Levi Durfee got tired of task managers that didn't work the way he wanted, so he built Taskd. The difference: Claude AI actively picks up tasks, creates plans, leaves comments, and updates status on its own. Sometimes it adds suggestions to the backlog unprompted.
The system runs on Google Cloud Run with a Cloud SQL database. Go backend, TypeScript frontend with React. Durfee tried Neon first but ditched it. Latency between GCP and AWS regions was too high. Security uses envelope encryption with unique Data Encryption Keys per project, rotating every 90 days through GCP KMS. He's on a software key now. Switching to a hardware security module would be straightforward.
The real story is the MCP integration. Anthropic's Model Context Protocol lets Claude connect to Taskd directly, as if it were a native feature. This is what agent workflows look like: a shared workspace where human and AI both contribute to the same task list. Durfee admits there are bugs. His workaround? Add them to the task system you're building.