TL;DR
These are not competing products. Zapier Agents is a separately metered add-on that puts judgement into a low-volume workflow. n8n is a source-available workflow engine you can self-host with no execution ceiling. Choosing between them is a hosting and volume decision before it is a feature decision. Best for: framing the evaluation correctly.
Zapier Agents bills in activities, not tasks. Free gives 400 activities per month with a 10-activity cap per run, Pro gives 1,500 with a 40-activity cap, Enterprise is custom. An activity is counted every time an agent uses a tool, searches the web or queries a knowledge source. Best for: teams that want an agent that cannot eat the production automation budget.
n8n Cloud publishes three self-serve tiers billed annually: Starter around €20 to €24 per month for 2,500 executions, Pro around €50 to €60 for 10,000, and Business at €667 for 40,000. Self-hosting removes the execution ceiling entirely. Best for: volume, data residency and anything air-gapped.
Zapier redefined its meter in mid-2026 so AI steps, Code, MCP and the SDK all draw from one shared task pool at variable per-action rates, with an MCP tool call counting as 2 tasks and a Lead Router lead counting as 5. Agents stayed on the separate activity quota. Best for: anyone whose Zapier bill changed without their workflows changing.
A useful architecture combines an agent for variable research with a workflow engine for defined routing and execution. Choose the boundary according to the job, rather than adding an agent to every step.
Contents
- What each product actually is
- The two meters, and why they are not comparable
- Triggers and connectors
- Agent state, approvals and debugging
- Hosting and data boundaries
- Side by side summary
- Additional tools to consider
- Which one fits which sales operation
- FAQ: Zapier Agents and n8n
What each product actually is
The comparison is frequently framed as two automation platforms competing for the same budget line. That framing produces bad decisions, because the two products sit at different layers.
Zapier Agents is a standalone paid add-on with its own tiers and its own meter, layered on top of the Zapier platform. Zapier positions Agents alongside Chatbots and MCP as an AI orchestration layer, and the September 2025 Copilot launch put plain-English Zap building on every plan. The thing you are buying is a bounded agent that can decide which of your connected tools to use, inside a workflow surface you already run.
n8n is a source-available workflow automation platform with an AI agent node, over 500 pre-built integrations, custom API connections for everything else, and the entire source code on GitHub. It can be deployed with Docker on-premises or air-gapped, or bought as managed cloud. The AI positioning is explicit and technical: multi-agent setups, retrieval-augmented generation systems, swapping cloud or offline models without rebuilding workflows, connecting to anything that speaks MCP, and inspecting every decision with humans in the loop. n8n closed a $180 million Series C in October 2025 at a reported valuation above $2 billion, which is context for how quickly the connector library and the agent tooling are moving.
The distinction that matters for sales operations: Zapier Agents gives you judgement inside somebody else's runtime. n8n gives you a runtime you control, into which you install judgement. Everything else in this comparison follows from that.
The two meters, and why they are not comparable
This is the section to read twice, because most published comparisons of these two products are pricing a state of the world that no longer exists.
Zapier: tasks and activities are separate pools
Zapier moved AI steps, Code by Zapier, MCP and the SDK onto the same task-based model as everything else. An AI step inside a normal Zap now consumes tasks like any other step, with nothing separate to buy. Zapier Agents, the standalone product, is a paid add-on with its own tiers and its own meter, and that meter is activities.
The published Agents tiers are Free at 400 activities per month, Pro at 1,500, and Enterprise custom. Testing counts toward quota on Free but not on Pro or Enterprise. The per-run cap is 10 activities on Free and 40 on paid, and Zapier's documentation states that at the cap the agent asks before continuing. Agent sharing is available only on Enterprise.
For Zapier, separate workflow task consumption from agent activity. Estimate both using the actual branching and retry behaviour of the proposed operations workflow.
The definition of an activity is the part that decides your bill: an activity is counted every time an agent uses a tool, searches the web or queries a knowledge source. It is not one activity per task you assign. A single agent run that checks a CRM, searches the web, reads a knowledge base and writes a record has consumed four activities, not one.
The structural advantage of the separate meter is real and worth naming, because it removes the most common objection to agents inside an automation platform: because activities are a separate pool from Zap tasks, an experiment with an agent cannot consume the task budget your production Zaps depend on.
The June 2026 meter redefinition changed everything else. Zaps, Tables, Forms and MCP were bundled into one plan with a task-volume slider running from 100 to 2 million, which repriced the Professional and Team tiers. Then AI steps, Code by Zapier, MCP and the SDK folded onto one shared task pool, and a task became any successful action that runs, at variable per-action rates: AI steps priced by model tier, Code costing more for extended runtime, an MCP tool call counting as 2 tasks and a Lead Router lead counting as 5. Plan prices and task allowances were unchanged, and pay-per-task overage now caps at three times the plan's task limit before runs are held.
For a sales operation, the Lead Router rate is the one to model. Five tasks per lead means a 4,000-lead month is 20,000 tasks on routing alone.
n8n: executions, with unlimited steps inside them
n8n Cloud meters workflow executions, and the crucial detail is that an execution carries unlimited steps. A 40-node workflow and a 3-node workflow both cost one execution. That is the opposite of the credit-per-module approach used by Make, where every module that executes uses at least one credit including conditional logic like filters and routers, and where some actions such as running code or certain AI features can use multiple credits.
Published n8n Cloud figures differ slightly by source and currency, so here is the range. One source gives Starter at €20 per month for 2,500 executions, Pro at €50 for 10,000 and Business at €667 for 40,000, all billed annually, with n8n advertising annual billing as saving 17% and quote-only pricing above 40,000 executions. Another gives Starter at €24 per month for 2,500 and Pro at €60 for 10,000. n8n's own pricing page confirms Business at €667 per month billed annually for 40,000 workflow executions with unlimited steps, and lists Pro as including 7 days of insights and up to 13,700 AI credits per month, admin roles, global variables, workflow history and execution search. Business adds 6 shared projects, SSO, SAML and LDAP, 30 days of insights, different environments, scaling options and Git-based version control. Enterprise is quote-only with unlimited shared projects, 200 or more concurrent executions, 365 days of insights, external secret store integration, log streaming and extended data retention.
The comparison you can actually make
You cannot convert activities to executions. What you can do is price your own workflow twice.
Take one real sales operations workflow. Count the tool calls, web searches and knowledge lookups an agent would make per run, multiply by monthly runs, and check it against 400 or 1,500 activities. Then count the same workflow as one n8n execution per run, multiply by monthly runs, and check it against 2,500, 10,000 or 40,000.
Triggers and connectors
Zapier's connector library is the largest in the category and that has not changed. For a sales operation whose stack is entirely mainstream SaaS, the probability that every tool has a maintained Zapier connector is close to one. The June 2026 bundling of Zaps, Tables, Forms and MCP into one plan also means the trigger surface, the storage layer and the form layer no longer arrive as separate purchases.
n8n publishes over 500 pre-built nodes for common apps plus custom API connections for everything else, and the practical difference is what happens at the edge. When a tool has no node, n8n's HTTP Request node plus its credential system gets you there in minutes, and the community node ecosystem fills gaps quickly. On Zapier, a missing connector means a webhook plus Code by Zapier, and Code now costs more tasks for extended runtime.
For sales operations specifically, three trigger patterns matter and both platforms handle them:
Form submission to routing. Zapier's Lead Router exists as a first-class object and costs 5 tasks per lead. n8n has no equivalent named product, so you build the routing logic in nodes, which is more work upfront and free of per-lead metering thereafter.
Scheduled research. Both support cron-style scheduling. The difference is what happens on failure at 3am, which is a debugging question rather than a trigger question.
Webhook from a signal provider. Both handle it. n8n's advantage is that you can inspect and replay the exact payload; Zapier's is that the connector probably already parses it.
Agent state, approvals and debugging
This is where the products differ most and where published comparisons say least.
Approvals. Zapier Agents has one approval mechanic and it is the per-run activity cap: at 10 activities on Free or 40 on paid, the agent asks before continuing. That is a budget guardrail rather than a business-logic guardrail. It stops runaway spend and it does not let you require sign-off before a specific action such as writing to a CRM or sending an external email. n8n's positioning is explicit about keeping humans in the loop, and because you build the graph, an approval gate is a node you place wherever the risk actually is.
For a sales operation, this asymmetry is the single strongest argument for n8n on anything customer-facing. An agent that can email a prospect needs an approval gate before the send, not a spending cap at forty tool calls.
Debugging. n8n's marketing claim is that you can inspect every decision, and the product backs it with per-node execution data, execution search, workflow history and 7 to 365 days of insights by tier. Zapier surfaces Zap history and Agents run logs. The practical difference emerges at the third layer of nesting: when an agent chose tool B instead of tool A and you need to know why, n8n shows you the node input and output, and Zapier shows you that the activity occurred.
State. Neither product gives you durable agent memory as a first-class primitive without wiring one. n8n supports retrieval-augmented generation systems and connecting any model, which means you attach a vector store or a database and own the memory layer. Zapier Agents queries knowledge sources, and each query is a billable activity, which makes chatty memory expensive by design.
An agent can research an account and propose a message while a workflow engine handles approved routing. Keep the qualification rule and evidence requirements explicit. This architecture is an editorial recommendation, not a claim about a measured adoption trend.
Hosting and data boundaries
n8n's answer here is the reason a material share of its adoption exists. Deploy on-premises, air-gapped if needed. Deploy with Docker. Access the entire source code on GitHub. Hosted version also available. For any sales operation handling data that cannot leave a jurisdiction or a network, this is not a preference, it is the requirement, and Zapier cannot meet it.
The corollary is operational cost. Self-hosted n8n is a service you run, with a database, a queue, upgrades and an on-call expectation. Third-party analysis is direct that self-hosting does not reliably win on raw dollars at small scale once you price the infrastructure, which contradicts the usual pitch.
Before buying another orchestration platform, describe the job as inputs, rules and outputs. A scheduled transformation against a small database may be enough when no judgement is required.
Side by side summary
| Tool | Layer | Best for | Entry price |
|---|---|---|---|
| Zapier Agents | Metered agent add-on on a hosted automation platform | Low-volume judgement inside an existing Zapier stack | Current scoped quote; confirm term and included capacity |
| Zapier platform | Hosted workflow automation, largest connector library | Mainstream SaaS stacks with no hosting appetite | Task-slider pricing from 100 to 2 million tasks |
| n8n Cloud | Managed source-available workflow engine with AI agent nodes | Teams wanting control without running servers | Starter €20 to €24 per month, 2,500 executions |
| n8n self-hosted | Self-run workflow engine, unlimited executions | Volume, data residency, air-gapped deployment | Current scoped quote; confirm term and included capacity |
Additional tools to consider
The orchestration layer has more credible options than this comparison implies, and three of them price in ways that change the maths.
Make uses credit-based pricing where every module that executes uses at least one credit, including conditional logic such as filters and routers, and some actions like running code or certain AI features can consume multiple credits.
Gumloop publishes a Free tier at 60,000 credits per year with 1 seat, 1 active trigger, 2 concurrent runs and 5 agent interactions, and Pro at $37 per month with 240,000 credits per year and unlimited seats.
Relay and Lindy both sit in the approval-first agent category, with Lindy marketed around always-on assistant workloads rather than scheduled pipelines.
Clay is an adjacent option for organising enrichment and account research in a table-based workflow. Compare its data-source and integration costs with the job you would otherwise build in an automation platform.
Clay’s orchestration layer does not make every returned record authoritative. Preserve identifiers and source provenance, deduplicate results and decide what constitutes a correct match before using the output.
Which one fits which sales operation
Choose n8n Cloud when you want per-node visibility, approval gates placed on business logic rather than on spend, Git-based version control of workflows, and an execution model where workflow complexity is free. Business at €667 per month is the tier where SSO, SAML, LDAP, environments and Git arrive, and that is the tier most RevOps teams of any size actually need.
Choose self-hosted n8n when execution volume is above 40,000 per month, when data cannot leave your infrastructure, or when you need to swap in offline models. Accept that you have taken on an operational service.
Choose neither when the job is adequately served by a scheduled script and a database. A more capable platform earns its cost when it removes meaningful maintenance or supports a workflow the simpler system cannot handle.
FAQ: Zapier Agents and n8n
Is Zapier Agents included in a normal Zapier plan?
No. Zapier Agents is a separately metered add-on with its own tiers, and its meter is activities rather than tasks. Zapier's documentation states that add-on billing is separate from the core plan, that agent activities and chatbot limits do not draw from the task pool, and that adding an add-on does not change the plan's task allowance. There is a free Agents tier at 400 activities per month. Note that any article citing Zapier Agents at $3 or $20 per month is describing a pricing state that no longer exists.
What exactly counts as an activity on Zapier Agents?
An activity is counted every time an agent uses a tool, searches the web or queries a knowledge source. It is explicitly not one activity per task you assign, which is the most common budgeting error. There is also a per-run cap of 10 activities on Free and 40 on paid tiers, and at the cap the agent asks before continuing. On the Free tier, testing counts toward the monthly quota; on Pro and Enterprise it does not.
Does n8n charge per step like Make does?
No, and this is the main structural difference between them. n8n Cloud meters workflow executions with unlimited steps inside each execution, so a 40-node workflow costs the same as a 3-node workflow. Make uses credit-based pricing where each executing module consumes at least one credit, including filters and routers that perform no external action, and where code execution and some AI features can consume several. For a complex sales operations workflow with heavy branching, that difference compounds quickly.
Is self-hosted n8n actually cheaper?
Compare hosting, execution volume and maintenance with the managed alternative. Self-hosting can reduce a subscription expense while adding operational work; the licence and workload also need to fit.
Which one handles human approval before an agent sends an external email?
n8n, because you place the gate where the risk is. n8n's stated design keeps humans in the loop, and since you build the node graph you can require an approval step immediately before any send or write action. Zapier Agents has one approval mechanic, the per-run activity cap at 10 or 40, and that is a spending guardrail rather than a business-logic one. If an agent in your workflow can contact a prospect or write to a CRM, that distinction should decide the choice.
Did Zapier's pricing change in 2026, and does it affect existing workflows?
Yes, twice, and yes. In June 2026 Zaps, Tables, Forms and MCP were bundled into one plan with a task-volume slider from 100 to 2 million tasks, repricing the Professional and Team tiers. Later that month AI steps, Code by Zapier, MCP and the SDK folded onto one shared task pool, and a task became any successful action that runs at variable per-action rates, with an MCP tool call counting as 2 tasks and a Lead Router lead counting as 5. Plan prices and task allowances were unchanged, and overage now caps at three times the plan's limit before runs are held. If your bill moved without your workflows changing, that is why, and Lead Router volume is the first thing to check.
The agent still needs an execution workflow. Our comparison of n8n vs zapier vs make covers the adjacent options.
Work with Forma Nôrden
We build signal based outbound systems for B2B companies selling into the enterprise and upper mid market. The orchestration choice looks like a feature comparison and behaves like a hosting and metering decision, which is why so many teams end up paying twice for the same automation. Explore how we work.
For enquiries about this article: partnerships@formanorden.com
