How to automate Make.com workflows with AI agents

Make.com (formerly Integromat) is a visual automation platform with a massive library of integrations. It connects to over 1,500 apps, and the visual builder is genuinely good for mapping out multi-step workflows. But most Make scenarios are static. They follow the same path every time.

AI agents change that. When you add an LLM call into a Make workflow, the scenario can classify, summarize, route, and respond based on the content it processes. Not just "if subject contains X, do Y" but "read this email, understand the intent, and take the right action."

How Make blueprints work

Make uses blueprints to define workflows. A blueprint is a JSON file that describes modules (nodes), their configuration, and how they connect. You can export any scenario as a blueprint and import it into another Make account.

This is what makes templates shareable. You download a blueprint, import it, plug in your credentials, and the workflow runs. No rebuilding from scratch.

The Runbooks gallery includes Make blueprints alongside n8n workflows and OpenClaw runbooks. Same concept, different runtime.

Template: Email to CRM with AI classification

The make-email-to-crm template watches your inbox for new messages and uses an AI module to classify each one. Is it a lead? A support request? A newsletter? Based on the classification, it routes the email to the right destination.

Leads get created as contacts in your CRM with extracted name, company, and email. Support requests get forwarded to your helpdesk. Everything else gets archived.

The AI classification step uses the OpenAI module in Make. You provide a system prompt that describes your categories and the model handles the rest. No regex, no keyword matching, no brittle rules that break when someone writes their email slightly differently.

Template: Slack Digest Bot

The make-slack-digest-bot template collects messages from one or more Slack channels over a configurable time window, sends them through an AI summarization step, and posts a digest to a summary channel.

This is useful for teams with high-traffic channels. Instead of reading 200 messages in #engineering, you read a 10-line summary of what happened. The AI picks out decisions, action items, and important announcements.

The summarization prompt is included in the blueprint and tuned for Slack-style messages. It handles threads, reactions, and file attachments gracefully.

How to import a Make blueprint

1. Download the blueprint JSON from the template page at tryrunbooks.com
2. Log into Make.com and create a new scenario
3. Click the three-dot menu and select Import Blueprint
4. Upload the JSON file
5. Connect your accounts (Gmail, Slack, OpenAI, CRM) in each module
6. Set your schedule and activate the scenario

Here is a snippet of what a Make blueprint looks like under the hood:

{
  "name": "Email to CRM with AI",
  "flow": [
    {
      "id": 1,
      "module": "google-email:TriggerNewEmail",
      "parameters": { "label": "INBOX", "maxResults": 10 }
    },
    {
      "id": 2,
      "module": "openai:CreateChatCompletion",
      "parameters": {
        "model": "gpt-4o-mini",
        "messages": [
          {
            "role": "system",
            "content": "Classify this email as: lead, support, or other. Return only the label."
          },
          { "role": "user", "content": "{{1.body}}" }
        ]
      }
    },
    {
      "id": 3,
      "module": "builtin:BasicRouter",
      "routes": [
        { "label": "Lead", "filter": "{{2.result}} = lead" },
        { "label": "Support", "filter": "{{2.result}} = support" }
      ]
    }
  ]
}

Tips for AI-powered Make scenarios

• Use gpt-4o-mini for classification tasks. It is fast and cheap. Save the larger models for summarization.
• Keep your system prompts short and specific. The more constrained the output format, the more reliable the routing.
• Add error handling modules after every AI call. Models occasionally return unexpected formats.
• Use Make's built-in data store for deduplication. Without it, reprocessed emails create duplicate CRM records.
• Start with a low execution limit (10 per run) until you trust the workflow. Scale up after testing.

Browse more templates

The Runbooks marketplace has Make blueprints for email routing, Slack automation, CRM pipelines, and more. Each template includes the blueprint JSON, a setup guide, and the AI prompts already tuned. Browse the full collection at tryrunbooks.com.