# UiPath opens the platform to AI agents: Claude Code and Codex are already in > UiPath for Coding Agents - native integration of Claude Code and OpenAI Codex in the enterprise platform. Governance, audit trails and RBAC as standard. Find… - Source: https://snok.ai/en/news/blog/uipath-claude-code-codex-enterprise/ - Author: Jacek Bugajski - Published: 2026-05-13 --- ## One day that changes how companies build automations On 12 May 2026, UiPath announced something the market had been waiting for for months: a native, production-ready integration with AI coding agents. At launch: Claude Code (Anthropic) and Codex (OpenAI). Further models are planned for the second half of 2026. This is not a sandbox experiment. **UiPath for Coding Agents is already available to enterprise clients.** On the very same day that SAP, at SAPPHIRE, was announcing its Autonomous Enterprise vision. If you manage [process automation](https://snok.ai/en/offer/ai-automation/) within your organisation, you now face a rather different question to consider: not "whether to introduce AI into automation", but "which coding agent to start with - and how to keep it under control". --- ## What UiPath announced **UiPath for Coding Agents** is a platform-level integration that allows external AI agents to run directly within automation processes - with full governance, audit and access control built in as standard. Key facts: - **GA: 12 May 2026** - available immediately to enterprise customers - **At launch:** Claude Code (Anthropic) and OpenAI Codex - **Further integrations** planned throughout 2026 - **Orchestration layer:** Maestro, built on Temporal durable execution - **Governance as standard:** policy enforcement, audit trails, credential vault, RBAC, runtime controls UiPath CEO and founder Daniel Dines put it plainly: *"Now anyone can describe what they want, instruct a coding agent to produce it, and take it through the stages to production."* --- ## How it works - the architecture in one sentence UiPath is not building its own AI agent. Instead, it is creating a **neutral orchestration layer** that connects any agent to enterprise systems - while preserving full observability and control. In practice: - A developer (or a business analyst) describes in natural language what the automation should do - The coding agent (Claude Code or Codex) generates the automation code - **Maestro** takes this code through the testing, deployment and governance pipeline - At every stage: logging, audit trail, credential vault - regardless of which agent generated the code An important consequence: you can run **Claude Code in one department and Codex in another**. UiPath governs both equally. There is no lock-in to a single model vendor. --- ## Two groups this changes ### Developers and automation teams Previously: you wrote code in UiPath Studio, tested it, deployed it. Now: you describe the desired outcome, the agent generates the code, and you review and approve it for production. Prototyping speed increases many times over - particularly for recurring types of automation (API integrations, data transformations, RPA processes). What does not change: responsibility for code quality and process correctness remains with the human. The agent generates, the human approves. ### Process owners and business analysts This is potentially the more significant change. Previously: describing a new automation required a developer, a specification, and a sprint. Now: you can describe what you need in natural language, the agent produces a working draft, and IT approves it for production. The barrier between "I know what I need" and "I have a working automation" is dramatically lower. --- ## Governance - this is the substance, not an add-on Discussions about coding agents most often skip the question every CISO and every COO asks: *who is accountable for AI-generated code running in production?* UiPath answers this differently from most vendors. Rather than trying to "secure the agent" after the fact, **governance is built into the orchestration layer**: - **Policy enforcement** - every automation entering the platform passes through defined policies, regardless of source (human or AI) - **Audit trails** - a complete history: who approved it, when it was deployed, which model generated the code - **Credential vault** - the agent never has direct access to credentials; everything is mediated by Maestro - **RBAC** - role-based access control at the process level, not merely the platform level - **Runtime controls** - the ability to stop or roll back an automation without affecting other processes For organisations covered by [NIS2 and DORA](https://snok.ai/en/offer/sap-security/nis2-dora-audit/): this is not optional. Any automation system processing critical data must be auditable. UiPath for Coding Agents delivers this natively. --- ## Why this matters now - the market context A 2026 Deloitte report states that 79% of companies have AI agents "to some degree", but **only 11% have them in production**. The gap between proof of concept and production stems from precisely the issues UiPath has just addressed: governance, auditability, integration with existing processes. At the same time, SAP announced at SAPPHIRE 2026 more than 200 specialised Joule agents for finance, HR and procurement - partnering with Anthropic (Claude), Mistral and Cohere. UiPath is positioned at SAPPHIRE as the orchestration layer connecting Joule with legacy systems and external applications. In other words: **the two largest enterprise software ecosystems are opening up to AI agents at the same time** - SAP from the ERP process side, UiPath from the automation and orchestration side. Companies running both systems find themselves at the centre of this shift. --- ## How to act within 30 days **Step 1: Identify a single pilot process** The best candidate: an automation you already run in UiPath, but one whose maintenance takes up a disproportionate amount of developer time. Processes with a high volume of repetitive code (API integrations, reports, data transformations) respond best to coding-agent support. **Step 2: Check your governance readiness** Before running a coding agent in production, answer three questions: - Do you have defined access policies for automations in UiPath Orchestrator? - Does your IT team have an established code review procedure for AI-generated automations? - Do you know which data processed by your automations falls under NIS2 or DORA? If not - that is work to be done before deploying the agent, not after. **Step 3: Choose an agent and start with a prototype, not production** Claude Code and Codex have different characteristics - Claude Code performs better on complex, multi-step processes requiring reasoning; Codex on generating code from a precise specification. Start with one agent, one process, full human oversight. --- ## What SNOK recommends SNOK, as a [UiPath](https://snok.ai/en/offer/ai-automation/) Platinum Partner, works with Claude Code internally - the very same tool now natively entering the UiPath platform. We know from practice how these models behave with real enterprise processes. Three observations for organisations considering implementation: **First: UiPath's vendor-agnostic approach is a genuine advantage, not a marketing slogan.** In environments where different departments may prefer different AI models (IT vs operations vs finance), the ability to orchestrate multiple agents through a single platform eliminates the problem of governance fragmentation. **Second: NIS2 and DORA are not "later" problems.** If automations touch critical infrastructure, financial data or operational systems, auditability is a legal requirement, not an option. UiPath for Coding Agents meets this requirement natively. **Third: the barrier to entry has just been lowered, but integration expertise still carries value.** A coding agent generates automation code - but it does not understand your business processes, legacy systems or integration dependencies. An integrator who understands how everything connects is still required here. --- Would you like to see this in practice or discuss implementation within your organisation? [Get in touch](https://snok.ai/en/contact/) - we respond within 48 hours.