After eight months in beta, Make.com's AI Agents feature became generally available this week to all customers on paid plans. The feature, which lets users build autonomous workflows that combine LLM reasoning with Make's existing 1,800+ app integrations, marks Make.com's biggest product expansion since its acquisition by Celonis in 2024.
I've been testing AI Agents in production over the past three weeks across two client deployments. Here's what I found.
What AI Agents actually do
An AI Agent in Make.com is a special module that wraps an LLM (Claude, GPT-4, Mistral, or Gemini) and gives it access to "tools" — which are essentially mini Make.com scenarios the agent can call autonomously to accomplish a goal.
The classic example: instead of building a deterministic scenario that says "if customer asks about pricing, send template A; if they ask about features, send template B," you build an agent that has access to your CRM, knowledge base, and email tool, and let it decide what to do based on the customer's actual question.
How it differs from beta
Three changes are worth noting compared to the beta version that's been around since September:
- Pricing model finalized. AI Agent calls now consume 5 operations per invocation (plus tool call operations). During beta, this was variable. The new pricing is predictable but more expensive than basic scenarios.
- MCP support added. You can now expose Make scenarios as MCP-compliant tools for use in Claude Desktop, Cursor, and other MCP clients. This was missing from beta.
- Visual debugger added. You can now see the agent's reasoning chain step-by-step, which makes debugging much easier than the beta's black-box output.
Where it actually works in production
From my testing, AI Agents shine in three use cases:
Customer support triage. Inbound emails or tickets → agent classifies urgency, routes to right person, drafts initial response. Works reliably once you give it 5-10 examples in the system prompt. Saves 60-80% of support team time on triage.
Lead qualification. Inbound form submissions → agent enriches via Apollo/Clearbit → scores against ICP → routes to sales or self-serve. Particularly powerful for B2B SaaS with complex qualification rules.
Internal knowledge retrieval. Slack questions → agent searches Notion + Google Drive + Linear → drafts answer with sources. Replaces "ask in #general" with instant accurate responses.
Where it doesn't yet work well
I had problems in a few areas:
Long-running tasks. Anything taking more than 5-10 minutes hits timeouts or context limits. The agent gets confused on multi-step workflows that span hours.
Cost predictability. A complex agent run can consume 50-100 operations easily. If you're on the Core plan (10,000 ops/month), you can burn through budget fast.
Hallucinations on factual data. Don't ask the agent to "look up customer's last 3 orders" without giving it specific tool access. It will sometimes invent data rather than say "I don't have access to that."
What it means for SMBs
If you're already on Make.com Pro or Teams plan, AI Agents are worth experimenting with this week. Start with a low-stakes use case (internal knowledge retrieval, draft generation) before putting agents anywhere customer-facing.
If you're on the Free or Core plan, the operations consumption will probably push you to upgrade. Budget accordingly.
If you've never used Make.com, AI Agents are not where you should start. Build a few traditional deterministic scenarios first to learn the platform — then add agents where they genuinely add value.
What I'm watching next
Three things to watch in the next 3-6 months:
- Will Make.com release pre-built agent templates for common use cases? They mentioned this on the launch call but no timeline.
- How will pricing evolve? 5 ops per call is reasonable now but will face competitive pressure from n8n's self-hosted approach.
- Will Zapier respond with their own agent framework? Their current "AI Actions" is much more limited.
I'll cover each of these as they develop.