Tasks & Automations
Agent Tasks
Two ways to put agents to work outside of chat, and how an Agent Task runs one agent on a single database record.
Think of your agents as employees who can work two different shifts. Sometimes you hand one of them a single file and say "take care of this one"; other times you set up a routine that runs on its own every morning. Both reuse the agents you've already built:
| Term | What it is |
|---|---|
| Agent Task | One agent runs once on a single record (one row of one of your databases), using that row's data as context, and writes the result back. This page. |
| Agent Automation | Agents chained into a multi-step routine that runs on a recurring trigger — a schedule, an incoming alert, an event, or a button you press. |
In a task, you pick an agent, pick one record, and write a one-line instruction — "summarize this," "score this," "draft a reply for this." Then the agent runs once with that row's full content in front of it. It can read every field, then update fields, write into the body of the record, or update a different record to set off the next agent in line.
This page is the short version, from the agent's side. The full guide — how to turn Agent Tasks on for a table, the different ways a task can run, what each status means, agent chains, how many can run at once, and the run history viewer — lives in the Databases section.
Let an AI agent work on this record and write its results back here.
Demo AgentBefore you start
- A saved agent you've already built — any agent will do, there's no special "task" type to create.
- The agent must be connected to the database where the record lives. You add that connection in the agent's Databases section.
- The agent needs at least read permission on that database. Give it write or full if you want it to update fields, write into the record body, or update other records to set off a chain.
If those three are in place, the agent shows up in the agent picker on any record in that database.
Step-by-step
- Open the database and click the record (row) you want the agent to work on. The record drawer slides open.
- Find the Agent Task section in the drawer.
- Pick the agent you want from the agent picker.
- Write a one-line instruction — what you want the agent to do with this record, in plain words ("Summarize the notes and set a follow-up date").
- Run the task. The agent reads the record, does the work, and writes the result back.
What the agent sees and does
When a task runs, the agent receives:
- The record's full content — every field name and its value, handed over as context.
- The instruction you wrote in the record drawer.
- Access to all of its normal tools and projects — so it can search the web, look up other databases, or reach other connected services as part of finishing the job.
The agent then writes its results back to the same record (or to another one) using its database tools.
Task or automation?
| You want to… | Use |
|---|---|
| Run a one-off job on a specific row | Agent Task |
| Use the row's own data as context | Agent Task |
| Set off another agent when this row's status changes | Agent Task |
| Run the same job every morning / whenever an outside alert arrives | Agent Automation |
| Coordinate several agents across multiple steps | Agent Automation |
| Send the flow down different paths based on a condition | Agent Automation |
The two work well together. Imagine new leads arriving through a Facebook form:
| Layer | Tool | What it does |
|---|---|---|
| Where leads come in | Meta Forms connector | New form submissions land as records in the Leads database |
| Per-record work | Agent Task on Leads | Each new row sets off a "Qualify this lead" task — the agent reads the lead's notes, scores them, and sets a Status |
| Recurring batch work | Agent Automation | Every morning a routine scans for high-score leads, writes personalized outreach copy, and sends it to Slack for the sales team |
Troubleshooting
| Problem | Fix |
|---|---|
| My agent isn't in the picker on a record | Connect the agent to that database in its Databases section, and confirm it has at least read permission. |
| The task ran but nothing was written back | The agent likely only has read permission. Raise it to write or full so it can update fields or the record body. |
| The task didn't set off the next agent in a chain | Chains need the agent to update the other record, which requires write or full permission on that database. |
| I need the task to run on a schedule or repeat | A task runs once. For recurring work, build an Agent Automation instead. |
What's next
- Agent Automations — combine agents into multi-step routines that run on schedules, outside events, or a button.
- Databases → Agent Tasks — the full guide: the ways a task can run, status meanings, chain protection, run limits and the run history viewer.