From AI reasoning to human action
Making an agentic system's reasoning visible at the moment someone has to approve or override it.
Vyasa sits above the systems a company already runs and reasons its way to the decisions their ERP can't make.
The routine clears itself. Only real exceptions reach a person, and when one does, that person has to decide whether to trust Vyasa's recommendation.
What I work on.
An order arrives. Credit, policy, inventory and pricing all run at once, and most of the time Vyasa deals with it automatically. When it can't, it holds the order and sends it to a person.
That handover is my job. Vyasa knows what it checked, which rule it applied, and what it would have done either way. The interface used to show only the outcome, so the person approving it was agreeing to something without being told why.
The reasoning now sits in the same place as the decision: what was checked, what passed, what didn't, the policy behind the hold, and a way to overrule it, laid out so it helps the person deciding instead of overwhelming them. Four clicks down to one, shipped and in daily use.
You can see what the product does at vyasa-ai.com.
The full case study is being written.
This is the newest of the four projects and the only one still running, so I'm writing it up while the work is live. The three below are finished and go into more detail on how I work.
- Decision fatigue in AI agentsAI UX patterns · Adobe
- Designing for usage-based AI pricingEnterprise AI · Icertis
- Making invisible work visibleMasters thesis
If you'd like to talk about this one before it's written up, email me.