GPT-6 Astra got +80% on a Physics benchmark. It insisted the expert results were wrong and its calculations were right.
A team checked. And yes. GPT-6 Astra was right. New score: 98%
No one knows what "agents" means in their company.
"You need to implement agents." How?
"Automate that with agents." How?
"Agents" is like "we're training a model": a lie.
An agent is an LLM that can change company data in production, usefully: reply to a customer, issue a refund, change something in an app.
It's still a very technical thing that non-technical executives have a hard time with, because it makes them look dumb.
I haven't found an easy way to teach C-levels how to implement agents in their companies. I found a hard one: teach them in practice, on their computers, how to build simple agents. They hate it.
Imagine a C-level forced to do pivot tables in Excel for the first time. It makes them feel dumb.
You need to accept that, for now, you'll feel dumb. Then AI agents get easier. Most execs won't do it, though :(