In this guide
TL;DR: Pick by what you’re billed for. Zapier charges per successful step (“tasks”) and connects 9,000+ apps. Make charges per module action (“credits”) and now has fallback AI connections. n8n charges per full workflow run (“executions”) and is the only one of the three you can self-host. n8n and Zapier both let you require human approval per agent tool.
Disclosure: The Make link on this page is an affiliate link: we may earn a commission if you sign up through it, at no extra cost to you. It doesn’t change our picks, and every option below has a free plan or free edition. We have no affiliate relationship with n8n or Zapier. See our affiliate disclosure. This is a sourced comparison: every price and feature comes from the vendors’ own pages, read on October 18, 2026. We haven’t built agents in these tools for this article. Where something isn’t published, we say so.
Quick verdict
- Pick Zapier if you want the widest app coverage (9,000+ apps, per its pricing page) and a familiar Zap editor, and you’re comfortable paying per step.
- Pick Make if you build visually, want an agent that retries on a backup AI provider when the main one fails, and your scenarios are short. Try Make (free plan available) (affiliate link)
- Pick n8n if your agents make many tool calls per run (n8n bills the whole run as one execution), or you need data on your own servers.
- Pick none of them if a simple scheduled script does the job. Free options: n8n Community Edition (self-hosted) or Make’s and Zapier’s free plans.
What changed in the last month
- Sep 25, n8n Agents (preview): a new Agents tab where you describe the job, pick a model, and attach tools, MCP servers and existing workflows. Approvals, per-tool credentials, sessions and versions are built in. “One turn with an agent is one execution,” and tool calls to workflows or sub-agents don’t count separately. Agents are on n8n Cloud for everyone on the latest stable version, on self-hosted with extra setup, and in preview on Enterprise. (n8n blog)
- Oct 6, Zapier Agents became “AI by Zapier”: agents now run as one AI step inside a Zap (details below). (Zapier blog)
- Oct 9, Make AI Agent update: fallback AI provider connections, where “Make retries the run once using the fallback connection,” plus prompt caching for supported Anthropic Claude and OpenAI models. Both are in the Make AI Agent app on all plans. (Make release note)
- Oct 14, n8n security update: 15 advisories (5 rated High), fixed in stable 2.43.5, beta 2.44.1 and v1 1.123.85. Cloud instances are patched automatically; self-hosted users should upgrade. (n8n security update)
Comparison table
| Feature | n8n | Make | Zapier |
|---|---|---|---|
| Billing unit | Execution: "a single run of your entire workflow," however many steps | Credit: "most actions consume 1 credit"; some Make AI Provider features use more. Routers and error handlers are free | Task: each successful step. Triggers, filters, paths, formatter and other built-ins are free |
| AI steps | One agent turn = 1 execution; Assistant credits are separate (1,600/mo on Starter) | Make's AI Provider on all plans; other AI providers on paid plans | 1, 3 or 5 tasks per AI step by model tier, plus 1, 3 or 5 per tool call on tiers that support tools |
| Entry paid price | Starter €20/mo billed annually ($20 in some regions) | Core from $12/mo billed annually ($16 monthly) at 10,000 credits/month; Pro $21, Teams $38 billed annually | Professional from $19.99/mo billed annually |
| Free option | Self-hosted Community Edition; Cloud trial with 1,000 executions, no card | Free plan, 1,000 credits/month | Free plan, 100 tasks/month, two-step Zaps |
| Agent features | Agents tab (preview) + AI Agent node; workflows, MCP servers and sub-agents as tools; knowledge files (Cloud) | Make AI Agent (New) app, in open beta; module, scenario and MCP tools; fallback connection; prompt caching | AI by Zapier step; tools picker; structured outputs; model tier choice |
| Human approval | Mark a tool sensitive → Approve/Reject before use | No per-tool approval in the AI Agent app; a separate Human in the Loop app exists on Enterprise only (closed beta) | "Require approval before running," per tool |
| Runaway protection | AI Agent node 'Max Iterations' setting (default 10); agent turns count against your execution quota | "Steps per agent call" limit; scenarios stop when credits run out (webhooks queue) | AI step pauses for approval at 75 tasks per run by default (raisable to 500); account usage stops at 3× your plan even with pay-per-task |
| Self-hosting | Yes (Community Edition; Business/Enterprise license keys) | No | No |
| Security notes | Self-hosted = you patch (Oct 14: 15 advisories). Cloud data stored in Frankfurt, Germany | Vendor-managed | Vendor-managed; SAML SSO from Team plan |
How we read the prices: Make’s prices appear only in its interactive pricing table; at the default 10,000 credits a month we read Core at $12, Pro at $21 and Teams at $38 a month billed annually ($16, $28 and $51 monthly) on make.com/pricing. n8n lists 2,500 executions on Starter and 10,000 on Pro (n8n.io/pricing).
Coming from Zapier Agents? What changed on Oct 6
Zapier says “every capability that made Zapier Agents powerful… now lives inside a single AI step in the Zap editor.” Its own mapping:
- Agent instructions → the prompt field on the AI by Zapier step
- Tools → the tools picker in that step
- Agent trigger → any Zap trigger; a chat/on-demand agent → a webhook trigger
- One agent calling another → Sub-Zap by Zapier as the trigger on the child Zap
- Human intervention → “Require approval before running,” now set per tool instead of all-or-nothing
- Run history → Zap History, showing tools called, model tier and tasks used
- Cost → tasks instead of “activities”
Two limits to know: some knowledge sources (HubSpot, Asana, Zendesk, Zoom, Jira) aren’t supported yet, and “a few capabilities are still rolling out.” Eligible agents show a migration banner.
One inconsistency: on October 18, Zapier’s pricing page still listed Zapier Agents as an add-on billed in activities (400 a month free). The Oct 6 post says agents now bill as tasks. Check your own account before budgeting.
A cost example you can redo yourself
Say one lead-triage agent runs 500 times a month and calls three tools per run:
- n8n: about 500 executions, assuming one agent turn per lead (tool calls to workflows don’t count separately).
- Zapier: 500 × (one AI step at 1, 3 or 5 tasks by model tier + each tool call at the same tier rate + one task per successful action step after it). Standard-tier models don’t support tool calls. Built-in filters and formatters are free.
- Make: 500 × (one credit per module action, more for some AI Provider features). An agent run’s internal credit use depends on the model and tools.
We haven’t run these numbers on live accounts. Plug in the current prices from each pricing page, and count runs that actually happen, not runs you plan for. If you’re weighing a move away from one of these, see our Zapier alternatives and Make alternatives guides for how each option bills.
Who should pick which
- Non-technical owner, a few hundred runs a month: start on Zapier’s or Make’s free plan and move up when you hit limits.
- Agent that calls many tools per run: n8n’s per-execution billing is easier to predict.
- Data must stay on your servers, or you’re on a tight budget with technical skills: self-hosted n8n. Budget time for upgrades (see Oct 14).
- Already deep in Zapier: migrate to AI by Zapier and set the per-run task limit (75 tasks by default, up to 500).
- You need resilience against an AI provider outage: Make’s fallback connection is the only built-in one of the three in the pages we checked.
Whichever you pick, an agent is only as good as what its tools return. Our copy-paste verification prompt tells it to check tool results before acting, with setup steps for n8n and Make.
FAQ
Is n8n free if I self-host? The Community Edition is free to download; you pay for the server and do the upgrades yourself.
What’s the difference between a task, a credit and an execution? Zapier task = one successful step. Make credit = one module action (most of the time). n8n execution = one full workflow run.
Do I need to rebuild my Zapier Agents? Zapier says to recreate each agent as an AI by Zapier step and shows a migration banner on eligible agents.
Which is safest to let near my email? n8n and Zapier offer per-tool approval. Make’s AI Agent app doesn’t; Make’s Human in the Loop app is Enterprise-only and in closed beta. Require approval on anything that sends or deletes.
Can I cap AI spending? Zapier pauses an AI step for approval when a run reaches its task limit (75 by default, raisable to 500), and stops all usage at 3× your plan’s task limit even with pay-per-task. Make lets you limit steps per agent call and stops when credits run out. n8n’s AI Agent node has a Max Iterations setting (default 10), and agent turns count against your execution quota. For model API keys, set caps at the provider too.
Sources & method
- n8n pricing
- n8n blog: Introducing n8n Agents (Sep 25, 2026)
- n8n docs: Tools Agent (AI Agent node)
- n8n security update, 14 October 2026
- Make pricing
- Make Help Center 2026 release notes index (lists the AI Agent release note as October 9, 2026)
- Make release note: AI Agent fallback connections and prompt caching
- Make AI Agent (New) app docs
- Make Human in the Loop (Enterprise) app docs
- Zapier pricing
- Zapier task usage rates
- Zapier Help: AI by Zapier model tier pricing
- Zapier Help: How pay-per-task billing works
- Zapier blog: Zapier Agents is now AI by Zapier (Oct 6, 2026)
Method. Last checked October 18, 2026. Sourced comparison from vendor pages only; no hands-on testing claimed. Affiliate links: Make only (marked). How we work.
