In January 2026, n8n shipped version 2.0 with native LangChain integration and 70 AI nodes. In February, Make promoted AI Agents from a side experiment to first-class citizens inside the Scenario Builder. By mid-year, Zapier split its AI agent product into separate billing entirely. All three of the major workflow automation platforms now sell you AI agents. None of them charge for it the same way.
If you picked your automation tool in 2024 based on per-task pricing and connector count, the ground has shifted under you. The question is no longer "which tool connects my apps." It is "which tool lets an AI make decisions inside my workflows, and what happens to my bill when it does."
What "AI agent" actually means in these tools
Before getting into pricing, it is worth being precise about what changed. These are not chatbots embedded in your workflows. The agent layer means the tool can now reason about what to do next, not just execute a fixed sequence of steps.
In Zapier, an AI Agent can browse the web, read documents, and decide which of your connected apps to call based on what it finds. In Make, AI Agents live inside the visual builder as reusable components that can branch, loop, and call other scenarios dynamically. In n8n, the dedicated AI Agent node wraps LangChain primitives — tools, memory, output parsers — into the same canvas you use for regular automations, with full JavaScript available when the visual builder is not enough.
The practical difference: a traditional Zap fires when a form is submitted and always runs the same five steps. An AI agent can receive the form submission, read the content, decide whether it is a sales lead or a support request, draft a personalized response using your brand voice, route it to the right team, and log a summary in your CRM — adjusting its behavior based on what it learned from the last hundred submissions.
That is genuinely useful. It is also genuinely expensive if you pick the wrong pricing model.
The three pricing philosophies
Zapier: separate product, separate bill. Zapier Agents runs on activity-based pricing, completely decoupled from the task-based pricing that covers standard Zaps. The free tier gives you a handful of agent activities. The Pro tier runs $400 per year. If you were already paying for Zapier tasks and now want agents too, you are paying for two products. The upside is clarity — you always know what your agent costs because it is a line item on its own. The downside is that you are paying for the same integrations twice.
Make: included on every paid plan. Make AI Agents shipped as first-class components in February 2026, available on every paid tier. You build them inside the same visual Scenario Builder you already use, and they consume the same credits as everything else. The cost is proportional to how many operations the agent runs, which means a simple agent that checks a condition and calls one API is cheap, and an agent that loops through a thousand records and makes a decision on each one is not. Make's approach rewards efficient agent design.
n8n: self-hosted, no per-agent premium. n8n charges for its cloud offering by workflow executions, and the AI Agent node is just another node — it does not carry a surcharge. If you self-host, the software is free under the fair-use license, and you pay only for the LLM API calls your agents make. For teams with engineering capacity, this is the cheapest path by a wide margin. For teams without a developer on staff, the learning curve is the real cost.
The numbers that matter
Zapier's own survey, published in 2026, reports that 72 percent of enterprises now use AI agents for tasks like data management and customer support. Gartner predicts 40 percent of enterprise applications will integrate task-specific AI agents by the end of 2026, up from less than 5 percent in 2025. McKinsey found that 23 percent of organizations are actively scaling an agentic system in at least one business function, with another 39 percent experimenting.
These are not hypothetical future numbers. They describe a market where your competitors are already running agents inside their workflows. The question is not whether to adopt, it is how to adopt without your automation bill tripling overnight.
What to actually do
If you are a small team without a developer, start with Make. The agent features are included, the visual builder is accessible, and the credit-based pricing is predictable. Build one agent for your highest-volume manual process — lead routing, support triage, invoice matching — and measure the time it saves against the credits it consumes.
If you already use Zapier and want agents, budget for the separate product line. Do not assume your existing Zapier plan covers agent activity. It does not.
If you have even one developer on your team, evaluate n8n seriously. The self-hosted option means your agent costs are just LLM API calls, and the LangChain integration gives you access to the entire ecosystem of AI tools without waiting for a platform to add native support. The tradeoff is maintenance — you own the infrastructure.
The automation market spent five years competing on connector count. In 2026, it is competing on intelligence. That is a better competition for everyone, as long as you read the pricing page before you build.