Tools and vendor stack
2026 pricing snapshot for an operator's own build — verify at build time, these shift often
2 min read
| Category | Vendor examples | 2026 pricing |
|---|---|---|
| Orchestration | n8n | Starter €20/mo (2,500 executions) to Pro €50/mo (10,000 executions); self-hosted Community edition free, runs on a ~$5–20/mo VPS [Established, official pricing page] |
| Orchestration (alternative) | Make.com | Core roughly $9–12/mo, Pro roughly $16–21/mo, both at 10,000 credits/mo; AI-agent steps consume credits faster than classic steps [Established tier pricing / Directional on AI credit multiplier] |
| Orchestration (alternative) | Zapier | Professional $19.99/mo (annual) for 750 tasks, billed per action inside a Zap; standalone Agents product $33.33/mo for 1,500 activities, billed separately [Established, official pricing page] |
| LLM API | Anthropic Claude and comparable frontier APIs | Entry-tier roughly $1/$5 per million tokens in/out; mid-tier roughly $2–3/$10–15; top-tier roughly $5/$25; prompt caching and batch processing cut these further [Established, verify live — pricing moves] |
| Support point-solution | Intercom Fin | $0.99 per resolution [Established, vendor-disclosed] |
| Support point-solution (alternative) | Zendesk AI | Roughly $2.00 per automated resolution [Established, vendor-disclosed] |
| Vector database (if a build needs RAG over your own documents) | A managed Postgres/pgvector offering | Free to roughly $25/mo at typical small-business scale |
| Hosting (if self-hosting n8n, or hosting a custom build) | A standard VPS or app-hosting platform | $5–20/mo entry tier |
| Contractor build (one-time, per automation) | A freelance automation builder | $1,500–15,000+ depending on scope — see ROI and payback |
See The 2026 tooling landscape for the full comparison of how these platforms' billing units actually behave at volume, not just the headline monthly price.
What's different about this stack versus a client-facing build
Unlike AI Agency's stack, there's no client-facing infrastructure here — no MSA, no per-client workspace, no voice-AI or telephony line unless your own operations specifically need one. What replaces that overhead is an internal discipline this course treats as equally mandatory: access control on whatever API keys and integration credentials your automations use, and a defined owner for each one. Failure modes covers the sourced 2026 security data on exactly what goes wrong when that discipline is skipped.
Up next
KPIs and kill switches
The specific gates that tell you an automation is actually working — not the ones a vendor dashboard shows you
2 min