Replit Agent
A cloud IDE where an AI agent builds, tests, and deploys a full-stack app in the same workspace it will run in, billed per checkpoint instead of a flat seat.
Who it's for
Non-developers and solo founders who want the whole stack, editor, database, hosting, and deploy, running in one persistent cloud workspace instead of stitched together from separate tools.
Standout details
- Agent 3 can run unsupervised for up to 200 minutes, testing its own work in a real browser against buttons, forms, APIs, and login flows before handing control back
- Every plan runs on effort-based pricing: simple edits land under the old $0.25 flat checkpoint fee, complex ones can run to several dollars, and the cost is not shown until after the checkpoint is debited
- Core dropped from $25 to $20 a month in February 2026 when Replit retired its $40-a-seat Teams plan in favor of Pro, a $100 flat rate for up to 15 builders on pooled credits
- Database rollback reaches back 28 days on Pro versus 7 on Core, and Replit now separates development and production databases by default after a 2025 incident where Agent deleted a live production database
- Agent can now build and wire up its own automations, Slack bots, Telegram bots, email flows, on request, not just the app itself
What it is
Replit Agent is the AI builder inside Replit’s browser-based cloud IDE. You describe an app in chat, and the agent writes it, provisions a Postgres database, wires up auth, and deploys it, all inside the same workspace where the app then runs. Agent 3, the current version, adds a browser-based testing loop: it periodically drives the running app itself, checking buttons, forms, and API responses, and can log into a Replit Auth flow to verify a login actually works before it calls a task done.
Replit says Agent 3 can run for up to 200 minutes with minimal check-ins, a Max Autonomy beta pushes that further, and the agent can now build subsidiary automations, a Slack bot, a scheduled email job, on top of the main app.
Who should use it
Non-developers who want a working app, database, and live URL without assembling separate services, and solo founders who would rather stay in one workspace than juggle an editor, a host, and a database console. It sits next to Bolt.new and v0 in the same category; the practical difference is that Replit’s workspace is persistent and includes the database and hosting natively, where Bolt spins up a browser sandbox and v0 hands off to Vercel.
Developers can use it too, for the same reason non-developers do: scaffolding CRUD, auth, and a deploy pipeline by prompt is faster than assembling it by hand, and Replit’s multiplayer workspace supports up to 15 collaborators on Pro if more than one person needs to be in the project.
What stands out
The pricing model changed twice in under a year. Agent operations used to bill as a flat $0.25 “checkpoint” regardless of how much work the task took. Replit moved to effort-based pricing, where a small edit can cost less than that and a complex feature can cost several dollars, bundled into a single checkpoint rather than split across many. Then in February 2026 Replit cut Core from $25 to $20 a month and retired the $40-per-seat Teams plan entirely, replacing it with Pro at a flat $100 for up to 15 builders on a shared credit pool.
The database story is the other half of the pitch. Every app gets a built-in Postgres database with point-in-time rollback, 7 days on Core and 28 on Pro, and Replit now separates development and production databases by default. That default did not exist in July 2025, when Agent deleted a live production database during a testing session reported publicly by SaaStr’s Jason Lemkin; Replit’s CEO confirmed the incident and shipped the dev/prod split, one-click restore, and a chat-only planning mode in response.
The verdict
For someone who wants an app, a database, and a live URL without touching separate tools, Replit Agent is genuinely the most complete package in this category. The workspace persists, the agent tests its own output in a real browser, and the database and deploy target are already there.
What you are trading for that convenience is bill predictability. Effort-based pricing means you find out what a checkpoint cost after it runs, not before, and multiple 2026 reports of active builders describe monthly spend running well past the plan’s included credit once a project moves past scaffolding into real debugging. Set a spending cap before your first serious session. If a variable AI bill is not something you want to think about, or you are happy with the database and host you already use, this is not the tool to switch for.
Honest verdict
Replit Agent is the closest thing in this category to a finished product: the editor, the database, the deploy target, and the agent all live in one place, so there is nothing to stitch together. The effort-based pricing is the real cost of that convenience. You commit to a checkpoint before you know what it costs, and 2026 coverage of user bills describes sessions running well past the plan's included credit once real features and debugging are involved. Buy it if you want one workspace that does everything and you are comfortable setting a spending cap before you start. Skip it if an unpredictable bill would actually stress you out, or if you already have a database and host you are happy with and just want an editor.
Published 1 Sept 2026