Skip to main content

Agents

Getting started with Coworker's Agent Builder and scheduling an agent

Updated this week

What is Agent Builder?

Agent Builder is Coworker's no-code interface for creating custom AI agents that work for you autonomously. With Agent Builder, you can give an agent instructions in plain language, connect it to your data sources and tools, and set it to run on a schedule or trigger - all without writing a single line of code.

What Can Agents Do?

Agents built in Coworker can be designed to handle a wide variety of recurring tasks. Here are some examples from our early testers and internal users:

  • Deal Re-Engagement: Automatically scan your CRM for stale deals and send re-engagement summaries or Slack alerts on a schedule

  • Accounts Payable Automation: Forward invoices (e.g., from Stripe or email) to an agent that routes and processes them automatically

  • Legal Document Review: Set up an agent to review incoming customer redlines against standard contract terms

  • Slack & Email Digest: Build a 30-minute heartbeat agent that summarizes activity across Slack and email and delivers a digest

  • Sales Pipeline Reports: Generate regular pipeline reports from your CRM data and post them to a Slack channel

  • Custom Workflows: Connect to tools like Zapier, Make, or your own API to trigger agents from external events

How to Create an Agent

Follow these steps to build your first agent in Coworker:

  1. Navigate to Agent Builder
    Go to your Coworker workspace and open the Agents section. Click Create New Agent or access the Builder from the agents panel.

  2. Name Your Agent
    Give your agent a clear, descriptive name that reflects its purpose (e.g., "Stale Deal Re-Engagement Assistant" or "Weekly Pipeline Digest").

  3. Write Your System Prompt
    In the system prompt box, describe what you want your agent to do in plain language. Be specific about:

    • The task or workflow to execute

    • What data to look at or act on

    • How to format its output (e.g., Slack message, report, email)

    • Any constraints or rules it should follow

    Tip: The more detailed your instructions, the more reliably your agent will behave as expected. You can always edit and refine the prompt later.

  4. Select Your Tools & Data Sources
    Choose which tools and data sources your agent can access. This may include:

    • CRM tools (HubSpot, Salesforce)

    • Communication tools (Slack, Email)

    • Productivity tools (Google Drive, Notion, Jira)

    • Custom tool packages configured for your organization

    Tip: For best results, only grant access to the tools your agent actually needs for its task.

  5. Choose an AI Model
    Select the AI model that will power your agent. Different models may be better suited for different tasks (e.g., reasoning-heavy workflows vs. fast, high-volume processing). Options currently include models from Claude, ChatGPT, and Google Gemini.

  6. Configure a Trigger or Schedule
    Decide when and how your agent runs:

    • Manual: Run your agent on-demand from the Coworker interface

    • Scheduled (Cron): Set your agent to run at a recurring interval (e.g., every 30 minutes, daily at 9am)

    • API Trigger: Trigger your agent programmatically via the Coworker API, or connect it to tools like Zapier or Make to fire it from external events

    • Custom Event: Configure your agent to respond to specific events in your connected systems

  7. Save and Test Your Agent
    Save your agent configuration and run a test to make sure it behaves as expected. Review the output and adjust your system prompt or tool access as needed. You can edit your agent at any time.

Configuring the API Trigger

If you want to trigger your agent from an external system or workflow automation tool (e.g., Zapier, Make), you can use Coworker's API trigger.

API Endpoint: https://api.coworker.ai/agents/trigger

Accepted fields for API calls:

  • question - The input or prompt to send to the agent

  • agentId - The unique ID of the agent to trigger

  • conversationId - (Optional) To continue an existing conversation thread

  • chatType - The type of chat context

  • userTimezone - The timezone for the triggering user

  • idempotencyKey - A unique key to prevent duplicate triggers

To find your API key, navigate to Settings > API Keys in your Coworker workspace.

Note: Webhook output (the ability for agents to push results to an external webhook) is on the product roadmap and coming soon.

Using Agents via Slack

Once configured, your agents can also be accessed and invoked directly from Slack. Coworker's Slack integration allows you to:

  • Receive agent outputs and notifications in Slack channels or DMs

  • Configure which channels an agent can post to (scoped access is recommended to prevent agents from posting to unintended channels)

  • Set up Slack-based triggers to run agents from Slack messages or events

Best Practice: When configuring an agent that writes to Slack, always scope its write access to specific channels or DMs. This prevents unintended messages from being posted to the wrong audience.

Tips for Writing Effective Agent Prompts

  • Be specific about the output format. If you want a Slack message, say so. If you want a structured report, describe the format.

  • Specify what NOT to do. For example, if your agent has Slack write access but should only DM - explicitly instruct it: "Do not post to public channels. Only send direct messages."

  • Start simple, then iterate. Begin with a focused, single-task agent and expand its capabilities once it's working reliably.

  • Use clear data references. Tell the agent exactly where to look (e.g., "Check HubSpot for deals that haven't had activity in 30+ days").

  • Set context boundaries. Let the agent know what time range, team, or dataset is relevant to its task.

Go to Agents

Related Articles
How to add Slack to Coworker
How to add Ashby to Coworker
How to add ClickUp to Coworker
Scheduled Agents
When should I use an Agent?
Did this answer your question?