I’m building FloForge, a CRM/workflow automation application for small businesses. I’m currently working on the backend of the automation engine and would appreciate feedback from people who have built workflow systems, CRMs, or event-driven applications. The current system supports: Event-based triggers — lead created, deal won, status changed, etc. Conditions attached to triggers Actions such as creating tasks and recording activities Automation activation/deactivation Automation run history Linking actions to the record that triggered the automation For example: Lead created → automation triggers → follow-up task is created and linked to the lead → run is recorded I'm now focusing on making this reliable enough for real-world use. The areas I'm specifically concerned about are: Idempotency: preventing duplicate automation runs/actions when the same event is processed more than once Failure handling: properly recording failed actions and deciding whether subsequent actions should continue Observability: useful run history, statuses, timestamps, errors, and action-level results Record relationships: ensuring actions remain correctly associated with the record that triggered them Conditions: making sure conditions are evaluated consistently before actions execute Concurrency: preventing race conditions when multiple events arrive at roughly the same time Retries: handling temporary failures without accidentally performing an action twice Execution architecture: whether the current event → automation → action approach will scale reasonably I'm not looking for compliments on the UI. I'd rather have people point out what's wrong with the architecture or what I'm overlooking. If you've built an automation engine, CRM, workflow system, job queue, or similar event-driven application, what would you consider essential for a production-grade implementation? Repository: Json-error9/FloForge-Automations-website-v3.0-launch Screenshots of the current automation UI are attached for context.