UiPath's stock swings and earnings calls have made headlines all year as the company repositions itself from "RPA vendor" to "agentic automation platform." Less covered is what that actually means at the product level — the specific features UiPath has shipped, and how they work. Here's a look at four of them: UiPath for Coding Agents, Autopilot for Agents, the split of Maestro into two products, and two newer additions from FUSION 2026.

UiPath for Coding Agents: letting Claude Code and OpenAI Codex ship to production

Announced on May 12, 2026 and reaching general availability at UiPath's FUSION 2026 conference, UiPath for Coding Agents is a platform integration aimed at a specific gap: AI coding agents like Claude Code and OpenAI Codex can write code quickly, but a business still needs governance, testing, and deployment infrastructure around whatever that code does — especially when it's automating a real business process.

The integration is built on three stated principles:

  • Open architecture. UiPath isn't requiring one specific coding agent. Claude Code and OpenAI Codex are supported at launch, with UiPath saying more integrations are planned.
  • A shared orchestration layer. Governance, observability, and execution controls sit underneath whichever coding agent or model version is doing the actual code generation, so switching agents doesn't mean rebuilding the guardrails.
  • Built-in governance. Policy enforcement, audit trails, credential vaults, and role-based access control apply to automations built this way, the same as automations built through UiPath's existing low-code tools.

In practice, the pitch is that a developer (or a less technical "citizen developer") describes what they want in natural language, a coding agent produces the implementation, and UiPath's platform carries it through testing, deployment, and ongoing operation — evaluations gate whether a given build is allowed to deploy. UiPath CEO Daniel Dines summarized the intent bluntly: "Now, anyone can describe what they want, direct a coding agent to produce it, and carry through every stage to production."

By FUSION 2026, UiPath for Coding Agents had also been tied into a second new feature, UiPath Cartographer (covered in more detail in RPA's Agentic Makeover: HCLTech's Robotiq.ai Deal and UiPath's Cartographer) — Cartographer's "Map of Work" (a governed record of how a business process actually runs) is meant to inform what a coding agent builds, rather than the agent working from a developer's guess at the process.

Autopilot for Agents: building an agent by describing it

UiPath's generative-AI assistant, Autopilot, first shipped in July 2024. Its newest form is Autopilot for Agents, a conversational assistant built directly into Agent Builder in Studio Web — currently in Preview.

Rather than manually wiring up an agent's prompts, tools, and data sources one by one, a builder works from a persistent chat panel alongside the agent they're editing. According to UiPath's own documentation, the panel includes:

  • A chat input for prompts and questions.
  • Dynamic prompt suggestions that update based on the agent's current state.
  • A model selector, so a builder can choose which LLM handles a given session.
  • File upload support (documents, images, code files, and data files, up to 15 MB each) for grounding a request in real material.

When a builder describes a change, Autopilot proposes specific edits — to the agent's prompts, its tool definitions, its schema, or its context sources — which the builder reviews before anything is applied. That review step matters: it keeps a natural-language request from silently rewriting an agent's behavior without a human seeing the diff first.

Splitting Maestro: Orchestrate vs. Automate

UiPath Maestro launched in 2025 as the orchestration layer sitting above UiPath's agents and RPA robots — the piece responsible for coordinating a multi-step process rather than running any single step. At FUSION 2026, UiPath split Maestro into two distinct products aimed at different kinds of work:

  • Maestro Orchestrate handles long-running, cross-team processes — the kind that can span days or weeks and need to persist state across that whole span (an insurance claim moving through multiple departments, for instance).
  • Maestro Automate is built for high-volume, fast-moving tasks on the more familiar builder canvases UiPath customers already use for conventional automations.

The split is UiPath's answer to a real tension in agentic-automation platforms generally: a system tuned for a process that unfolds over weeks has different state-management and observability needs than one built to process thousands of short transactions per hour, and trying to serve both with one product tends to compromise on both.

Two more from FUSION 2026: Delegate and Dark Testing Factory

Two additional FUSION 2026 announcements are worth knowing about even though they're each narrower in scope:

  • UiPath Delegate, now generally available, is described as a governed agent that runs on a user's own desktop, taking on tasks a person would otherwise do by hand. It accepts instructions either as natural language or by having a person demonstrate the task once, and it operates under that person's existing identity, permissions, and audit controls — it doesn't get a separate, broader set of access.
  • Dark Testing Factory, still in Preview, flips the usual test-automation script: instead of automating individual test cases, it automates the testing system itself, built on UiPath's existing Test Cloud. People set the goals and boundaries; agents execute the testing within them, with what UiPath calls "human-accountable governance" — a person is still the one accountable for what the agents were told to do.

The pattern underneath all of this

Taken together, these releases point at the same underlying bet: UiPath is trying to make itself the governance and orchestration layer that sits underneath AI-generated and AI-driven work, regardless of which AI actually produced it — a coding agent, a conversational agent-builder, or an autonomous test-runner. Whether that bet pays off commercially is a separate, ongoing story (and one this site has covered from the earnings and investor-day side). But at the product level, the direction is consistent: less "UiPath's own AI does the work," more "UiPath governs the work, whoever's AI is doing it."

Sources

AI Coding AssistantsAgentic AIEnterprise AutomationRPAUiPath