Agents
Agents are AI systems you work with in plain language. They draw on your connected tools and company knowledge, act within the right permissions, and handle anything from answering a question to running a multi-step task on your behalf. Anyone can create an agent by describing what they want, no coding or training required. Agents in Sana are created in two different ways:
Ready-made domain agents are built by Workday for a specific area of work - like preparing for meetings, handling IT requests, or taking care of your travel bookings. They're available out of the box: just open one and start chatting. Think of them as on-tap experts for a particular domain.
User-built agents are ones you or a colleague create yourself, in plain language, for a task you want handled the same way every time - like drafting a performance review, or updating your CRM after a call.
The rest of this guide focuses on that second bullet point, building and running your own agents and frequently asked questions.
Agents built by users
Use Sana to build AI agents that carry out multi-step work for you - pulling from your connected tools, taking action, and pausing for your approval whenever a human decision is needed.
Introduction
A user agent is a reusable AI agent you (a user) create in plain language to handle an end-to-end task: preparing a performance review, drafting a response to an RFP, prepping for an account meeting, updating your CRM after a call. You describe what you want to build to the agent builder, and the agent builds an explicit, tracked, step-by-step plan it will follow every time it runs.
You stay in control through approvals (human-in-the-loop) that land in your Tasks inbox, where everything your agents need from you shows up in one place. This is also where you can find outputs your agents have finished working on for you.
Building a User Agent
Agents are built through conversation — no code and no workflow expertise required.
To create an agent:
The primary way to create an agent is to simply write what you want in the main chat. You can also press "Create Agent" from the Discover page or your agent overview page
Describe what you want it to do in plain language - for example, "Create an agent that pulls together a briefing with attendee roles from Workday, notes from our last meeting, and a suggested agenda and key talking points for me before each Google calendar meeting."
Sana generates a structured, multi-step plan, with a live preview building as you describe it, showing the trigger and the steps it will follow. A single agent's steps can span Workday and your connected third-party apps
Keep refining in the chat - add a step, change the wording, adjust tone of voice or output format, set the trigger - and the preview updates as you go. Everything is done in natural language by chatting.
When you're happy, save the agent. You'll give it a name, an icon, and a color so it's easy to recognize later
Editing an agent
You can change an agent at any time:
Edit the steps and logic: Click Edit on the Process panel to reopen the builder chat. Adjust the steps, trigger, or instructions in plain language, then save your changes.
Edit the details: Update the name, description, category, and icon from the edit details icon next to the agent name
The Agent Page
Users need a consistent and fast way back to their most important agents, a way to see what an agent is doing, and stay in control; without that visibility it's hard to trust an agent to run unattended.
This is why every agent has its own agent page - its home base. One consistent place to see how the agent is set up, check it's working, and catch and fix failures. View full run history with status, completion rate, and recent activity, and open any run to see exactly what it did step by step. From here you can also manage the agent's configuration and switch its automatic trigger on or off.
At the top of the page, you'll find the agent's icon and name on the left, and the main controls on the right:
Running (see Running an Agent below)
Run now - immediately start a run for manually run agents.
Use the Run with event option on your agent page to select any past event to immediately test-run the agent, instead of waiting for the next event to automatically trigger.
Pin to sidebar - this will pin the agent to your sidebar, for easy access.
Appearance editor - edit the agent’s name, category, description and icon.
Trigger toggle (automated trigger agents only) - turn an agent's automated trigger on or off. When off, the agent won't run on its schedule or event until you switch it back on.
The Process panel shows how the agent is set up:
Trigger - how this agent starts. This can be Manual, on an automated Schedule (e.g. Mondays at 9am), or based on an event in a third party system (e.g. a new email in your Gmail inbox). See Triggers below.
Steps - "The sequence of actions this agent performs." Each step is shown as a numbered card describing what the agent does at that point. Steps that pause for your sign-off carry an Approval badge, so you can see at a glance where you'll be asked to weigh in
The Runs area shows the agent's track record:
A completion rate overview and an activity heatmap for the last three months, so you can see how often you are running this agent, and its completion rate.
A list of every time the agent has run — see Run history below.
Running an Agent
There are two ways an agent runs:
Manually - click Run now on the agent's detail page. The agent starts immediately and opens a chat where you can watch it work through its plan step by step.
Automatically - when the agent has an active trigger (a schedule or an event), it runs on its own. Use the On / Off toggle to control whether automatic runs are active.
Use the Run with event option on your agent page to select any past event to immediately test-run the agent, instead of waiting for the next event to automatically trigger.
When an agent runs, it works through its plan in order. If it reaches an approval step or needs your input, it pauses and sends the request to your Tasks inbox (see The Tasks inbox and Human-In-The-Loop section). Otherwise it carries on until the task is complete.
Run history
Every run is recorded in the Runs list on the agent page. For each run you'll see its title, when it ran, and its status. You can sort by time or status and filter to a specific status. Click any run to open its chat and see exactly what happened.
A run can have one of these statuses:
Status | What it means |
Queued | Waiting to start. |
Running | Currently working through its steps. |
Action required | Paused, waiting for you (or another approver) to act. |
Completed | Finished successfully. |
Failed | Stopped because of an error. |
Stopped | Halted before finishing. |
Cancelled | Cancelled by you or the system. |
Skipped | The run was skipped. |
Triggers
A trigger decides when an agent runs. Each agent has one trigger, set up while you build it.
Manual - The agent only runs when you start it.
Schedule - The agent runs automatically on a recurring cadence (for example, every Monday morning). The detail page shows the schedule in plain language.
Event-based triggers - The agent runs based on an event, for example when a new email is received in Gmail or after a meeting is recorded. Useful for agents that summarize calls or update your CRM after a meeting.
Use the On / Off toggle on the detail page to activate or pause an agent's trigger at any time.
The Tasks inbox & Human-in-the-Loop (HITL)
Agents do the legwork, but you stay the decision-maker. The Tasks inbox is where everything your agents work on for you lands: outputs your agents finish, and any actions waiting on your review or approval. Nothing your agents do gets lost.
The Tasks inbox
Open the Tasks inbox from the left sidebar. A badge shows how many items need your attention.
Two categories in the Tasks inbox help with surfacing what matters and automatically filtering out completed tasks that do not require the user’s attention:
For your review: When an agent needs you to review something, it shows up here, clearly flagged as needing your attention. This can include getting your confirmation before it proceeds with an action (like sending an email on your behalf) or finished agent outputs that are ready for your review (like a drafted brief).
Completed: A record of all completed agent processes.
You can filter the inbox by All, Unread, Action required, or by a specific agent. Click an item to open it and see the full conversation and context behind the request.
Approving an agent's action
When you open an item that needs you, the agent's request appears as an approval card right inside the conversation. The card shows:
What the agent wants to do - the action and the system it affects (for example, "Salesforce — Create Deal").
The details to review - the fields the agent has filled in (such as deal name, amount, stage). You can expand with View more to see everything before deciding.
You then choose:
Approve - let the agent carry out the action and continue its plan. (The button may be labeled to match the action, e.g. Send or Approve or Submit.)
Deny - reject the action so the agent doesn't proceed.
The agent waits at this checkpoint until you respond - it won't take the action on its own.
FAQ
Q: What's the difference between an agent and just chatting with Sana?
A: Chat is best for one-off questions. An agent is a reusable, multi-step process you set up once and run again and again - on demand, on a schedule, or after a meeting - with built-in checkpoints where it pauses for your approval.
Q: Do I need to know how to code or build workflows?
A: No. You build agents entirely in plain language by describing what you want. You can refine them the same way at any time.
Q: Will an agent ever take action without asking me?
A: No. Any action in a third-party system like Workday, Gmail, or Sharepoint will require a human-in-the-loop approval, which will be sent to your Inbox. The agent will wait until it receives your approval to take this action, or abort mission if you deny the action.
Q: How do I stop an agent from running automatically?
A: Switch its trigger Off using the toggle on the detail page. It won't run on its schedule or event again until you turn it back on.
Q: Can I see what an agent did on a past run?
A: Yes. Open the agent and click any run in the Runs list to see the full conversation and every step it took.
Q: An agent run says "Action required" — what do I do?
A: It's waiting on you. Open your Inbox (or click the run), review the approval card, and choose Approve or Deny. The agent continues as soon as you respond.
Q: Can I send an approval to someone else, like my manager?
A: Routing approvals to other people and roles is part of the product vision but isn't available yet - today, requests go to the person the agent is running for.