News for hackers. Stuff that matters.
Re:Ask HN: What's your current AI agentic setup? by zmcddn on Tuesday November 10, 2026 @01:29PM (#50444232)
I use codex and claude pro 100 subs, and I do use deepseek and other models when needed. I mostly just use codex, claude code, opencode, and dsh for the harness.

However, I use a workflow orchestrator around them myself. Once I’ve worked out how I want a particular task handled, I want to save that procedure and run it again with different inputs.

For example, I have research workflows, which uses several agents running in parallel to do separate research, one agent to synthesize the results, and another agent to write a design doc from the research. Similarly, for debugging, I have one (or more) agent inspecting the code and identifying the issue, another agent designs the fix, and either the same agent fixing it (if it's a small fix) or another agent fixing it (if it's big), after that another agent review the fix, make sure the issue is fixed and not reproducible, and corresponding tests are updated/added.

So the thing is that all these workflows can be reused and composed into larger pipelines, which gives me more freedom than a simple structure. If you are interested, I use the opensource orchestrator Crewplane (crewplaneai/crewplane) for my workflows.

There are no comments on this comment yet.