UiPath Ships the Unglamorous Half of Agentic AI
UiPath used its FUSION 2026 conference in Las Vegas to ship the unglamorous half of the agentic-AI promise. The headline items were not smarter models but plumbing: governance controls, governed data access, a Linux port and an autonomous testing roadmap. For a company fighting platform giants, that choice is the story.
Announced on September 23, the releases cover two tracks. The platform release brought UiPath Coding Agents to general availability. It introduced the Delegate desktop agent and added governance controls for enterprise AI. The Test Cloud release pushed testing toward autonomy, with staged rollouts beginning in October 2026.
What Actually Shipped
Three agent products are available now. Coding Agents, first announced in May 2026, let teams build and run UiPath automations from the coding tools they already use. Supported tools include Claude Code, Cursor, Codex and Antigravity. A command-line tool handles authentication, scaffolding and publishing. UiPath says nearly 60 percent productivity gains have shown up in early coding-agent use, according to its September earnings call.
Delegate is the second product. It is a desktop agent that acts on a user’s behalf across applications, files and business systems, on Windows and macOS. Users can describe, say or demonstrate a task, and Delegate executes it under the user’s own identity and audit controls. It ships with pre-authenticated access to more than 100 connectors and support for Model Context Protocol servers.
Cartographer is the third, and the one the press release sent to Indian media barely mentions. Founder and CEO Daniel Dines called it the most consequential announcement in the company’s history from the keynote stage. Cartographer builds what UiPath calls a “Map of Work”: a written record of how a business process actually runs. That record includes the exceptions and judgment calls employees handle but rarely document.
A companion feature, the Decision Ledger, records human corrections to agent decisions and feeds them back into the map. The Decision Ledger remains preview and roadmap, not product.
Governance Is the Product
Dines’ keynote argument was blunt. Large language models cannot bear consequences, learn on the job or be exact, he told the audience. His answer is not better models but a control plane around them.
The new governance stack backs that position. Model Hub gives teams central visibility into which models are in use and how they are routed. Runtime Checker validates agent behaviour against defined policies while an agent runs, not only at build time. An LLM-as-Judge guardrail evaluates outputs against custom criteria.
Compliance Packs map regulatory standards into trackable controls. Identity policies govern who, or what, can reach a given model, tool or resource.
One policy framework applies to every actor, agent, robot or person alike. Model calls route through UiPath’s AI Trust Layer regardless of the underlying model. No availability date was published for the governance controls, and no pricing accompanied any of the announcements. UiPath’s existing agent licensing charges 0.2 Platform Units per LLM call on standard models and 0.4 on premium ones.
Two quieter additions matter for enterprises with old systems. Integration Service now includes a Database Hub for SQL Server, Oracle and Databricks. Its Relay component reaches on-premises systems from cloud automations.
An MCP Connector lets workflows call any compatible MCP server as a native activity. Data Fabric adds zero-copy “Federated Entities” for sources like Salesforce and SAP. And the Automation Suite now runs the full agentic stack on Linux, on premises, for organisations with sovereignty requirements.
Testing Without Testers
The Test Cloud announcements point at a “dark testing factory”: testing that runs continuously, maintains itself and reports results without human hands. UiPath Autopilot can interpret a manual test and execute it against an application without prebuilt automation, from October 2026. Autonomous Exploration lets testers set a goal and send an agent to investigate an application and report findings, also from October.
Testing AI-infused applications arrives in public preview in December 2026, with semantic verification for outputs that vary between runs. A Playwright integration lands in public preview in October, letting teams run existing Playwright tests through Test Manager. SVP Gerd Weishaar framed the driver plainly: AI-generated code is widening the gap between software development and testing capacity.
The Numbers Behind the Pivot
The launch lands on improving financials. UiPath’s second quarter of fiscal 2027, ended July 31, brought revenue of $410 million, up 13 percent year-over-year. ARR reached $1.938 billion, up 12 percent. It was the company’s fourth consecutive quarter of GAAP profitability, with $32 million in GAAP operating income. Eighteen of its twenty largest deals in the quarter included AI components.
Two years ago, that growth rate would have read as a disappointment for a category once expected to compound much faster. Today it reads as a turnaround. The stock’s long slide and the RPA market’s maturation forced a reset, and the agentic pivot is the answer. The company also announced leadership changes alongside the quarter, and reaffirmed full-year guidance of $1.789 billion to $1.794 billion in revenue.
The Squeeze From Above and Below
UiPath still holds roughly 30 percent of the global RPA market and has been named a Gartner Magic Quadrant Leader for seven consecutive years. The pressure comes from both directions.
Microsoft bundles Power Automate with the Power Platform and sells Copilot Studio into every Microsoft 365 tenant. Entry pricing sits near $15 per user, against UiPath’s $420 per unattended bot. Salesforce’s Agentforce reached roughly $800 million in ARR by early 2026 and is absorbing CRM-adjacent automation into its own platform. ServiceNow pushes from the ITSM side.
Dines has acknowledged the asymmetry, noting his company competes against rivals with marketing budgets an order of magnitude larger. The strategy is visible in this launch: compete where complexity, legacy systems and governance requirements are highest, not where bundling wins. The Linux port, the compliance packs and the runtime policy enforcement are aimed squarely at the regulated enterprises the giants serve less well.
What to Watch
Three questions decide whether FUSION 2026 was a milestone or a trade-show. First, pricing: nothing shipped this week carries a public price, and buyers cannot model costs without one. Second, the Decision Ledger, the mechanism that turns human corrections into machine learning, exists today only as a roadmap item. Third, the December public preview of AI-application testing will show whether semantic verification can handle outputs that vary between runs.
Editor’s Note: This article draws on two press releases distributed by UiPath. TechRecast cross-verified them against the company’s FUSION 2026 announcements and documentation. Independent reporting from The Next Web, SiliconANGLE, Reworked and industry analysis publications informed the keynote and product context. Financial figures come from UiPath‘s second-quarter fiscal 2027 results, filed September 3, 2026.
The releases sent to Indian media carry a September 28 dateline. The announcements themselves were made on September 23 at FUSION in Las Vegas. Figures for three million developers, the 60 percent coding-agent productivity gain and availability dates are company claims. No pricing was published for any product announced this week. TechRecast contacted no parties for comment before publication.