Make added a step that picks its own next move

The setup

Make's AI-automation page, describing the platform as it stands today, positions AI agents as an addition to the company's existing scenario-based automation, built to be assembled on the same visual canvas Make customers already use for fixed if-this-then-that flows. A separate AI agents product page states the feature is available on all plans and is built directly inside the same Make canvas as a standard scenario, so setup for an existing customer means adding a new kind of module rather than adopting a new tool or paying a separate fee.

What the documents show

Make's own FAQ on the agents page draws the line explicitly: ChatGPT can respond. Make AI Agents take action, and, on the difference from an ordinary scenario, states that unlike a fixed scenario an agent can reason, choose what to do next, and trigger real workflows, complementing fixed logic with AI decision-making rather than replacing it. The AI-automation page frames the same distinction in build terms, listing adaptive decisions — letting an agent choose its route instead of hard-coded logic — as a difference from Make's traditional automation, alongside complete transparency, described as step-by-step logs and reasoning for every action the agent takes.

The friction

The agents page's own FAQ discloses the guardrail question directly: asked what happens if the AI makes the wrong decision, Make's documentation answers that a builder can set clear rules, add manual approvals, or stop the agent at specific points, and states agents work alongside deterministic logic rather than instead of it. That is Make's own account of where oversight has to be built in by the person configuring the agent — it is not automatic, and the same FAQ recommends a standard scenario instead of an agent whenever a step does not require judgment on unstructured input.

What changed in the work

The documents support a narrower claim than Make becoming autonomous: a builder who previously chained fixed steps can now hand one step a goal and let it pick its own path, with Make's own materials treating the resulting loss of a fully predictable flow as something to manage with rules and approvals rather than something already solved. The editorial read is that the setup cost shifts from mapping every branch in advance to defining the boundaries an agent is allowed to operate inside, which is a different kind of design work, not obviously a smaller one.

• Does this step require judgment on messy input, or would a standard scenario branch handle it just as well?
• Which specific actions has the workspace configured to require a manual approval before an agent can take them?
• Can every decision the agent makes actually be inspected in the reasoning log, not just assumed?

Make's own framing — reasoning that shows its work, alongside a plain admission that unsupervised agents can go wrong — is a useful marker for what agentic is meant to add over a scenario that already worked.