The line that stuck with me is that the harness matters as much as the model. The Opus example shows why. Every model release forced a retune and a full rerun of evals before it could ship.
That is a recurring cost, and the piece never prices it. Multiply it across every agent an enterprise runs and the harness turns into a standing budget line. Nobody has priced that yet.
One more question. This comes from a Microsoft VP describing Microsoft's own platform. What happens to the harness when a customer has 40 agents built by 12 teams on 3 frameworks?
سلام
✌️
The line that stuck with me is that the harness matters as much as the model. The Opus example shows why. Every model release forced a retune and a full rerun of evals before it could ship.
That is a recurring cost, and the piece never prices it. Multiply it across every agent an enterprise runs and the harness turns into a standing budget line. Nobody has priced that yet.
One more question. This comes from a Microsoft VP describing Microsoft's own platform. What happens to the harness when a customer has 40 agents built by 12 teams on 3 frameworks?
Happy to connect, let's talk more about it.
A worbook to monitor agent at scale in Sentinel
https://youtu.be/GYHOH9GUcDc
Detections of AI agents https://youtu.be/GYHOH9GUcDc
Greetings this is irrefutably a sublime instigate crystal with unperturbed apparently it's obligatory to perpetuate a perpetual presumption