Coworker's Agent Builder lets you create a custom AI agent that works autonomously on your behalf — no coding required. There are two ways to build one: using the guided AI creation flow (voice or text), or building manually by filling in the form yourself.
Two ways to build an Agent
Option 1: Build with AI (Recommended)
The easiest way to create an agent is to describe what you want it to do — either by speaking or typing. Coworker's AI will interpret your description and pre-fill the agent configuration for you.
Go to Agents in your Coworker workspace and click Create New Agent.
Choose your input method:
Voice — Click the microphone and describe your agent out loud. For example: "Create an agent that scans my HubSpot pipeline every Monday morning and sends me a Slack summary of any stale deals."
Text — Type your description directly into the prompt field. Same idea — just describe the task, the data sources to use, how to format the output, and when it should run.
Review the pre-filled configuration. After you submit, Coworker will generate an agent name, system prompt, suggested tools, and trigger settings based on your description. Review each section and adjust anything that doesn't look right.
Save and test. Run the agent to confirm it works as expected, then iterate on the instructions as needed.
ℹ️ Tip: The more specific your description, the better the generated config. Include details like which tools to use, what to look for, and how you want the output delivered.
Option 2: Build manually
If you prefer to configure every setting yourself, you can skip the AI creation flow and fill in the form directly.
Go to Agents and click Create New Agent, then select the option to build manually (or skip past the voice/text step).
Name your agent with a clear, descriptive label — for example, "Weekly Pipeline Digest" or "Morning Slack Summary."
Write your system prompt. This is the core instruction set for your agent.
Be specific about:
The task or workflow to execute
What data to look at or act on
How to format the output (e.g., Slack message, email, report)
Any constraints or rules the agent should follow
Select tools and data sources. Grant the agent access only to what it needs — for example, HubSpot for deal data, Slack for messages, or Jira for ticket status.
Set a trigger. Choose how the agent gets kicked off:
Manual — Run on demand from the Agents page or by calling it from chat
Scheduled — Run on a recurring schedule (e.g., every Monday at 9am)
Heartbeat — Run at a frequent interval (e.g., every 30 minutes) to monitor for changes
When mentioned in Slack — Run when a user calls the agent via its Slack command
Configure output delivery. Decide where the agent should send its results — Slack, email, or the Coworker interface.
Save and test. Use the Run Now button to trigger a test run and review the output. Adjust your instructions as needed.
Starting from a template
Not sure where to start? Agent Builder includes a library of ready-to-use templates for common workflows — from daily digests and pipeline monitoring to PR reviews and customer health tracking. Templates come pre-filled with a system prompt, tools, and trigger settings that you can customize to fit your needs.
To use a template, click Browse Templates on the Agents page and select one that matches your use case.
Tips for building effective agents
Start simple. Build a single-task agent first, confirm it works reliably, then expand its scope.
Be specific in your prompt. Vague instructions lead to inconsistent results. Tell the agent exactly what to do, what to use, and how to present its output.
Limit tool access. Only connect the data sources the agent actually needs — this keeps it focused and reduces noise.
Use draft mode for write actions. If your agent sends messages, creates tickets, or updates records, enable draft mode so you can review actions before they execute.