Autopilot is now a coding agent in Studio Web
Public preview. Available now in Studio Web — nothing to install, nothing to update. Already generally available in Studio Desktop. We want to share it early because the preview is exactly when your feedback immediately shapes what ships.
TLDR: the coding-agent Autopilot is now in Studio Web. It plans, builds, runs, debugs and explains automations alongside you — across the project types you build in the browser: Maestro Flow, agents, API workflows, RPA workflows, and much more. Open Studio Web and tell it what you need. No new subscription, no extra accounts, nothing to install.
Why
We rebuilt Autopilot as a full coding agent on the same skills and CLI tooling we ship for third-party coding agents. It landed in Studio Desktop first. Studio Web is where a growing share of building happens — and where the newer automation components live: Maestro Flow and BPMN, agents, API workflows, to name a few. Those are the ones with the least muscle memory behind them, which makes them the ones where a coding agent that already knows UiPath conventions helps most.
Same agent, same skills, same governance. Different surface.
Used Autopilot in Studio Web before? Try it again. The earlier capability sets of Autopilot generated and assisted. The new Autopilot plans a job, does it, checks its own work, and iterates, just like a developer does.
How
Open Studio Web. That is the whole prerequisite.
There is no version to check and nothing to install — that is the difference from Studio Desktop, where you needed to be on a recent STS. Autopilot loads the UiPath skills and connects to your project, your tenant, and Orchestrator from the browser.
What it does across the project types you build in Web
Maestro Flow
Build the flow, wire nodes to published processes, add approval and script nodes, validate it, run it, then read the runs and incidents afterwards. Evaluation sets and evaluators too.
Maestro BPMN
Generate or edit the diagram, validate and debug it, keep it and the implementation in step
Maestro Case
Draft the case plan and its stages, then manage instances, processes and incidents
Agent
Scaffold the agent with its prompts, tools and memory · debug it running on Studio Web · build evaluation sets, evaluators and test cases · set guardrails
API Workflow
Build it from a description, look up and wire connector activities, validate, run it locally before deploying
RPA Workflow
Build and edit workflows that follow UiPath conventions · find the right activity · run validation and Workflow Analyzer · strip unused items · step through a debug session
Coded App
Scaffold the app, push and pull the source, pack, publish and deploy it
Function
Write the deterministic Python unit with its input/output contract, run it, publish it
These are not aspirations — they are what UiPath skills and the uip CLI expose, and Autopilot runs on exactly those. You can check any of them yourself.
And the things that are not tied to one project type:
-
Plan from a spec — hand it a PDD, an SDD or a plain requirement and watch it plan, then build
-
Publish & deploy — package a solution and put it in Orchestrator
-
Run & troubleshoot — start it, read the failure, name the cause
-
Explain — walk through a project you have never seen, step by step
-
Improve — restructure, add the missing error handling and logging
A few things you can say to it:
-
Build the eligibility check as an API workflow and call it from this flow
-
Replace the mocked node with the published process from Shared/Finance and map the fields from the loop
-
This flow run faulted — why?
-
Explain what this BPMN actually does, stage by stage
-
Turn this requirements doc into a solution I can run
Why Autopilot
-
Included — no separate subscription and no token bill, within a monthly usage allowance per user
-
No extra accounts — no third-party sign-up, no API keys; it is already pointed at your project and tenant
-
Governed — admins control which models are available, how autonomously it acts, which tools are enabled, and which MCP servers are allowed, centrally through Automation Ops. Every model request routes through the AI Trust Layer
-
Studio Web stays the visual layer — everything Autopilot does is something you can open, inspect, edit, and govern yourself
If your team prefers a different editor or a third-party coding agent, that works too — UiPath for Coding Agents opens the platform to them. This is for the people who build in Studio Web and want the least setup with the most control.
Try it
Open a project in Studio Web and give it something real — build a flow from a spec, point it at a run that faulted, or ask it to explain the project nobody on the team wants to touch.
Worked well? Send that too — knowing what to double down on before GA matters as much as knowing what broke.
Prefer to talk about it in public? Reply in this thread: your prompt, what you expected, and what happened instead. Either route reaches us.
