Septembre 2026

Notes de version de septembre 2026 pour les agents dans Automation Cloud et Test Cloud.

September 25, 2026

Build and debug agents with Autopilot Agent (Preview)

Autopilot Agent, a coding agent built into Studio Web, is now available. For agents, it can:

• Scaffold an agent with its prompts, tools, and memory
• Debug the agent as it runs in Studio Web
• Build evaluation sets, evaluators, and test cases
• Set guardrails

Autopilot Agent runs on the same UiPath skills and command-line interface (CLI) tooling as UiPath coding agents, and connects to your project and tenant from the browser with nothing to install. Autopilot Agent is now the default Autopilot experience. Classic Autopilot remains available, and you can switch between the two experiences.

For details, refer to Autopilot for Agents (Preview).

Correction, September 25, 2026: This feature was released on September 23, 2026, but was not included in the release notes at the time.

September 22, 2026

Using Data Fabric entities as agent context is now generally available

Agents can now use Data Fabric entities directly as a structured context, letting them query and reason over live or near-real-time, structured system of record without relying on static data snapshots. This lets users query this data in natural language, while existing enterprise access rules and permissions are enforced.

For details, refer to Contexts.

Introducing advanced agents (Preview)

Agents can now be built on the advanced harness, an alternative to the standard reason-and-act loop for long-running, document-heavy work. An advanced agent keeps its intermediate work in a file workspace instead of in the prompt, delegates slices of work to sub-agents and calls its tools programmatically from a sandboxed JavaScript interpreter.

During public preview, advanced agent consumption is 50% off the standard rate.

For details, refer to Advanced agents (Preview).

16 septembre 2026

Build agents with Coding Agents is now generally available

Building UiPath agents with a coding agent is now generally available. You can build both low-code and coded agents entirely from a coding agent (Claude Code, Codex, Cursor, and others) and terminal.

The uipath-agents skill covers the full lifecycle for both agent types — scaffolding a project, editing prompts and input and output schemas, adding tools, running evaluation sets, and packaging and deploying to Orchestrator. The skill is available to install via uip skills install.

Pour de plus amples informations, consultez les pages suivantes:

• Créer des agents avec des agents de codage
• UiPath pour les agents de codage

September 15, 2026

UiPath Agents in Flow is now generally available

UiPath Agents in Flow is now generally available, following its public preview in July 2026. You configure the agent's system prompt, model, input and output schema, and tools from the Flow canvas, with no standalone agent publishing step. Flow variables pass to and from inline agents natively, and multiple agents can be chained in series within a single flow.

For details, refer to Agents in Flow.

Correction, September 22, 2026: This feature was released on September 15, 2026, but was not included in the release notes at the time.

14 septembre 2026

Introducing conversational chat agents in Maestro Flow

Chat agents can now be built as a Maestro Flow, alongside the existing single-prompt Agent Builder path. Modeling the chat experience in Flow unlocks capabilities like intent-based routing, hand-off between specialist agents, deterministic replies, and parallel branches that run while the conversation continues. Both build paths surface to end-users as the same chattable Conversational Agent, meaning you can switch to Flow without disrupting how users reach your conversational agents.

For details, refer to Chat.

Introducing conversational voice agents in Maestro Flow

Voice agents can handle real-time phone calls — answering inbound calls and placing outbound ones. Audio streams straight to a real-time speech model, so callers can interrupt the agent mid-sentence, and tools run during the call.

You can hand off between specialist voice agents without dropping the call, feed the call transcript or a structured call summary into downstream nodes, and let the agent press digits to navigate an external phone tree. Numbers connect over a SIP trunk or your own Twilio account.

For details, refer to Voice.

9 septembre 2026

Les politiques de conception d'agents dans Studio Web sont désormais généralement disponibles

Les politiques de conception d'agents de la politique de gouvernance Automation Ops pour Studio Web sont désormais généralement disponibles. Les politiques vous permettent d’appliquer les exigences de configuration d’agent suivantes avant qu’un agent ne soit publié:

• Score minimum de l’agent
• Plage de température maximale
• Nombre maximum de jetons par réponse d’agent
• Escalade avec intervention humaine
• Nombre maximal d’itérations par exécution d’agent

Pour de plus amples informations, consultez la section Paramètres des politiques Studio Web.

7 septembre 2026

Présentation du garde-fou LLM en tant que juge [Aperçu]

Un nouveau garde-fou intégré, LLM en tant que juge, est désormais disponible en aperçu. LLM en tant que juge utilise un LLM pour évaluer les invites, les réponses ou les appels LLM des agents par rapport à des instructions personnalisées que vous écrivez en langage naturel simple, en comblant l’écart laissé par d’autres garde-fous prêts à l’emploi, qui ne détectent qu’une catégorie fixe et prédéfinie par le biais d'un service de classification géré.

Chaque vérification effectuée par le garde-fou effectue un appel LLM réel, de sorte que le modèle juge consomme des Agent Units ou des Platform Units en plus de la propre utilisation de l’agent, suivie séparément. Cela s’applique partout où le garde-fou s’exécute, y compris les exécutions d’évaluation d’agent lorsque le garde-fou est activé. Pour plus de détails, consultez Garde-fous prêts à l’emploi et Concession de licences.