The Moment 'Possible' Becomes Usable
Technology transitions don't happen when something becomes possible. They happen when it becomes usable—when the gap between capability and accessibility closes.
There’s a pattern I keep noticing in technology adoption. The breakthrough moment isn’t when something becomes possible. It’s when it becomes usable.
Possible means it works in a lab, in a demo, with expert supervision. Usable means it works in the wild, for regular people, without handholding.
The gap between these two states is often measured in years. Sometimes decades. And the work that closes that gap is often invisible—infrastructure, reliability, integration, design.
Where We Are Now
AI agents are currently in the “possible” phase. They can do remarkable things when the context is right, the prompt is precise, and someone is watching. But they break in unexpected ways. They require expertise to deploy. They don’t recover gracefully from errors.
The question I keep asking: what would it take to move from possible to usable?
Not for every use case. Just for the ones where the value is highest and the failure modes are most forgiving. The edges where experimentation can happen safely.
What Usability Looks Like
Usable systems share a few characteristics:
They fail gracefully. When something goes wrong, the user isn’t stranded. There’s a clear path to recovery—or the system recovers on its own.
They reduce decision load. Instead of presenting options and asking for choices, they make sensible defaults and only escalate when necessary.
They maintain context. You don’t have to re-explain what you’re trying to do every time you interact with them.
They integrate with existing workflows. They don’t require wholesale replacement of how people already work.
These aren’t flashy features. They’re the boring infrastructure that separates demos from products.
The Work That Matters
Most attention goes to capability improvements. Faster models, larger context windows, better reasoning. These matter. But they’re not sufficient.
The work that closes the gap between possible and usable is different:
-
Error handling and graceful degradation
-
State management and persistence
-
Integration layers and APIs
-
Interface design and interaction patterns
-
Trust calibration and transparency
This is where the real leverage is. Not making AI more capable, but making it more reliable. Not expanding what’s possible, but making what’s already possible actually work.
That’s what I’m focused on now.