Yes. Workday Extend supports building custom agents: you build one in App Builder, either by describing it to Developer Agent or by writing its files yourself, deploy it to a tenant, and register it in Agent System of Record so people can use it inside Workday. Workday announced custom agents as "generally available as part of Workday Extend AI" on October 13, 2026, so the feature itself is not a beta.
The rest of this page is the documented scope: what you build in, what the agent can touch, what it needs from your contract, and what it cannot do.
Where do you build a custom agent in Workday Extend?
In App Builder on the Workday Developer Site. Workday's documentation says you can build the agent in an Agent app or in an Extend app, using Developer Agent or by hand.
An agent is two files in the open AgentSkills format: an agent definition (.agent) and one or more
skill files (.agentskills) that list the tools the agent may use. Developer Agent has a subagent,
Custom Agent Creator, dedicated to producing them. It can discover the available Agent-Ready Tools,
look up Custom Agent Actions you have already deployed, and write the skill file.
An Agent app is a stripped-down app type for exactly this. It supports only Agents, Skills, Attributes and Orchestrations, so it cannot hold pages (PMD) or Extend Model components. Workday states that Agent apps do not count against your Extend app entitlements.
Can you add a custom agent on top of an existing Extend app?
Partly, and the documented route is through orchestrations rather than through the app's pages or data model. Workday lets you build the agent inside an Extend app, so the agent can ship with an app you already have. And an orchestration in an Extend app can become a tool the agent calls: create it from the Custom Agent Action template, give it a display name and a description that tells the agent when to use it, and mark its effect level as Destructive, Idempotent, Read-Only or Open World. Workday's agents ask the user to approve every call to an action that is not marked Read-Only, so mark pure lookups Read-Only to avoid pointless approval prompts.
Workday's documentation does not describe an agent that drives an existing app's pages. The app's logic reaches the agent only through orchestrations you expose as actions.
What can a Workday custom agent call?
Workday documents three kinds of tools:
- Agent-Ready Tools. Workday's library of pre-built tools for reading and writing Workday data. In the tools explorer they come in three types: Basic (read-only), Advanced (write-only) and Prompt (read-only tools some Advanced tools require). The Agent-Ready Tools explainer goes through them.
- Custom Agent Actions. Your own tools, built as orchestrations. They can include connectors, your own processing logic and calls to Workday REST and SOAP APIs.
- Custom scripts. Scripts bundled in the skill, or scripts the model writes from the skill's instructions. Workday says these run in a secure environment with no network access, and that they cannot reach Workday APIs directly.
Workday adds that the agent runtime cannot make arbitrary API calls, even to Workday. Everything goes through the Agent-Ready Tools or Custom Agent Actions clients. If a task is not covered by a tool you registered, the agent cannot do it.
Who can use a custom agent once it is built?
Custom agents are delegate-mode only, meaning they act for the person who invokes them. After you activate one, people reach it from the Self-Service Agent chat panel in Workday, from Sana, through the business process framework, or over the Agent to Agent (A2A) and Model Context Protocol standards. Until an administrator configures and activates it in Agent System of Record, the agent is hidden from end users. You choose which security groups can use it with the Available To prompt.
How the agent's permissions combine with the user's is covered in the guide on securing custom agents. Workday's two pages describe the boundary in different words, so test with a user who has less access than you do before you rely on either.
What do you need to build one?
From Workday's build documentation:
- A connection to a Development tenant.
- Agent System of Record set up in that tenant.
- Workday Flex Credits, because Workday AI agents require them.
- Any Custom Agent Actions your agent uses, configured first.
On the Extend tier: Workday's Developer Agent page says it is offered through Workday Extend Professional. Workday's documentation changelog says it renamed the Extend Professional SKU to Extend AI, and its packages page lists Developer Copilot and the other AI features under Extend AI only. Ask your Workday account team to confirm what your contract includes, because the documentation does not state a price. The question of cost has its own page: is Workday Extend free?
Workday also says an agent deploys only to Development, Implementation and Sandbox tenants from App Builder. Moving it to a higher environment is done in App Manager, with Update Agents on the agent's configuration tab.
What a custom agent cannot do
- It cannot be deleted once registered. Deactivate it instead, which hides it from users.
- It cannot be used through a proxy. Workday says you cannot delegate agent interactions or use a proxy, so sign in as each test user.
- A script in the skill cannot call out to the internet or to Workday.
- An Agent app cannot contain pages or a data model.
Next step: build the first one
The how-to is building your first Workday custom agent: pick one narrow job, list its tools, and test it as different people before it goes live. If you do not have a tenant to build in yet, Workday now offers personal developer accounts; the getting started guide covers what that account includes.