Comment by adeebvaliulla
3 days ago
That makes a lot of sense, and I like that you’re being explicit about regret minimization rather than chasing local optima.
The Thompson Sampling + Wilson score combo is a pragmatic choice. In practice, most agent systems I see fail not because they lack metrics, but because they overreact to them. Noisy reward signals plus greedy selection is how teams end up whipsawing configs or freezing change altogether. Treating uncertainty as a first-class input instead of something to smooth away is the right move.
I also agree with your point on attribution. Perfect attribution is a trap. In real production environments, partial and imperfect outcome signals still dominate static configs if the system can reason probabilistically over time. This mirrors what we learned in reliability and delivery metrics years ago: trend dominance beats point accuracy.
One area I’d be curious about as this matures is organizational adoption rather than the math:
- How teams reason about defining outcomes without turning it into a governance bottleneck
- How you help users build intuition around uncertainty and regret so they trust the system when it routes “away” from what feels intuitively right
- Where humans still need to intervene, if anywhere, once the control plane is established
If this holds up across long-tail tasks and low-frequency failures, it feels like a real step toward agents that behave more like adaptive systems and less like fragile workflows with LLMs bolted on.
Appreciate the thoughtful reply.
These are excellent questions!
Outcome definition: Simpler is better. Teams that start with one binary signal like "did it work?" (call completed, meeting booked, etc.), get value immediately. Governance bottlenecks usually come from overthinking it upfront.
Building trust: When Kalibr routes away from what feels like the "right" model and it works, people are surprised. We capture and show outcome history so teams can see when a path started to degrade and when Kalibr shifted traffic. No LLM decision making means no black box around routing choices, it's all shown in your dashboard when you use Kalibr.
Human intervention: Defining new paths, adding goals, handling edge cases where signal is genuinely sparse. The goal isn't zero humans anywhere, it's getting them out of the reactive debugging loop so they can focus on strategic decisions instead of repeatedly patching failed agents.
Curious, have you built multi step agents and run into the challenge of repeated failures?