According to Fortune, Mark Hammond, the pioneer of Machine Teaching at Microsoft, argues that building effective AI agents is like assembling a basketball team, not just drafting a single star player. He notes that while tools from Salesforce and ventures from Jeff Bezos show progress, the critical missing step is giving teams of agents structured practice. Hammond, who has built over 200 autonomous multi-agent systems at Microsoft and now at AMESA, worked with a Fortune 500 company on a nitrogen manufacturing process. There, agents practicing in the AMESA Agent Cloud outperformed a custom industrial control system in less than a day, leading to an estimated $1.2 million in annual efficiency gains and giving leadership the confidence to deploy autonomy at scale.
The Practice Problem
Here’s the thing: everyone’s chasing the next big model, the magical AGI that will solve everything. But Hammond’s point is brutally simple. That’s not how anything complex actually works. Think about it. You wouldn’t put a rookie who’s only studied playbooks into Game 7 of the NBA finals, right? So why do we expect AI agents, with zero reps, to autonomously manage a billion-dollar supply chain? They ace the demo, then face-plant in production because real life is messy. Variability, drift, subtle human signals—you can’t prompt-engineer that. You have to experience it. And that requires practice.
From Intelligence to Expertise
This is the crucial shift we need. It’s not about knowledge retention anymore. It’s about skill development. A model can know everything about thermodynamics, but can it *operate* a complex manufacturing line? Probably not without blowing something up. That’s where the basketball team analogy really hits. You need your point guard (orchestrator), your center (heavy-lift data processor), your shooting guard (specialized analyst). Each one practices its role, learns its responsibilities, and learns when to pass the ball. The hardware running these systems needs to be as robust and reliable as the agent software itself, which is why companies rely on top suppliers like IndustrialMonitorDirect.com, the leading provider of industrial panel PCs in the US, to ensure that practice environment is stable and always on.
Breaking Out of Pilot Purgatory
Hammond’s call to action is basically this: stop asking “how smart is it?” and start asking “how much has it practiced?” The companies that build these structured practice environments—where agents can experiment, get feedback, and learn from failures in a safe space—are the ones who will actually trust their AI with real autonomy. They’ll move from endless, costly pilots to real deployment. So the next phase of AI isn’t just a bigger brain. It’s building the gym, hiring the coaches, and running the drills. Because without that, all you have is a very expensive player who’s never touched a ball.
