Zapier Broke at 50 Employees. Here's What We Built Instead.

When no-code automation becomes the bottleneck: migrating 120 zaps to an owned workflow engine.

No-code automation is genuinely excellent until the number of automations exceeds the number of people who understand them. For this client that happened at around 120 zaps and 50 employees.

The failure modes were predictable in hindsight. No version control meant no way to review a change. No environments meant every edit was a production edit. And silent failures — a zap that stopped firing three weeks earlier — surfaced only when a customer complained.

We did not rewrite all 120. We audited them and found that 38 were dead, 44 were trivial notification forwards that stayed exactly where they were, and 38 carried real business logic. Only those 38 moved.

The replacement is deliberately boring: a queue, typed job handlers in the existing application repository, and a retry policy per job class. It gets code review, tests, and the same deployment pipeline as everything else.

The lesson is not that no-code tools are bad. It is that automation carrying business logic deserves the same engineering rigour as the business logic already in your codebase.