The short answer
Make is a visual automation platform, formerly Integromat and now part of Celonis, where you build workflows called scenarios on a canvas with 3,000+ native apps. Maia by Make, upgraded at Waves '26 on 20 October 2026, builds, plans and fixes scenarios from a description. Make AI Agents, in open beta since February 2026, reason and call tools inside a scenario with a Reasoning panel. The free plan includes 1,000 credits a month.
| Made by | Make (formerly Integromat), part of Celonis |
|---|---|
| Free plan | Yes: 1,000 credits a month, Maia as a 30-day trial |
| Models | Make's AI Provider (Maia runs on OpenAI); paid plans can connect their own provider |
| Best at | Complex visual workflows mixing fixed steps with an AI agent |
| Shows sources | Not applicable; the Reasoning panel shows why an agent chose each tool |
| Team plans | Teams and Enterprise |
| Your data | Maia prompts and scenarios are not used to train the model; Maia requests are processed in the US |
What Make is now
Make lays a workflow out as a chain of circles on a canvas, one module per app action, with routers splitting the path and filters deciding what passes. Each workflow is a scenario. You pay in credits: most module actions cost 1 credit, routers and error handlers cost nothing, and some apps cost more, such as the Make Code app at 2 credits per second of run time.
Two AI features sit on top of that. Maia by Make is the assistant that builds scenarios with you. Make AI Agents are agents that run inside a scenario and decide which tools to call. Both are labelled beta, so expect behaviour and pricing to keep moving.
Maia: the co-worker that builds scenarios
Click + Create scenario in the top-right corner and the chat with Maia sits in the Scenario Builder. A mode selector offers Build together, which asks one question at a time and builds after each approval, and Plan first, which asks everything up front, checks your connections, app capabilities and plan access, and builds only once you approve the plan. Use Plan first for anything with more than three or four modules, because it catches missing connections before you waste a build.
The Waves '26 upgrade on 20 October 2026 added three things worth using. Memory keeps up to 50 private preferences of 200 characters each, such as "always use Google Sheets, not Airtable", and you can edit, clear or switch it off. Attach image takes up to three images per session, PNG, JPEG or WebP up to 1.5 MB, so you can photograph a whiteboard sketch and ask Maia to build it. Resolve error reads your run data, finds where a run stopped and proposes a fix you review before saving.
Maia is in public beta on all paid plans, and free accounts get a 30-day trial from signup. It has limits: it cannot see other scenarios, start from an existing scenario or call subscenarios or webhooks, and chat sessions are kept for 15 days.
Make AI Agents: reasoning you can watch
The new Make AI Agents app, announced on 11 February 2026, puts agents on the same canvas as everything else. Add the Make AI Agents (New) module called Run an agent (New), write its instructions and give it tools by selecting any app and module, choosing which inputs the AI fills and which you fix. Add files to its knowledge base or map file data in directly.
The standout is the Reasoning panel, which shows step by step why the agent called each tool. When an agent does something odd, you can see the exact decision instead of guessing. Live animations show when it is thinking and the order of its tool calls, and you can test it in a chat on the canvas. A Library of Agents gives ready-made examples for inventory, research, triage and reporting.
Agents run on Make's AI Provider on every plan, and paid plans can connect their own AI provider. Waves '26 also made agent orchestration, Knowledge and Make Grid, a map of everything built in your organisation, available, along with audit logs for agents and MCP. Human in the loop, data pinning and agent evals are listed as coming soon.
Concrete automations you can build
Lead research agent
A webhook from your website form starts a scenario. The agent, with a web search tool and your CRM module, researches the company, writes a two-line summary and scores fit from 1 to 5. A router then sends scores of 4 and 5 to Slack and logs the rest.
Support triage
A new email in a help inbox goes to an agent that reads it, checks the customer's plan in your billing app and returns a category and a draft reply. Fixed modules file the ticket and alert a person for anything urgent.
Weekly report
A scheduled scenario pulls last week's numbers from a sheet and an ads platform, and an agent writes a short commentary on the three biggest changes before a module emails it to the team.
Use your scenarios from Claude or ChatGPT
Make's MCP server lets Claude, ChatGPT and other AI clients run your scenarios. Set a scenario to active with on-demand scheduling, connect with OAuth or an MCP token, and select the run scope. Running scenarios works on every plan; the management tools need a paid plan.
Where Make falls short
- The learning curve is real. Routers, iterators and aggregators are powerful, but a non-technical user will be slower here than in Zapier.
- Beta everywhere. Maia and Make AI Agents are both beta, so features and credit costs may change.
- AI costs more credits. Make's pricing page says some AI features on Make's AI Provider may use more than the standard 1 credit, without giving a fixed rate.
- Fewer native apps. 3,000+ is plenty for most teams, but Zapier connects more.
How to get better results
- Describe the outcome, the trigger and the apps. "When a Typeform entry arrives, research the company and post a score to Slack" gives Maia everything it needs.
- Use Plan first for bigger builds, and read the plan before approving.
- Give agents few tools. Three well-named tools beat ten vague ones, and the Reasoning panel gets easier to read.
- Keep fixed actions in modules. Let the agent decide and draft; let ordinary modules send, file and update.
- Run Resolve error on every failed run before you start editing by hand.
Privacy and which plan suits whom
Make's Maia system card says your prompts, scenario structures and outputs are not used to train or fine-tune the underlying model, and Maia never has access to your connections or keys. Maia's provider is OpenAI, and Make's data zones page lists Maia requests as processed in the US even when your organisation uses an EU data zone. If you need strict EU residency, check the AI providers and data zones page before you switch Maia on. Enterprise owners and admins can turn Maia off in Feature controls.
Free, with 1,000 credits a month, is enough to learn the canvas. Core suits solo users. Pro adds custom variables and full-text execution log search, which you will want once agents run daily. Teams suits groups that build together and raises limits such as knowledge files for AI agents (150 per team instead of 100). Enterprise adds custom functions, enterprise apps such as Workday and ServiceNow, and the option to switch Maia off in Feature controls. See the Make pricing page for current prices.
Pick Make if you think visually and your workflows branch. If you want the simplest setup, read our Zapier guide; if you want self-hosting, read the n8n guide, or compare them on Zapier vs Make and Make vs n8n.
Prompts to copy
Ask Maia to plan a lead scenario
When a new entry arrives in [form app], research the company website, write a two-line summary, score fit from 1 to 5 against this ideal customer: [description], add the lead to [CRM] and post scores of 4 or 5 to the #sales Slack channel. Plan first and list every connection I need before you build.
You get: A full plan and a checklist of connections before anything is built
Agent instructions for support triage
You triage support emails. Use the billing tool to find the customer's plan. Return: category (billing, bug, how-to, other), urgency 1 to 5, and a reply draft under 120 words. If you cannot find the customer, say so and set urgency to 3. Never offer refunds or discounts.
You get: Consistent, structured triage you can route with a router
Fix a broken run
This scenario failed on its last run. Find the module where it stopped, explain the cause in plain English and propose the smallest fix. Do not change any other module.
You get: A diagnosis and a targeted fix to review
Build from a sketch
I have attached a photo of our approval process sketched on a whiteboard. Build it as a scenario using [apps]. Where the sketch is unclear, ask me before guessing.
You get: A first scenario built from your drawing
Weekly report commentary
Here are last week's and the previous week's numbers: [data]. Write a commentary of under 150 words on the three biggest changes, with the numbers, a likely reason for each, and one question the team should look into.
You get: A short, numbers-first summary to email the team
Save a useful memory
Remember: I always use Google Sheets for logs, Slack for alerts and British English in any text you write.
You get: Maia applies your defaults to every future scenario