Three questions I ask every team before we test their AI agent
What's the worst thing it can do if it's wrong?
Who can stop it mid-action?
When did you last run the same input twice and compare the outputs?
Most teams answer the first one well. Almost nobody answers the third
@random_walker Worth adding: floods are unmeasured by design. The cost lands on someone outside the system, so it never appears in the sender's evals. A flood and a win look identical from inside.
@chipro The part I'd want tested is "calibrated." Type-safety kills parse errors, not wrong decisions, and a well-formed wrong action trips nothing downstream.
The malformed output was doing free error detection.
@aparnadhinak Coding and math RSI because the feedback is free and instant. The skills that actually matter in deployment have feedback locked inside customer systems nobody can publish. That asymmetry decides what gets better.
@marcklingen The WhatsApp angle is bigger outside the US than most takes account for. In India it isn't a messaging app, it's the default interface for everything, including customer support for many businesses. An agent UI there skips an entire step in the adoption process.
@vfsglobalcare@USAndIndia@CGI_USFederal@CGI_USA As mentioned already both options (Mail & Phone) did not help. Kindly READ THE CONTENT AT LEAST before you reply. Same as how you close the tickets
@airindiain Flight AI938 (DXB-CCJ) on 2nd April is been cancelled. I have been trying to reschedule the flight as per the civil aviation rules and no one is replying even after contacting in [email protected].
@RNTata2000@DGCAIndia
@TataMotors I have been waiting for past 100+ days for my Nexon even though I was promised delivery within 2.5 months. The dealers are telling it is because there is no allotment from Tata. Why is it delayed so much?