Make AI Agent Governance: Is It Ready for Marketers?

Marketing teams build a lot of automation on Make (formerly Integromat): lead routing, CRM syncs, email triggers, reporting pipelines. For a broader primer on the category, DevByteDaily’s marketing automation guide for scaling revenue covers the fundamentals. As those workflows add AI agents that make their own decisions, a new problem shows up fast: nobody can see what the agent actually did, or why. Make’s answer, announced at its Waves ’26 event on October 20, 2026, is a governance layer it calls MakeOS, built around a live dependency map named Make Grid. Make AI agent governance is the right idea for marketing automation. It is also, right now, only partly built.

This is a breakdown of what that governance layer actually covers today, what Make is still promising for later, and whether a marketing team running real automation should care yet.

What Make’s New AI Agent Governance Layer Actually Does

Make’s pitch, in the company’s own words, is that “businesses need a glass box, not a black box.” The idea is simple: every action an AI agent takes inside a scenario, whether it sends an email, updates a CRM deal, or triggers an ad budget change, should be visible, traceable to a specific agent or MCP client, and reversible if something goes wrong.

That sits on top of the AI Agents feature Make introduced in February 2026, which let users build agents directly on the same visual canvas as their automation scenarios. The original pitch there was a Reasoning Panel, a real-time log showing which tool an agent picked and why. Waves ’26 extends that idea from a single agent to an entire automation stack, with a few concrete pieces:

• Make Grid: a live map of scenarios, agents, and connected automations across an account, including some built on other platforms.
• Audit logs: a record of changes attributed to a specific person, agent, or MCP client, instead of a generic “something changed.”
• Custom Roles and Private Spaces: permission boundaries so a marketing team’s agents can’t quietly touch finance or ops scenarios.
• MCP toolboxes: a way to limit what an agent is allowed to connect to, run, or read, instead of giving it blanket access.
• Human-in-the-loop approval: a pause point that holds a scenario for sign-off before it continues.

Make also introduced Maia’s Plan mode, which shows a proposed set of changes to a workflow before building anything, and Make Process Intelligence, a task- and process-mining feature built on Celonis technology that is still on a waitlist.

Why This Matters for Marketing Automation Specifically

For a sense of how fast agents are spreading across marketing stacks generally, see DevByteDaily’s primer on what AI agents for marketing actually do. Marketing teams hand agents some of the riskiest jobs in the stack: deciding which lead goes to which rep, which contact gets suppressed from a send, or how a budget shifts between campaigns. A wrong call rarely causes a system outage. It causes a bad customer experience, a compliance problem, or a wasted ad budget, and often nobody notices until a client complains. Governance tooling matters here precisely because the failures are quiet.

What’s Actually Live Right Now

This is where the honest pros-and-cons split starts. Make’s own Waves ’26 post organizes the release into four pillars: Discover, Build, Govern, and Observe. Not everything in each pillar shipped on announcement day.

According to Make’s release page, these pieces are available now:

• Agent orchestration and Maia upgrades (Build pillar)
• Custom Roles and Private Spaces (Govern pillar)
• Make Grid, event subscriptions, and audit logs for agents and MCP (Observe pillar)

Make’s help center changelog corroborates part of this independently: an October 9, 2026 entry lists AI Agent fallback connections, prompt caching, and MCP tools for audit logs, which lines up with the audit-logging claim in the Waves ’26 post.

What’s Still Marked “Coming Soon” or on a Waitlist

Several of the features that make the governance pitch compelling are not shipped yet. Make’s own page lists them as “coming soon” or “coming later”:

In plain terms: the approval step that would actually stop a bad agent action before it happens, the toolboxes that would actually limit what an agent can touch, and the spend tracking that would tell a marketing lead what agents are costing, are all still pending. What shipped on day one is mostly visibility (the Grid, the logs, the roles). The control layer that sits on top of that visibility is the part still being built.

A Marketing Example: Lead Routing With an Approval Gate

Picture a common marketing automation setup: an agent reads an inbound lead form, scores it, and assigns it to a sales rep, with a separate branch that drafts a follow-up email for high-value leads. Before this governance layer, a marketing ops lead could only see that the scenario ran and check the output after the fact. With Make Grid and audit logs live today, that same ops lead can see which agent made the call and when, attributed by name rather than by scenario ID alone.

What they still can’t do, until human-in-the-loop approval actually ships, is make the agent pause and wait for a human okay before it sends that high-value follow-up email. Right now the honest version of “control” is after-the-fact visibility, not before-the-fact prevention. That is a meaningful gap for any team using agents on customer-facing sends.

The Pros of Make AI Agent Governance

A few things here are genuinely useful for a marketing automation team today, already built rather than promised:

1. Real attribution in the logs. Knowing that “Agent X” made a change, rather than a timestamp that only says “something changed at 3:47pm,” makes post-mortems on a bad send or a misrouted lead much faster.
2. Permission boundaries exist from day one. Custom Roles and Private Spaces mean a marketing team’s automations can be fenced off from finance or engineering scenarios without waiting for the rest of the roadmap.
3. It builds on a workflow marketers already use. Because agents sit on the same canvas as regular Make scenarios, a team that already automates in Make doesn’t need to learn a separate agent-building tool.
4. The Reasoning Panel is a real debugging aid. Seeing which tool an agent picked and why, in the same place the scenario runs, cuts down on guesswork when an agent-driven workflow misfires.

The Cons and Open Questions

The weaker parts of this release matter just as much as the strong parts:

1. The control features aren’t here yet. Human-in-the-loop approval, the single most important safety feature for customer-facing agent actions, is still “coming soon” with no public date.
2. No pricing clarity. Make’s pricing page doesn’t say which plan tier includes Make Grid, audit logs, or Custom Roles. A team can’t budget for this without talking to sales, a gap similar to the plan-tier confusion DevByteDaily found in its Klaviyo Composer pricing review.
3. Vendor-reported numbers aren’t independently verified. Make’s boilerplate cites customers cutting manual work “by a factor of 5” and reporting “50% fewer manual processes in under 6 months.” Those are Make’s own figures, not third-party audited results, and the company doesn’t name the source study.
4. “Reversible” is undefined. Make’s announcement says actions should be visible, permissioned, and reversible, but the post never explains what reversing an agent’s action actually looks like in practice.
5. AI spend tracking is missing. For a marketing team running agents against a budget, not knowing what each agent run costs until “Observe” matures further is a real planning gap.

Make AI Agent Governance vs. the DIY Alternative

Most marketing teams without this kind of tooling handle governance the manual way: a shared spreadsheet of which scenarios touch which systems, a Slack channel for “did anyone change this automation,” and a post-incident scramble through scenario history when something breaks. That approach scales badly past a handful of workflows, and it has no concept of per-agent attribution at all.

Make’s version is ahead of that baseline even in its current, partly-built state, mainly because of the Grid view and the audit logs. It is not yet ahead of a dedicated enterprise governance platform that already has mature approval workflows and spend controls, if a team already has one of those. For a mid-size marketing team that automates heavily inside Make already, the upgrade is close to free visibility on top of a tool they’re paying for anyway. DevByteDaily’s look at Sprinklr’s marketing AI agents beta shows a different vendor wrestling with the same visibility question from the social-platform side.

Make AI Agent Governance at a Glance

The table below lines up what matters for a marketing automation decision: current readiness, who it fits, and where the gaps sit.

Make AI agent governance at a glance: what shipped with Waves ’26 versus what marketing teams are still waiting on.

That split is the whole story of this release: the visibility half shipped, the control half is still being built.

Who Should Use This And Who Should Wait

It’s a reasonable fit for a marketing team that already runs lead routing, campaign reporting, or CRM sync scenarios in Make and wants to add agents without losing track of what those agents are doing. The Grid and audit logs alone justify turning agents on for lower-stakes, internal-facing tasks, like report generation or data enrichment, where a mistake is annoying rather than customer-facing.

It’s a weaker fit for a team planning to hand agents customer-facing actions, like sending emails or changing live ad spend, before human-in-the-loop approval actually ships. For that use case, the safer plan is to keep a human review step outside Make for now, built with existing scenario branching, and revisit once Make’s approval gate is confirmed live rather than “coming soon.”

Frequently Asked Questions

What is Make AI Agent governance? It’s a set of visibility and control features Make announced at Waves ’26 on October 20, 2026, including Make Grid, audit logs, Custom Roles, and Private Spaces, aimed at letting teams see and restrict what AI agents do inside automation scenarios.

Is Make’s agent governance available now? Partly. Make Grid, audit logs for agents and MCP, Custom Roles, and Private Spaces are live as of the Waves ’26 announcement. Human-in-the-loop approval, MCP toolboxes, and AI spend tracking are listed as coming later, with no confirmed date.

Does Make charge extra for these governance features? Make’s pricing page lists plan tiers (Free, Core, Pro, Teams, Business, Enterprise) but does not state which tier includes Make Grid, audit logs, or Custom Roles specifically. Confirming cost requires checking with Make directly.

How is this different from the AI Agents Make launched in February 2026? The February 2026 release introduced agents on the scenario canvas with a Reasoning Panel for a single agent’s decisions. Waves ’26 adds account-wide visibility and permissions across every agent and scenario, going well beyond one agent’s reasoning.

Should a small marketing team wait for the full MakeOS release? Not necessarily. The live pieces, especially Make Grid and audit logs, are useful today for low-risk internal automations. Waiting makes more sense specifically for customer-facing agent actions that need an approval gate before they fire.

The Bottom Line For Your Make AI Agent Governance Checklist

Make AI agent governance is a real step toward making agent-driven marketing automation auditable, and it reads as more than another AI feature announcement. The attribution and visibility pieces, Make Grid and the audit logs, work today and are worth turning on for teams already building in Make. The control pieces that would make agents safe to hand customer-facing work, especially human-in-the-loop approval, are still roadmap items without a date.

The practical next step: if a marketing team already automates in Make, enable Custom Roles and Private Spaces now, route new agents through low-stakes internal workflows first, and hold off on customer-facing agent actions until approval gates are confirmed live rather than “coming soon.” Check Make’s own changelog before trusting any third-party review that claims the full governance suite is already available, since several of its most important pieces are not.