Jiyuan Ludong CTO Han Kai at Yunqi Conference: From Multi-Model Routing to an Agent Feedback Loop

On September 22, at the enterprise agent summit of the 2026 Yunqi Conference in Hangzhou, Jiyuan Ludong co-founder and CTO Han Kai delivered a talk titled From Harness to RSI Flywheel.

As the theme reported by QbitAI indicates, the session was about moving from multi-model routing toward a feedback loop, and about exploring how agents can keep evolving. The focus, in other words, is not on the quality of a single model call, but on letting an agent distribute work across several models and feed the results of a run back in as the signal for the next improvement.

That direction speaks to the questions enterprise agent deployments keep running into: once a system is running in real business processes, how does it stay stable, how do you know which step actually improved, and how do you turn production feedback into reusable capability. Closing the loop between routing and feedback is essentially a way of asking whether an agent can be relied on over the long run.

The report only conveys the framing of the talk. It does not set out technical details of the approach, and it does not say whether the work has been productized or open sourced, so what can be confirmed is a technical session at the conference rather than the launch of a new product.

In the current enterprise agent context, the value of this kind of talk is in putting the engineering problems on the table: multi-model routing means trading off cost against latency, while a feedback loop involves evaluation, data return paths and permission boundaries, and neither works without observable infrastructure underneath.

What to watch next is whether Jiyuan Ludong turns the idea into a usable tool or service, and whether enterprise customers can verify the gains of continuous improvement in real deployments.