The 2026 tooling landscape

#The three orchestration platforms, as they actually price in 2026

[Established — official vendor pricing pages, verify at build time], pricing in this market moves often enough that these numbers should be re-checked directly before you commit spend.

#What the billing-unit difference actually means for your bill

The billing unit, not the headline price, is what determines your real cost at volume. n8n counts an entire multi-step workflow as one execution; Zapier counts every action inside a Zap as a separate task. A 10-step workflow run 10,000 times a month is 10,000 executions on n8n but up to 100,000 tasks on Zapier — at that volume, n8n can undercut Zapier by a wide margin, while at low volume (a handful of simple automations) the three platforms are roughly cost-comparable. [Directional] — the direction of this effect is a straightforward consequence of the disclosed billing models above; the specific magnitude will depend on your own workflow shape, so model it against your own step count before choosing a platform on price alone.

#The AI-agent layer, and what it costs on top

All three platforms now ship native AI-agent capability rather than requiring a separate tool, but none of it is free on top of the base subscription. Make's AI Agents consume credits per module call — a single AI-agent execution can burn roughly 43–50 credits, against a handful of credits for a classic deterministic step, so an AI-heavy scenario should be budgeted like a much larger workflow than its step count suggests. [Directional] — consistently reported across multiple 2026 Make pricing guides, not an official per-execution multiplier Make itself publishes. Zapier's standalone Agents product is billed entirely separately from your core plan's task allowance, on its own "activity" meter. n8n's AI nodes run on your existing execution-based billing, with the underlying LLM API cost metered separately by whichever model provider you connect. [Established for the billing-model split, Directional for the exact per-run costs]

#Point solutions, distinct from the general orchestration layer

Not every automatable function needs a general-purpose orchestration platform. For a well-defined, common enough function — customer support being the clearest 2026 example — a purpose-built point solution may be the better buy: Intercom's Fin prices at $0.99 per resolution, Zendesk's AI agent at roughly $2.00 per automated resolution. [Established — vendor-disclosed pricing models, verify at build time] What actually automates covers why "resolution" in these vendors' own dashboards means something narrower than it sounds, and Build vs buy vs DIY compares a point solution like this against building the equivalent yourself in n8n or Make.

#LLM API cost, the layer underneath all three

Whichever orchestration platform you pick, most AI-agent steps ultimately call an LLM API metered separately. Entry-tier models run roughly $1 per million input tokens / $5 per million output tokens as of this research; mid-tier roughly $2–3/$10–15; top-tier roughly $5/$25 — prompt caching typically reads at a fraction of base input cost, and batch processing typically halves both rates. [Established, verify at build time — provider pricing changes frequently] For a single narrow automation running at typical small-business volume, this cost is usually the smallest line item in the stack, well under the orchestration-platform subscription itself — see Tools and vendor stack for the full build-cost picture.