Engineering Journal · 2026-08
Why background jobs become operational systems
A few workers and a status column are not a platform. The moment business depends on them, you already have an operational system — usually an accidental one.
The pattern
It starts harmlessly. A feature needs an import. Someone adds a queue and a worker. Then exports. Then nightly reconciliation. Then an AI batch. Each piece is reasonable. Together they are a distributed system with no owner, no recovery model, and no shared language for 'what is running'.
What changed
The work left the request/response boundary. Latency, partial failure, and redelivery became normal. The product team still reasons in HTTP terms; operations reasons in incidents. The gap between those models is where outages hide.
The useful response
Name the execution plane. Give it a control surface, a definition of done for recovery, and a place for logs that operators can read without SSH. Whether that plane is Dagychu or another system is secondary. Treating background work as an accidental side effect of the web app is the failure mode.