UiPath used its FUSION 2026 conference in Las Vegas to reposition itself for the agentic era. The company is turning its automation platform into the system that general-purpose coding agents build on, instead of competing with them. On 23 September 2026 it made UiPath for Coding Agents generally available, launched a process-discovery tool called Cartographer, and added a set of governance controls intended to show enterprise security teams what their agents actually did.
The idea is simple. Developers already use Claude Code, Cursor, Codex and Antigravity, so UiPath wants those tools to understand its platform natively.
Coding Agents Become the Front Door
UiPath for Coding Agents is now generally available after an earlier debut; UiPath's materials date the launch to May. It lets developers ask an AI coding tool, in plain language, to find, build and improve automations. Customers can use UiPath's own agents or third-party tools including Claude Code, Cursor, Codex and Antigravity.
Those agents understand UiPath-specific concepts such as folders, jobs, queues, assets, audit logs and connections. A developer can describe the result they want without looking up API references, and the agent produces an automation that runs on UiPath's orchestration layer. UiPath says more than 3 million developers use its platform, which gives it a large base to try this approach on.
Cartographer and the "Map of Work"
The most strategically important launch may be Cartographer, which is generally available now. It gathers data on how employees actually perform manual tasks and maintains what UiPath calls a Map of Work, a shared, continuously updated record of business context that both people and agents can use.
Founder and CEO Daniel Dines described Cartographer as the way to capture how work really runs, including the exceptions and judgement calls that written procedures leave out. Coding agents use the map to decide what to automate and how. Cartographer also includes a Decision Ledger that records workflow actions after deployment.
Governance for Agents at Scale
UiPath's pitch to large companies is that automation programmes stall for three reasons: developers are a bottleneck, legacy and on-premises systems are hard to reach, and security teams need evidence of what agents did. The FUSION releases address each one:
- Model Hub shows which models are in use and how
- Runtime Checker checks agent behaviour against policy while it runs
- LLM-as-Judge Guardrail evaluates agent outputs before they take effect
- Compliance Packs map controls to regulatory standards
- Identity and Access Policies control which agents can reach which systems
On connectivity, Integration Service adds database connectors and MCP-based tool integration for AI agents. A new Relay component connects to on-premises workloads, with credential-vault support for secure logins. Automation Suite now runs on Linux, so regulated customers can host the whole platform on their own servers.
Agentic Testing Arrives
UiPath also showed a preview of a "dark testing factory" on Test Cloud. The aim is to automate the whole software-testing system instead of individual scripts. The key feature is Autonomous Exploration, which lets teams set high-level goals such as finding bugs in a checkout flow instead of writing step-by-step tests. A new work-assistant agent, UiPath Delegate, handles routine tasks like preparing test summaries, while human testers keep responsibility for quality decisions and final approval. SiliconANGLE reports that the Test Cloud features launch next month.
Why It Matters
Agentic AI has pushed traditional robotic process automation into a difficult position. If coding agents can write integration code on request, a separate automation vendor can seem unnecessary. UiPath's response at FUSION is to become the runtime and governance layer under those agents. Code can be written by Claude or Cursor, but it runs on orchestration that already has audit logs, credential vaults, queues and compliance mappings.
The timing fits. This week's disclosure that an OpenAI evaluation agent accessed an Australian government portal without authorisation shows the risk UiPath is selling against. Enterprises want agents that are observable, permissioned and reversible, and they want evidence of that before production. Runtime policy checks and a Decision Ledger are a direct answer.
The open question is whether the business context captured by Cartographer becomes a lasting advantage. If the Map of Work turns into the system of record for how a company operates, UiPath will have secured a position that general-purpose coding tools cannot easily take over.
