Autopilot Agent

Autopilot Agent, a coding agent and the default Autopilot experience in Studio Web for building, debugging, and evaluating agents.

Autopilot Agent is a coding agent built into Studio Web, and the default Autopilot experience for agents. It plans and builds agents alongside the rest of your solution, then runs and debugs them to check its own work. Autopilot Agent runs on the same UiPath skills and command-line interface (CLI) tooling as UiPath coding agents, so there is nothing to install.

What Autopilot Agent can do for agents

For agents, Autopilot Agent can:

• Scaffold an agent with its prompts, tools, and memory
• Debug the agent as it runs in Studio Web
• Build evaluation sets, evaluators, and test cases
• Set guardrails

Importante:

Autopilot Agent support for Agent projects in Studio Web is experimental. You need to review the generated prompts, tools, and guardrails before you run or publish the agent.

For details, refer to Autopilot Agent in Studio Web in the Autopilot guide.

Classic Autopilot

Classic Autopilot is a conversational assistant that works on your agent one step at a time, with you driving each step. By default, it proposes targeted edits to the agent definition, which you review before applying. The Switch to classic Autopilot button in the Autopilot panel header changes from Autopilot Agent to classic Autopilot. For details, refer to Classic Autopilot.

Autopilot Agent settings

The Autopilot Agent settings control what Autopilot Agent can use and how it responds. The settings are grouped into the following tabs:

Tabulación	Qué controla
Herramientas	The tools Autopilot Agent can use, grouped by type, such as file, local, web, and utility tools. Each tool can be turned on or off.
Aptitudes	Built-in UiPath skills and your own, which you create with New skill or add with Import. A skill turned off stays out of Autopilot Agent's context for the session.
Conocimiento	Information Autopilot Agent saves from your conversations, which you can review and manage.
Servidores MCP	Model Context Protocol (MCP) servers that expose more tools to Autopilot Agent. Local servers run on your machine, and remote servers are accessed over HTTP.
Personalización	Custom instructions added to every prompt, follow-up suggestions after each turn, and the communication style, tone, and proactivity of Autopilot Agent.
Seguridad	Websites Autopilot Agent uses without asking (Trusted) and websites it never accesses (Blocked), including your organization's website rules. Choosing Always trust during an approval adds a site to Trusted.

Licensing and governance

Autopilot is included in your user license, within a monthly usage allowance per user. Administrators govern Autopilot through Automation Ops, and every model request routes through the AI Trust Layer. For details, refer to Autopilot licensing.