If 2025 was the year automation platforms added AI steps, 2026 is the year they rebuilt around agents. Make, Zapier and n8n all shipped agents, AI assistants that build workflows from a description, and MCP support that lets AI tools like Claude and ChatGPT call your automations directly. Here is what changed on each.
Make
- AI Agents on the scenario canvas (February): build, run and debug agents inside the same visual canvas as your scenarios, and share them across the organization.
- MCP tools and the Make MCP Server (June): agents can use MCP tools, and AI clients like ChatGPT, Claude and Cursor can call Make scenarios as tools. Make says it is included on every plan, and you pay for the credits the scenarios use.
- Claude in the Make AI Provider and AI Sub-Agents (July): Claude models are available without your own Anthropic key, billed in Make credits, and one agent can hand tasks to specialist sub-agents.
- Maia (August): a conversational assistant that turns plain-language requests into scenarios and agents, step by step in the visual builder.
- Builder improvements through the year: If-Else and Merge, scenario recovery, a command-line tool for CI/CD, reusable functions, undo and redo, and AI explanations of errors.
Zapier
- AI by Zapier: agent features now live in one AI step inside the Zap editor, with approval per tool, a choice of model tier and structured outputs. Long runs pause for human review.
- Model-based pricing (from June 15): each AI step costs tasks according to the model tier, drawn from one shared pool with automation and MCP.
- Next Gen Zaps (September, early access): describe a workflow in an AI tool and deploy it to Zapier, or build with natural language, code or a canvas, with branches that rejoin, nested loops, continue-on-error and human-in-the-loop pauses.
- One MCP server for many AI clients, plus AI Guardrails that redact personal data and flag prompt injection.
n8n
- n8n Agents (September, preview): give an agent a goal, a model and tools, and it works out the steps. Reach it through Slack, run it on a schedule, or call it from a workflow.
- n8n Assistant (September): describe a workflow and it plans, builds, runs and debugs it on the canvas.
- Spend caps and Browser Use (October): monthly and per-session spend limits for agents, and browser automation for all users.
- Pricing by execution: one full run counts as one execution however many steps it has. The Business plan is self-hosted only, and the Community Edition stays free to self-host.
What it means for your automations
- Agents need guardrails. All three platforms now offer approvals, spend caps or human-in-the-loop steps. Use them from day one, especially for anything customer-facing.
- MCP changes how automations are used. Your scenarios, Zaps and workflows can become tools that your team’s AI assistant calls directly. Design them with clear inputs and outputs.
- Watch AI costs per step. Credits, model tiers and executions price AI differently. Compare them against your real volumes before you commit.
- AI builders speed up drafts, not design. Assistants that build from a description are great for a first version. Error handling, naming, documentation and testing still decide whether an automation lasts.
Not sure which platform fits your next automation? Read how we choose between Make, Zapier and n8n, or book a free discovery call.
