UiPath FUSION 2026 shipped three things automation buyers can switch on today — the Delegate desktop agent, Coding Agents and Cartographer — plus a governance layer, with no price for any of it. If you are on a UiPath renewal this quarter, the useful move is to pilot the pieces that are actually shipping, treat Decision Ledger and the autonomous testing story as roadmap, and get the unit Delegate consumes written into the order form before you sign.
The conference ran September 22–25 at Wynn Las Vegas. The product news landed on September 23 in a press release and a longer recap post on UiPath's own blog. Read side by side, they sort cleanly into what you can buy, what you can watch, and what nobody has told you yet.
What Is Actually Available From FUSION 2026?
Three agent products are available now, alongside a set of new governance controls; two headline ideas are not. UiPath's recap post lists Cartographer, UiPath for Coding Agents and Delegate as "generally available today," and describes Decision Ledger as "previewed today and on our roadmap."
| Announcement | Status per UiPath | What it does |
|---|---|---|
| Delegate | Available | A governed agent on an employee's desktop that does hand-work across apps |
| Coding Agents | Generally available | Lets Claude Code, Cursor, Codex and others build and run UiPath automations |
| Cartographer | Available | Drafts and maintains a "Map of Work" from documents, systems and people |
| Governance controls | Announced; no availability status given | Model Hub, Runtime Checker, LLM-as-Judge guardrail, compliance packs, identity policies |
| Automation Suite on Linux | Available | The full agentic stack on premises |
| Decision Ledger | Roadmap | Records agent decisions for continuous improvement |
| Test Cloud autonomous testing / "Dark Testing Factory" | Planned / building blocks | Agents that explore and test applications on their own |
The governance line is the one I expected to be vapor and, on paper, is not. The press release groups Model Hub — "central visibility into which models are in use, where they run, and how they're routed" — with a Runtime Checker that "validates agent behavior continuously against defined policies while an agent is running" and an LLM-as-Judge guardrail that "evaluates outputs against custom criteria." TechNode Global's report lists the same five controls. That is a real answer to the question a security team asks first.
The testing story is where the language softens. UiPath says of the Dark Testing Factory only that "we're showing its building blocks" on Test Cloud, and TechNode files autonomous test execution, Autonomous Exploration and AI-based verification under planned work. A demo on a keynote stage is not a line item you can put in a statement of work.
What Is UiPath Delegate, and Who Should Pilot It?
Delegate is a desktop agent: an employee describes a task in plain language, voice or a screen recording, and the agent does it across the apps they already use. A desktop agent, in this sense, is an AI agent that operates the user's own applications under that user's identity rather than running on a back-office robot. UiPath's product page says it works across desktop apps, browsers, terminals and mainframes on Windows and macOS, with "pre-authenticated access to 100+ connectors" through Integration Service plus MCP server support.
The pitch that matters to a Centre of Excellence lead is on that same page: Delegate "gives every employee a personal, on-tap agent, so your program reaches every desktop without adding developer capacity." That is the attended-automation promise from a decade ago, rebuilt on a language model. It also puts Delegate head to head with general-purpose computer-use agents — Claude Computer Use, Gemini's version, which we covered as a threat to the RPA market in June — that do not come with UiPath's audit trail.
Steel-man UiPath here, because the argument is good. Every action, the page says, "runs under enterprise identity, permissions, and audit on the UiPath Platform," with "autonomy modes, per-app approval, managed identity, permissions, and a full audit trail." A raw computer-use model gives you none of that out of the box; you build it. If your automation program already runs on UiPath, Delegate is the cheapest way to get a governed desktop agent into production, because the governance is the part you already paid for.
The weakness is the one we found when we compared Browserbase, Playwright and computer use for vendor-portal work: for a repeatable, stable task, a deterministic script still beats a model that re-reasons every click. Delegate's own answer is that it can save repeatable workflows "as reusable routines, then run them on demand, on a schedule, or from defined triggers," per the press release. Pilot it on the long tail of tasks nobody will ever script, not on the ones your CoE already automated.
What Do Coding Agents Change for Your Automation Team?
Coding Agents moves automation authoring from UiPath Studio to whatever coding agent your engineers already use. UiPath announced the integration on May 12 and the September release says it is "now generally available on the UiPath Platform," working with "UiPath Coding Agents, Claude Code, Cursor, Codex, Antigravity, or other supported agents" and fluent in UiPath folders, jobs, queues, assets, audit logs and connections.
Mechanically, UiPath's documentation describes three parts: the coding agent you prompt, a uip command-line tool that "authenticates to your organization, scaffolds projects and solutions, runs and publishes automations," and UiPath Skills — "packaged instructions that teach your coding agent how to do specific UiPath tasks."
That last sentence is the security review. A Claude Code, Cursor, Codex or Antigravity session holding an authenticated uip CLI can publish to your Orchestrator. We have written about what happens when agent allowlists are looser than they look — a "Python only" rule that meant any command, and permission prompts that stopped being asked. Scope the CLI's credentials to a development folder before anyone tries it.
Diginomica's Alyx MacQueen raised the sharper gap back in May: it is not clear whether governance covers "what the coding agent does before submission – the prompting, the reasoning trace, the iterations during code generation," as she wrote at launch. UiPath's audit log sees the automation that got published. Your coding agent's session is a separate record. If you are in a regulated industry, you need both.
What Will Any of This Cost?
Nobody outside a UiPath sales call knows, because neither the press release nor the recap post published a price or a licensing unit for Delegate, Coding Agents or Cartographer. The press release contains no pricing, and the Delegate page offers "Try it free" and "Request demo," not a rate card.
What UiPath does publish is how its agents are metered today, and that is the frame your quote will use. Under Unified Pricing, Agent Units, AI Units and Robot Units collapse into a single Platform Unit consumable. The agent licensing documentation charges 0.2 Platform Units per LLM call on standard models, 0.4 on premium models and 0.16 on basic ones, with calls counted "in 64k input token increments." Enterprise users also get a monthly AI and agentic usage pool for agent development and evaluations, assigned per named user and not poolable at tenant or organization level.
Now apply that to a desktop agent. Delegate re-reasons across screens, so one task is many LLM calls, and a premium model doubles the per-call rate. If Delegate draws on the same per-user pool — something UiPath has not said — your heavy users would burn out their allocation while light users strand theirs. None of that is a flaw in the model — it is simply the shape of the bill, and the same shape we mapped when we argued agentic AI pricing needs a cap before you buy consumption.
UiPath has reasons to want that consumption. Its second-quarter fiscal 2027 results, released September 3, showed ARR of $1.938 billion, up 12% year over year, net new ARR of $37 million and dollar-based net retention of 109%. Full-year ARR guidance is $2.065–2.070 billion. A vendor growing ARR at 12% needs expansion inside its installed base, and a per-employee desktop agent is the widest expansion surface it has.
Should Public-Sector and On-Prem Buyers Wait?
No — the on-premises path is real, but check which of the September products it includes before you plan around it. The September release says Automation Suite is "now available on Linux with the full UiPath Platform agentic AI stack." In May, UiPath's public-sector announcement said agencies could run agentic workflows "using UiPath Maestro™, Agent Builder, GenAI Activities, and context grounding" on open-source models run "entirely within their own data centers," and cited ISO/IEC 42001, FedRAMP and AIUC-1.
Delegate, Coding Agents and Cartographer are not named in that May list. "Full agentic stack" is a marketing phrase; the component list is the contract. Ask for it by product name.
What to Do Before Your UiPath Renewal
This Week:
- Pull your current order form and find which plan you are on — Flex (Agent Units) or Unified Pricing (Platform Units). The consumption units differ, and so will your Delegate quote.
- Ask your UiPath account team, in writing, for three answers: which unit Delegate consumes, whether it is licensed per user or per run, and whether Cartographer is included in your current subscription.
- Scope a development-folder credential for the
uipCLI before any engineer connects a coding agent to your tenant.
This Month:
- Run a two-week Delegate pilot with ten users on the long-tail tasks your CoE declined to automate. Log LLM calls per task from the agent licensing view, and multiply by the premium and standard rates to get a real per-task cost.
- Turn on Model Hub and the LLM-as-Judge guardrail in a non-production tenant and see what they actually catch. UiPath gave them no availability status — confirm they are in your tenant, then evaluate them rather than accepting a slide.
- Have one automation developer rebuild an existing Studio workflow through Coding Agents and compare time-to-publish and the audit record it leaves.
Before Renewal:
- Put a consumption cap and an overage rate into the order form. An uncapped desktop agent across every employee is an open line item.
- Get Decision Ledger and Test Cloud's autonomous testing written as roadmap, with no fee attached. Do not pay a platform uplift for features UiPath itself describes as roadmap and building blocks.
- If you run on premises, get Delegate and Coding Agents named in the Automation Suite entitlement, or plan on cloud.
The Bottom Line
FUSION 2026 was a stronger shipping conference than most, and a weaker pricing one. UiPath put working agents and working governance on the table, with the roadmap items labelled as roadmap — the same pattern as Dreamforce, which shipped agent operations while its headline model stayed in pilot. Delegate is attended automation again, rebuilt on a language model, and whether it is billed per seat, per run or per LLM call is the one thing the launch did not say. That answer decides whether "every desktop" is a productivity win or a budget line nobody can forecast.
Pilot what ships. Price what you pilot. Sign nothing you cannot meter.
Continue Reading
- Gemini 3.5 Flash Computer Use Threatens the $35B RPA Market
- Agentic AI Pricing: Don't Buy Consumption Without a Cap
- Browserbase vs Playwright vs Computer Use: The Script Still Wins
- True Cost of a Copilot Seat: The Licence Is the Floor
- Agent Orchestration Platforms: Score Exit, Not Features
- Dreamforce 2026 Shipped Agent Operations, Not the Headline Model
