Welcome to the first issue of The Dispatch.
Every Friday: AI tools we actually tested, what changed, and what's worth paying for. No hype, no sponsored placements.
This week: n8n build methods, tested
We built the exact same lead-capture workflow (webhook → spreadsheet → email → WhatsApp template) three ways and measured everything:
Building with n8n-mcp: 3.4 seconds, validated before and after. The best primary route — with one catch: on a local instance you must set WEBHOOK_SECURITY_MODE=permissive first, or every management call gets blocked by SSRF protection. Also, the tested version can't activate workflows — that's one REST call or UI click.
Hand-writing JSON: 1.5 seconds to build, 5 runs to get right. Fast when you know the exact node JSON by heart. The traps we hit, one by one: Execute Command nodes excluded by default in n8n 2.x, expression prefixes, webhook bodies living under $json.body, downstream nodes seeing the previous node's output (not the webhook's) — with empty fields and no error.
Official template #5907: the slowest path. 7 structural changes, 0 of 6 functional nodes surviving unmodified. Templates are good for ideas; for a standard flow, direct building wins. And don't start from a template whose WhatsApp runs through an unofficial third party when Meta's official API is one HTTP node away.
Bonus: prompt caching cut our AI step's cost 78.9% on repeat calls (measured: $0.001378 vs $0.006546 per call with a ~5,490-token static prompt).
Full breakdown with all measurements: https://knowstackai.com/articles/n8n-build-methods-tested-mcp-manual-template-2026/
Site news
KnowStack AI now runs 34 in-depth guides, all with published test methodology.
Every article now shows a real "Updated" date when refreshed — no more stale pricing.
You're reading the first issue of this newsletter. If it was useful, forward it to one freelancer who'd like it.
See you Friday,
— Tomer
