GatwAI

Make AI Agents

Make AI Agents for visual business workflows: the new version, tool selection, credits, privacy and differences from fixed automation rules.

Provider sources checked

What does Make AI Agents do?

Make AI Agents add steps that select suitable tools for variable inputs to the visual automation platform. A tool can be a module, scenario, MCP tool or another agent. Knowledge files provide context, while execution inspection helps explain the actions used. Current documentation explicitly covers the new agent version released on 2 February 2026. Not every process needs this flexibility. A fixed transfer between systems is easier to predict with an ordinary scenario. An AI step with a fixed output schema may suffice for summarization. Agents become useful when a request determines which information source is needed. Selection remains fallible, so approvals and restricted write access belong in the process design.
Editorial research · 2026-11-14How we research →

Who is this tool for?

  • Operations teams experienced with Make scenarios.
  • Business teams converting unstructured requests into controlled workflows.

Strengths

  • Agents combine with fixed scenario steps.
  • Knowledge files and multiple tool types extend the available context.
  • Operation and model-call consumption is documented.

Limitations and requirements

  • The new app remains labelled open beta.
  • Older agent guides can describe different behavior.
  • Long texts and repeated tool calls can increase consumption.

Who is behind the product?

Company
Make / Celonis
Headquarters
Not clearly established by the primary sources reviewed.
Company founded
Not clearly established by the primary sources reviewed.
Product launch
The initial beta was announced in April 2025; the current new app is documented from 2 February 2026.

Make is part of the Celonis group in its privacy notice. The agent-feature launch is not the founding of Make or Celonis.

Suggested ways to use it

Suggested workflows or verification steps. These are not hands-on tests; for unverified products they are not a recommendation to use the service.

Answer internal knowledge questions

  1. Load approved instructions as a small knowledge collection.
  2. Permit read-only tools and require sources in the answer.
  3. Escalate ambiguous questions to a person and check fidelity.

Goal: A proposed knowledge pilot with inspectable evidence.

Prepare supply requests

  1. Feed test requests from a form into a scenario.
  2. Let the agent retrieve permitted inventory data and draft a recommendation.
  3. Place orders only after separate staff approval.

Goal: A prepared case without uncontrolled purchasing.

Pricing, allowances and data

What should you check before choosing a plan?

Make's free plan includes 1,000 credits monthly. The new agent app is available with Make's AI Provider on all plans; custom provider connections require a paid plan. With Make's provider, an agent run incurs an operation charge plus token-dependent credits. Tools, files and knowledge preparation can add consumption. A custom model connection adds the external model bill. Dynamically displayed paid subscription prices are not presented here as verified fixed amounts.

What should you check about data and permissions?

Make, connected systems and model providers jointly form the data path. The security page states a default 30-day log retention; contractual and plan details may differ. Knowledge files should contain only necessary information. Review permissions, processing agreements, deletion and model terms before production. Displayed reasoning is not a guarantee that sensitive transmission did not occur.

External reviews and available evidence

A Make-forum post by Fabiano_Ferreira on 14 April 2025 reports positive beta experience and requests more model controls. It is one historical account, not a test of the new app.

Reported positives

  • The post describes a working initial agent flow.

Reported cautions

  • It is not representative; GatwAI has not tested these workflows.

These reports reflect the linked authors’ experiences. Test date, plan and task affect results; they do not establish a representative overall rating.

Product development and notable changes

  • New agent app

    Current help identifies a new version; older guides need separate interpretation.

    Source for this change ↗

Which alternative fits which task?

n8n →

For customizable workflows with optional self-hosting.

Frequently asked questions

Is every AI feature an agent?

No. Fixed AI calls and ordinary rules suffice for many tasks; agents additionally choose tools depending on the situation.

Does an agent run equal one credit?

No. Tokens, tools and knowledge processing can add credits or model costs.