Kestra’s ability to orchestrate self-healing data pipelines is exceptional. I heavily utilize native features like pluginDefaults and polymorphic triggers to build resilient architectures. It goes beyond simple task scheduling; it acts as an automated guardian for our application's data layers, seamlessly handling memory-cached data and integrating flawlessly with local AI agents.

The thing I keep coming back to is the AIAgent plugin with native MCP tool support. I built a Discord forum triage bot: new bug report comes in, Kestra fires, an agent runs semantic search across GitHub and Discord via Coral MCP, decides if it's a duplicate or new, and acts on it. All of that is one YAML file. Plugin coverage is solid. GitHub issue creation, a KV store for per-guild config, parallel HTTP calls to Discord's API, JSONStructuredExtraction for pulling structured fields out of AI output. Stuff I'd normally spend a day wiring up is just a few lines in the flow. There was a real race condition where two upvote reactions landing simultaneously would both try to file a GitHub issue. behavior: QUEUE, limit: 1 fixed it. Literally one line in the YAML. I wasn't expecting that to be so clean. Everything's in git, flow changes go through PRs, and the whole thing deploys with a single docker run on a server I already own. No extra cost, nothing new to maintain. For a side project running on existing infrastructure, that's kind of the whole point.

I use Kestra for automation, which really helps reduce heavy workloads and makes the work easy. The initial setup was also easy.