Make: Visual Automation With AI Agents, Priced

Make is a visual automation platform where you build workflows called scenarios and, since its AI additions, agents that choose their own steps. The Free plan gives 1,000 credits a month; paid prices could not be read from its pricing page, and secondary sources disagree. It lists 3,500+ integrations, an MCP server and role-based governance. Checked 31 October 2026.

At a glance

What Make is

Make's product page calls it "The operating layer for the AI-era business" and promises "a glass box view of everything you build." Underneath the slogans it is a visual builder. You put modules on a canvas, one per app action, connect them, and run the result as a scenario. A trigger starts it; modules fetch, transform and send data; the execution history shows each run step by step.

Our comparison with Zapier and n8n in short: Make's canvas shows branching and data flow more openly than Zapier's linear steps, and it is cloud only, unlike n8n, which you can self-host.

How you build

From the product page:

• By prompt. Maia, Make's AI assistant, builds a first version of a scenario, which you refine on the canvas.
• From your AI chat, or by dragging and dropping modules.
• Custom code. Add JavaScript or Python inside any scenario.
• Error handling. Error handlers, automatic retries and exception routing are listed.
• Observing. Watch automations and agents as they run, and read execution history to see which step failed and why.

AI in Make

The product page lists two AI approaches, and the difference matters.

Other items on the page: an AI Toolkit, an MCP server and a category of AI apps.

What an agent is made of

The help centre's description of the earlier version of the app says each Make AI agent has five parts: an LLM as its "brain", a system prompt, context (memory), tools and reasoning. The reasoning part applies when you use a reasoning model.

• Instructions live in the system prompt, which sets the agent's role, limits and process.
• Knowledge is files in the agent's long-term memory, such as an internal knowledge base or company policies.
• Tools can be modules, scenarios and MCP tools. The help centre says to learn when to use each.
• Models. The app works with Make's AI Provider, and custom AI provider connections need a paid plan, per that page.

A newer "Make AI Agent (New)" app has its own help section, with key concepts, building a first agent, managing knowledge and a credit usage article. The older page carried a note pointing to it, and said the older app was in open beta so that functionality and pricing might change. We could not read the details of the new app.

The help centre's quickstart is a simple inventory management example, which gives a feel for the intended use: an agent with a few tools and a clear instruction, running inside a scenario.

Plans and prices

This is the part we could not pin down. Make's pricing page loaded but showed no paid prices in the text we read, and a direct text fetch was blocked. What the page does state:

• Free: 1,000 credits a month. "If you are on the Free plan, you have 1000 credits per month."
• Other plan names that appear: Core, Pro, Teams, Business (named in a start-up discount banner) and Enterprise.
• Annual payment in advance is cheaper than monthly, with no figures given.
• A Start-up Plan offers 50 percent off Business for companies with under 20 employees.
• For every 10,000 credits licensed, you get a usage allowance such as 5 GB of data transfer.
• Features that use Make's AI Provider "may use more" credits, with no amount given.

Search results from Make's own sites give conflicting numbers for the 10,000-credit tier: $9, $16 and $29 for Core, Pro and Teams in one place and $12, $21 and $38 in another. The likely explanation is annual against monthly billing, but no page we read confirms it. The same results mention one credit per module action and a 15-minute minimum interval between scheduled runs on Free.

Until you confirm on the live page, plan on a credit model: each module action in a scenario costs credits, AI features can cost more, and a scenario that runs every few minutes over several modules can use a plan's credits quickly. Work out credits per run, multiply by runs per month, then compare with Zapier's task tiers.

Governance

For teams, the product page lists:

• company SSO and 2FA enforcement
• role-based permissions: Owner, Admin, Member and Auditor
• private spaces for separate team workspaces
• Make Grid, a live view of running automations with ownership, status and audit history

These matter once agents act in business systems. Pair them with narrow connection permissions and an approval step for tools that send or change data.

Workflow or agent: an example

Take lead routing for a small sales team. The same job can be built two ways in Make, and the choice shows the difference between the two AI approaches.

As an AI workflow. You define the path. A form submission triggers the scenario. A module adds company data. An AI step reads the free-text message and returns one label: sales, support or spam. A router sends each label down its branch: sales leads go to the CRM and a Slack message, support goes to the help desk, spam is dropped. The AI makes one decision; everything else is fixed. It is cheap, predictable and easy to test.

As an AI agent. You give the agent a goal ("handle each new lead sensibly") and tools: look up the company, search the CRM for existing contacts, create a lead, post to Slack, create a ticket. The agent decides which tools to call and in what order. It copes with odd cases, such as a returning customer writing about a refund, but it uses more credits, behaves less predictably and needs tighter limits.

For most routing jobs the workflow is the better first build. Move to an agent only when the number of exceptions makes the router unmanageable. That mirrors the advice we give for code frameworks in agentic AI frameworks.

Estimating credits before you pay

Because the public pricing text is thin, estimate from your own scenario.

1. List the modules one run executes, counting each module in the path that actually runs.
2. Count runs a month. A scenario triggered by a form may run a few hundred times; one on a five-minute schedule runs about 8,600 times a month whether or not there is work.
3. Multiply, assuming one credit per module action unless the live pricing page says otherwise.
4. Add AI steps at a higher rate; Make says AI Provider features may use more, so test and read the credit log.
5. Add 30 percent for retries and testing.

A schedule-driven scenario is the usual cause of surprise bills on any credit platform. If the data arrives as an event, use a webhook trigger instead of polling. Make's help centre has a credit usage article for the new AI Agent app and for Knowledge; read it before building an agent that looks up files often.

Moving to or from Make

From Zapier. The concepts line up: a Zap is a scenario, an app action is a module, and Zapier's paths become routers. Rebuild one automation at a time and compare outputs.

From n8n. The same, with one difference: if you rely on self-hosting or on code-heavy nodes, check that Make's JavaScript and Python support covers it.

Away from Make. Scenarios can be documented step by step. Write down each scenario's trigger, modules and mappings, and the instructions of any agent, before you leave, since cloud scenarios do not export to another platform in a runnable form that we know of.

What it does well

• Visible logic. Branches, routers and data flow show on the canvas.
• Observability. Execution history shows what each run did and where it failed.
• AI in two flavours, agents for open-ended work and workflows for controlled decisions.
• Code when needed, with JavaScript or Python.
• Governance features for teams.
• A free tier to learn on.

Limits and cautions

• Unclear pricing in public text. Check the live page, and expect credit math.
• Cloud only. No self-hosted option in what we read.
• App-count claims differ between 3,500+ and 3,000+; check your apps individually.
• AI agents are evolving. The help centre has an older and a new app, with beta notes.
• Credit use of AI features is not specified on the pages we read.
• Vendor claims. "Trusted by 500,000+ teams" and "reduce manual work by a factor of 5" are marketing figures.

Make versus the others

See AI agent platforms for the full comparison, and AI automation agencies if you would rather hire the build.

Who should pick it

Make suits operations and marketing teams who want to see a workflow on a canvas, mix fixed steps with AI decisions, and keep an audit trail. It suits less well if you must self-host, if you need the largest possible app library, or if you cannot tolerate credit-based cost estimates.

A first scenario to build

Build something you can check by eye. For example: a new row in a spreadsheet triggers a scenario, an AI step summarises the row into one sentence, and the result is posted to a chat channel. It uses three modules, so you can compute credits by hand. Run it ten times, read the execution history for each run, and look at the credit log. If the numbers match your estimate, add a router and a second branch. If they do not, find out why before building anything bigger. The habit of checking credits on a small case prevents surprises on a large one.

Our take

This is an editorial reading of Make's product page and help centre, not a hands-on test.

Make's pitch is visibility: see every step, watch every run. That matters more once an AI agent is in the loop, because the hard part of agents is knowing why one did what it did. Its split between agents and workflows is also useful guidance. If you know the path, use a workflow and let AI decide one step. Use an agent when the route really depends on the input.

We would not decide on price from public pages today, since the paid numbers are not clear in what we could read. Start on Free, run one scenario with your real apps, record the credits it uses in a week, then price the plan from that. Keep the agent's instructions and tool list in your own document, so moving later, to Zapier or n8n, is a rebuild of wiring and not of thinking.

Checklist

• Decide whether the route is known (a workflow) or open (an agent)
• Start on Free and build one scenario with your real apps
• Record credits used per run and runs per month
• Confirm paid prices and credit rules on the live pricing page
• Set approvals for tools that send or change data
• Review execution history after the first ten runs
• Keep prompts and tool lists in a document you own