat, and the Team plan is listed for teams of 2 to 150. Your engineers still own code review, and nothing on the pricing page replaces that step.
Evidence. Anthropic pricing.
What it costs, and what is not published
Replit's pricing page prints the Pro plan at $90 per month billed annually, beside a $100 figure, and the Core plan at $18 per month billed annually, beside $20. Replit's Pro plan: $90 per month, billed annually according to Replit (checked November 17, 2026). Replit's plan help page lists a different annual figure for Core, so confirm the Core price at checkout before you budget. The pricing page lists Enterprise as custom, with single-tenant and private networking options. Missions does not appear on the pricing page at all.
| Vendor | Plan | Published price | Billing | Source |
|---|---|---|---|---|
| Replit | Core | $18 per month (page also shows $20) | Billed annually | Replit pricing |
| Replit | Pro | $90 per month (page also shows $100) | Billed annually | Replit pricing |
| Replit | Enterprise | Quote-based | Custom | Replit pricing |
| Replit | Missions | Quote-based | Not published | Launch post |
| Anthropic | Pro | $17 per month ($20 if billed monthly) | Billed annually, $200 up front | Anthropic pricing |
| Anthropic | Max | $100 per month | Monthly | Anthropic pricing |
| Anthropic | Team standard seat | $20 per seat per month ($25 if billed monthly) | Teams of 2 to 150 | Anthropic pricing |
Pricing checked November 17, 2026.
For Anthropic, Anthropic's Pro plan: $17 per month, billed annually according to Anthropic (checked November 17, 2026). The same page lists the Team plan's pooled API credits at up to $500 per month, which is a usage allowance and not a seat price.
Worked example: a five-person team's spending ceiling
Take a five-person sales-operations team that assumes it buys five Replit Pro plans at $90 per month each. That is $450 per month, or $5,400 over a year when billed annually. The pricing page says Pro allows up to 15 collaborators, but it does not say whether the Pro price is per seat, so confirm that before you budget. If the team also ran one Missions goal a month at the launch example's $1,000 cap, the spending ceiling would add $12,000 over the year, for a combined ceiling of $17,400 ($5,400 plus $12,000). A cap is a maximum, not an expected spend, and Missions pricing is unpublished, so treat $17,400 as an upper bound built from two published numbers, not a quote. If the same team also wanted a HubSpot contact.creation event to trigger an approved follow-up draft, that trigger adds no figure to this ceiling, so the 5 plans, 12 goals and $17,400 bound stay unchanged.
| Line | Basis | Annual ceiling |
|---|---|---|
| 5 Pro plans, billed annually | 5 × $90 × 12 months | $5,400 |
| 12 Missions goals at the example cap | 12 × $1,000 | $12,000 |
| Combined upper bound (2 lines) | $5,400 + $12,000 | $17,400 |
Where US Tech Automations fits next to a goal-based tool
A proposed US Tech Automations design, which is configurable and not a live deployment, could start with a HubSpot contact creation event as the trigger. The action reads the contact's fields and drafts a follow-up email and a call note. The output is a logged note on the contact and a task for the owner, and nothing is sent until a named person approves the draft. Two prerequisites matter. According to HubSpot's webhook docs, subscriptions are written to a journal that the app polls rather than a target URL, so the design needs a polling schedule. Creating subscriptions requires an OAuth token with the webhook write scope, plus the object scope crm.contacts.read to read contacts. The human review points are the approver's check on each draft, and a check for an existing note before any send, so a retry does not produce a second email.
A second proposed design covers a weekly pipeline check. The trigger is a scheduled weekly run. The action pulls open deals with no logged activity inside a window you set, and the output is a summary posted to a Slack channel with each deal's owner named. Prerequisites are read access to deals in your CRM, a Slack channel with bot posting permission, and a deal-owner mapping you keep current. The owner decides whether to act, and the design never changes a deal stage on its own. Our guide on connecting monday.com to Slack covers the Slack side of that pattern.
Zapier, Make and n8n can keep run histories, retries and error branches when you configure them, so a no-code build is a fair alternative. In those tools, or in in-house code, the buyer still designs and owns observability, idempotency so a retry does not double-send, escalation when an approval sits too long, access controls for each connection, and maintenance when the CRM's API changes. A proposed US Tech Automations design would configure those as named steps: a run history per contact, a duplicate check before each send, an escalation to the deal owner after a window you choose, and a separate scoped connection for each data source. You still have to approve and test each of those choices.
Who should choose which option
| Vendor | Option | Choose it if | Skip it if |
|---|---|---|---|
| Replit | Missions | You have one measurable goal, a budget owner and a written quote | You need a fixed monthly cost before any quote arrives |
| Replit | Agent 4 | You already build in Replit and want parallel tasks in one project | You need outbound approvals or an independent judge for non-software goals |
| Anthropic | Claude Code | Your engineers want coding help inside a paid plan | You need per-goal budgets or approval gates from the tool itself |
| Your current tools | Plus a workflow layer | Your work is CRM follow-up and reporting across tools you already pay for | Your problem is building a new product |
Who this is for
This is for founders, operations leads and engineering leads at a small B2B software or services company who already run a CRM and several SaaS tools. They need one of three things: a decision on whether to pilot Missions this month, a plan that keeps outbound work approved, or a simpler follow-up workflow across tools they already pay for. Teams in adjacent workflows often start with a narrower guide, such as automating vendor price comparison for restaurant purchasing. If your team tracks pilots in a work-management tool, our monday.com review is a related read.
Red flags: you cannot state the success metric in one sentence; nobody is named to approve customer-facing actions; or the budget owner will not set a cap before the first run.
Glossary
Mission: a goal you hand to Replit, with a contract that sets the budget, the approvals and the success criteria.
Mission Contract: the agreement that holds the spending limit, the actions that need permission and acceptable tradeoffs.
Independent AI judge: the check that compares results with your success criteria; a mission is complete when they are met.
Agent swarm: several agents working in parallel on parts of one goal and sharing findings.
Checkpoint: in Replit's Build Mode, the point where Agent completes the work on a request (Replit AI billing docs).
Effort-based pricing: Replit's model in which the cost of a request scales with its complexity (Replit AI billing docs).
Agent washing: Gartner's term for relabeling assistants, RPA tools or chatbots as agentic when they lack substantial agentic capability (W.Media).
Decision checklist
- Can you state the goal and its success test in one sentence that a judge could check?
- Do you have a written Missions quote, or a confirmed date for one?
- Is the spending cap set in writing, with a named budget owner?
- Is each outbound action (prospect contact, proposal, sales material, product commitment) assigned to a human approver?
- Have you confirmed Core or Pro pricing at checkout, since Replit's two pages disagree?
- Have you authorized only the data scopes the agent needs?
- Does a simpler tool you already pay for solve the problem?
Signal vs Speculation
Published by the vendors (signal):
Replit launched Missions on November 17, 2026, with a $1,000 example cap, configurable approvals, progress updates and an independent judge, according to the launch post.
Missions pricing is not published, and organizations are told to contact sales.
The DataCat and NumPy results are Replit's own controlled tests.
Replit's pricing page shows Pro at $90 per month billed annually, while the help docs give a different annual Core figure.
Independent context (signal, with limits):
Gartner forecasts more than 40% of agentic AI projects will be canceled by the end of 2027. That is a market forecast, not a measure of any one product.
According to Stack Overflow, 14.1% of respondents use AI agents daily at work, and 46% distrust the accuracy of AI tools.
Our inference (speculation, labeled):
We think Missions suits goals with a clear finish line, such as a revenue target or a feature build, better than open-ended operations. That is a reading of the launch, not a published fact.
We expect small teams to hit the quote step first, which makes "act this month" realistic only for teams that can start the sales conversation now.
We do not know how well the judge handles messy business goals, because the launch post gives no evaluation method.
Frequently asked questions
What is Replit Missions?
Replit Missions is a service launched November 17, 2026 that runs a goal you define with a swarm of agents, inside a budget and approval rules you set.
Is Replit Missions included in Replit's Core or Pro plans?
The vendor has not published whether Missions is included in any plan, and the launch post says organizations can contact sales about it.
Can a Missions agent contact prospects without my approval?
The launch example shows that you can require approval before the agent contacts prospects, shares a proposal, changes customer-facing sales materials or makes product commitments.
Are the DataCat results independent?
No. They are Replit's controlled tests, and the post says they do not establish performance against the original platform's hosted service or on Replit's production systems.
Should a small team switch from Replit Agent 4 or Claude Code this month?
Not on the published evidence alone. Pilot only after you have a quote, a written cap and a named approver, because Missions pricing is the main unknown.
When should you not use US Tech Automations?
A simpler tool wins when the job is a single step, such as a notification your CRM already sends natively. It also does not fit when you have no API or export access to the systems involved, because the workflow would have nothing to read. Building a new product is a third case, which Missions, Agent 4 or Claude Code address more directly.
Conclusion
Replit Missions is worth a serious look for one goal with a measurable finish line and a named approver. It is a poor fit for a small team that needs a fixed monthly price this quarter. Start with a quote, a written cap and one pilot goal, and keep every outbound action behind a human. If your bottleneck is follow-up across tools you already pay for, see how US Tech Automations configures this with approvals and run history built into the same design.
About the Author

Helping businesses leverage automation for operational efficiency.