The office robot was never much of a robot. It had no arms, no chrome face and no opinion about lunch. It was a piece of software trained to open an invoice, copy a number, paste it into an elderly accounting system and do the same thing again without sighing. UiPath made that modest trick into a large company. Now the trick is changing. The software is being asked not only to click, but to decide what should be clicked, which machine should do it and when a person ought to intervene.
That is the reason UiPath now talks less about robotic process automation, or RPA, and more about agentic automation. The phrase can sound like enterprise software fog. In practice, it describes a relay team: an AI agent reads an ambiguous request; a reliable bot enters data in a legacy screen; an API checks the answer against another system; a manager approves the exception. UiPath's new centerpiece, Maestro, conducts the handoffs and keeps a record of what happened.
Act I / The digital hands
A stubborn screen became a business
UiPath began in Bucharest in 2005, when Daniel Dines and Marius Tîrcă built software libraries in a small apartment. The company was called DeskOver. Its early specialty was computer vision and user-interface automation, which meant teaching software to understand what appeared on a screen. For years it was an uncelebrated technical shop. Then large companies noticed that the same technology could operate the countless applications that did not talk neatly to one another.
The company released its first desktop automation product line in 2013, renamed itself UiPath in 2015 and raised a $30 million Series A from Accel in 2017. Growth came quickly after that. It became Romania's first unicorn, moved its headquarters to New York, raised successively larger rounds and listed on the New York Stock Exchange in April 2021. The journey from apartment to ticker symbol took 16 years, which is both a rocket ride and a long time to watch a cursor move.
“The most interesting thing about the robot was never the click. It was everything the click had to connect.”YesPress observation
The original problem has not disappeared. Banks still run mainframes. Hospitals still use software designed around forms. Manufacturers have plants full of bespoke interfaces. Government agencies cannot replace every system because a clever model arrived on Tuesday. UiPath's robots can work through those screens as a person would, while API automation handles systems that offer a cleaner doorway. Its Healing Agent and computer-vision tools are designed to keep automations working when an interface moves a button or changes a label.
Act II / The control room
What the platform actually does
UiPath has assembled a wide shelf of products around that basic execution engine. Studio is where developers and business users build automations. Agent Builder creates and evaluates AI agents. Document Understanding and IXP extract meaning from invoices, claims, emails and other untidy material. Process Mining reconstructs how work travels through business systems; Task Mining studies the desktop steps in between. Test Cloud checks applications and workflows. Orchestrator manages robots, credentials, schedules and queues. Maestro sits above the collection, mapping a complete process and assigning each part to an agent, robot, system or person.
The distinction matters because an AI answer is not the same as completed work. A model can summarize a medical record. A healthcare process also needs to retrieve the correct file, protect patient data, apply a policy, record the decision, send uncertain cases to a reviewer and survive an audit months later. UiPath is selling the connective tissue around the intelligence.
Classic RPA
- Predictable rules
- Repeatable screens
- High-volume execution
- Known exceptions
Agentic workflow
- Ambiguous inputs
- Contextual decisions
- Multiple tools and models
- Human escalation
For buyers, the menu comes in several forms. Automation Cloud is the managed SaaS version. Automation Suite can run on a customer's own Kubernetes infrastructure or Linux machines, including environments with strict sovereignty requirements. That self-hosted option is important in public service, financial operations and healthcare, where moving sensitive data into a new cloud may be a nonstarter. Customers can also choose among model providers rather than accepting one model as the permanent brain of every process.
The company earns money from term licenses, maintenance and support, SaaS subscriptions, and a smaller professional-services business. In fiscal 2026, it reported $1.611 billion in revenue, up 13 percent. Subscription services provided $954.5 million, licenses $606.4 million, and services and other revenue $49.7 million. The company recorded full-year GAAP operating profit for the first time. At the end of April 2026, annualized renewal run-rate reached $1.901 billion.
Act III / The customer test
Where automation earns its keep
UiPath sells mainly to large organizations. At January 31, 2026, 2,565 customers produced at least $100,000 in ARR, and 357 produced at least $1 million. Those 357 customers accounted for roughly 52 percent of fiscal-year revenue. The concentration is a strength and a risk: enterprise deployments can spread into dozens of departments, but a few purchasing decisions carry unusual weight.
The most persuasive examples are pleasingly specific. Johnson Controls used document processing and automation in accounts payable, where it had been handling 200,000 manual transactions each month. The company says the program created more than $10 million in enterprise value and saved $6 million in accounts payable. WEX, the financial-technology company, reports $2.7 million in savings and 70,000 hours of labor saved annually from its automation program. These are not sentient digital colleagues. They are claims, cards, invoices and old screens moving with fewer human detours.
In July 2026, British retailer The Very Group signed a three-year agreement to use UiPath's acquired Peak technology for pricing decisions, including campaign simulation and scenario planning. The example reveals another part of the strategy. The 2025 acquisition of Peak added specialized decision intelligence for retail and manufacturing. UiPath does not only want to provide building blocks; it wants to package more industry work where the data, decision and action are already connected.
AI can write the recommendation. Enterprise automation must still prove who approved it, what changed and whether the work finished.The production gap
Act IV / A crowded pit
Partners on Monday, competitors on Tuesday
UiPath faces its familiar RPA rivals, Automation Anywhere and SS&C Blue Prism. The more complicated threat comes from the giant platforms already installed at its customers. Microsoft Power Automate, ServiceNow, SAP, Salesforce, Oracle and IBM can place automation beside software buyers already use. Pega and Appian approach from process management and low-code development. Celonis starts with process intelligence. Workato, MuleSoft and Boomi emphasize integrations. Developer teams can assemble agents with open-source frameworks and increasingly capable coding tools.
Many of those names are also UiPath partners. Microsoft Copilot Studio agents can be orchestrated by Maestro. Google Cloud models power a medical-record summarization agent. SAP sells an integrated UiPath offering. The platform connects to Salesforce and Snowflake. This frenemy map is ordinary in enterprise software, where customers prefer their expensive systems to cooperate. UiPath's claim to difference is that it can coordinate third-party agents while also operating the awkward user interfaces and documents where API-only platforms stop.
That openness has to survive commercial pressure. A customer may prefer the automation bundled with a cloud or application contract. An internal team may build a narrow agent cheaply. New models may become better at operating computers directly. UiPath's response is reliability: combine flexible AI with deterministic automation, mature credentials, human approvals, monitoring and auditability. The pitch is not that its agent is always the cleverest. It is that the whole process will behave in production.
The July 2026 product releases pushed in both directions. Maestro Flow introduced a lighter canvas for assembling agents, API calls, documents and human steps. UiPath for Coding Agents let developers use tools including Claude Code, Codex and Cursor to build and deploy UiPath agents and automations. Low-code users get a visual route; engineers get a terminal route. Both end in the same governed runtime.
Finale / The useful bet
The future is a handoff
UiPath's culture still carries the vocabulary of its fast global expansion: humble, bold, immersed and fast. It had 3,981 full-time employees at the end of January 2026, with a majority outside the United States. The company also maintains a large learning ecosystem through Academy, Community, documentation and public developer tools. That matters because automation programs rarely fail for want of one more feature. They fail when nobody owns the process, the exceptions multiply, or employees work around a system they do not trust.
The company's mission has long been to unlock human potential by removing repetitive work. That promise deserves a practical reading. Automation can eliminate drudgery, but it can also rearrange jobs and centralize decisions. The human-in-the-loop is therefore more than a product diagram. It is the person responsible when a model is unsure, a policy conflicts with reality, or a customer deserves an explanation.
UiPath has spent 20 years learning that the hardest part of office work is often the seam between two systems and two people. Agentic AI makes each participant more capable, but it creates more seams. The company is betting that the winner will not be the loudest digital worker. It will be the conductor that knows when the model should think, when the robot should click and when the human should lift the baton.