UiPath Sold Clicks. Models Now Own the Mouse

UiPath, Inc. reported fiscal second-quarter results on September 3 for the period ended July 31. Revenue was $410 million, up 13 percent. Annualized renewal run-rate reached $1.938 billion, up 12 percent. GAAP operating income was $32 million. Cash, cash equivalents, and marketable securities sat at $1.405 billion. Management raised full-year revenue and ARR guides by a few million dollars. Eighteen of the twenty largest deals included an artificial-intelligence component. Then PATH, which had already printed ninety dollars in May 2021, fell about 17 percent into the September 6 close.

The line the tape actually read was net-new ARR: $37 million, against $49 million in the prior quarter. Last year’s second quarter had added $31 million, so the year-over-year print improved. The sequential step-down did not. Dollar-based net retention held at 109 percent. That is a healthy installed base expanding slowly, not a category re-accelerating because agents arrived.

Studio Sold a Chair in Someone Else’s Software

Daniel Dines and Marius Tîrcă founded the company in Bucharest in 2005. The product that later filled the S-1 was a trio. Studio let a developer, or a business user with a visual canvas, record a process. Robots executed it: open the window, copy the field, paste into the next system, click submit. Orchestrator scheduled the fleet, logged the jobs, and told an operations team which bot had stuck on a popup. The economic object was a software occupant for a seat whose work was to drive applications that did not talk to each other.

That was a real disruption. Shared-services centers and business-process outsourcers sold people who lived in SAP, claims screens, and Excel. UiPath sold a license that lived there instead, without waiting for the ERP vendor to expose an API. Blue Prism and Automation Anywhere sold cousins of the same idea. UiPath won the American funding race, listed on the New York Stock Exchange on April 21, 2021, and raised $1.3 billion. For a few weeks the market capitalized the click as infrastructure.

The fragility was in the surface. A robot that impersonates a user inherits every brittle selector and every UI redesign. It also inherits a competitor who already owns the desktop. Microsoft folded Power Automate into the Microsoft 365 bundle. The per-bot price list met a per-user suite. Growth decelerated while the company still talked like a platform.

Models Took the Same Surface Without the License

Generative models then ate the remaining specialty. A language model that can fill a form is a cousin of Studio. A computer-use model that drives the browser and the spreadsheet the way a junior analyst does is Orchestrator without the RPA certification. The clerk’s chair was never a moat. It was an interface gap. Once the general model can sit in it, the specialized occupant has to explain why a second runtime is still rent.

Enterprises have been acting as if that explanation were already in. Mid-level cuts arrived as a bet on agentic software before the production systems were ready. UiPath’s reply is the same sentence every incumbent in this stack now uses: agents hallucinate; robots are deterministic; someone must orchestrate humans, models, and bots under an audit trail. Maestro is that product. Maestro Case targets investigations and approvals. Maestro Flow lets a coding agent design a process as one artifact. The WorkFusion acquisition, announced with fiscal 2026 results, bolted on pre-built agents for anti-money-laundering and know-your-customer work — the regulated exception pile where a wrong click is a fine.

The pitch is coherent. Governance is a product. Banks will not let an unmanaged model loose in a sanctions queue. The financials have not yet shown that governance is a growth product. ARR is still compounding at 12 percent. Gross margin is still an 80 percent software margin. That combination is the SaaS-margin story in miniature: the vendor can remain necessary, even GAAP-profitable, while the multiple that assumed the click was a scarce layer collapses. UiPath spent $268.5 million on Class A buybacks in the first half of fiscal 2027. Cash generation funding shrinkage of the share count is what a mature automation utility looks like, not what a 2021 IPO was sold as.

Orchestration Has to Outrun a Bundled Mouse

Dines now describes a two-year platform transformation. Ashim Gupta dropped the dual COO-CFO title to run sales and delivery; Hitesh Ramani took finance. Outcome-based pricing is under consideration — a quiet admission that per-bot tariffs are the wrong unit if the customer is buying a closed case, not a runtime.

ServiceNow, Salesforce, and Microsoft do not need to acquire that runtime to threaten it. They already sit on the workflow, the CRM, or the desktop where the click happens. Summer chatter about PATH as a takeout is logically tidy and still unreported. An acquirer would be buying an install base and a governance layer — plus a sales force that still has to explain why a second mouse remains in the diagram.

The remaining product is not the click. It is the audit trail around a click that anyone’s model can now make.

Treat UiPath as a preview, not a museum. RPA was the first commercial attempt to automate work at the glass because the systems behind the glass would not federate. Computer-use models are the general version of that attempt. The test for investors is narrow and sequential: does net-new ARR re-accelerate as Maestro and the vertical agents move from the eighteen-of-twenty-deals talking point into the run-rate, or does 12 percent become the terminal growth of a well-run click utility. The principle is older than PATH. When the interface becomes cheap, rent moves to whoever still has standing to say which click was allowed.