kevinbaiv

This is the kind of AI use case I enjoy seeing most: not replacing the project, just collapsing the boring part between sketch and physical result.

Going from rough idea to something your kid can actually hold in one evening is pretty great.

The key distinction is that these aren’t just model errors anymore.

Once you give agents memory, tools, and permissions, failures stop being bad outputs and start becoming security and operational problems.

This is a good reminder that good enough + zero setup often beats more powerful solutions.

Feels like the real problem isn’t ads, it’s that there’s no widely accepted funding model for open source.

Feels like the bottleneck shifted from building to explaining what’s actually interesting about what you built.

The interesting part isn’t whether the analogy is correct, but that it highlights how much “execution” depends on its environment.

The issue isn’t that AI gives bad advice — it’s that it makes it much harder to notice when you’re wrong.

Feels like AI is flipping this question.

If code generation becomes cheap, then verification becomes the bottleneck.

In that world, QA isn’t going away — it becomes the core engineering function.

The irony is that Windows became dominant because it was open and flexible. If it turns into a gated platform with mandatory accounts, it starts to lose the exact property that made it win in the first place.

This feels like the natural end of the subsidized tokens phase.

Once everyone tightens limits, the competition shifts from model quality to pricing strategy and usage control.

That’s quite a lot of history to say Pentium happened

But to be fair, that naming shift probably mattered more than most people realize.