Recently DeepSeek Harness got a lot of people in the AI circle excited, but I tried it once and didn’t keep going.
The reason:
DSH is still a preview, and the team says so explicitly — it can ship breaking changes at any time. Right now it’s a period of intense community churn and co-creation.
For someone like me who wants both novelty and stability: I want novelty, but I’m not that adventurous.
I want stability — I need an agent to reliably assist me every day as my second brain, so I can’t casually accept destructive changes.
I’ve noticed that a lot of people moving to DSH are actually coming from OpenClaw/Hermes, and for them that’s a big step up, of course.
But my OpenCode + Paseo + OpenViking setup runs so smoothly that I genuinely can’t think of a single hard requirement that would call for DSH.
I’d also like DSH to plug into Paseo, but the convenient option right now is ACP, and that doesn’t meet my bar.
If I were in a hurry, I could always rework opencode-agent.ts into a dsh-agent.ts, but the price is that in the short term I’d have to chase DSH’s frequent changes — and I can’t stand that.
Then Qwen3.8 27B took off in capability, but I didn’t deploy it locally.
The reason:
My local compute could run it, but after using local setups straight through last year and the year before, I’m burned out on it.
opencode go is still simple and blunt, generous with quota — the time I’d spend fiddling with a local deployment is better spent just using MaaS.
Running Qwen-embedding 0.6B locally in Ollama to do embeddings for OpenViking suits me just fine.
Reject FOMO. Any shiny new thing that doesn’t give me real, controllable, stable help is just noise.