Most of the AI tools I try don't last a month. I'll find something promising, hit the ceiling in an afternoon, and move on. From the user side, the disposability is exhausting. From the builder side, the same dynamic is thrilling. Both are true. And the tension between them tells you where the real value is migrating.
The essay maps five structural patterns I keep seeing underneath the churn: why most AI software is disposable (and why that's not purely a bad thing), why context is replacing judgment as the scarce resource, why the race shifted from attention to attachment, why screens are retreating, and why physical AI is where the defensible data lives.
Where I'm placing my bets:
- Interfaces that disappear. The screen should be where you go when you want to, not where you're forced to be.
- Systems that earn the right to remember you, and give you a real way to leave.
- One person plus agents as the default team.
- Verification as the hard problem: evals, audit trails, provenance.
- Physical presence as a product. Attendance is proof of commitment no algorithm can fake.
- Physical agents that actually work in the real world.
Most predictions about where this goes will be wrong. Mine included. But these aren't predictions. They're where I'm putting my time when I can't be certain.