On the new bottleneck nobody is automating.

Implementation used to be where startups ran out of time to be wrong. You had an idea, and then six weeks of engineering between you and any signal about whether the idea was worth having. Most teams spent the bulk of their energy on that stretch. The building, the breaking, the shipping.
That stretch is collapsing. A small team with the right tools now closes in days what used to take a quarter. The work hasn't disappeared, but the cost of doing it has dropped dramatically.
What's left is the work that was always harder: deciding what to build, knowing why it matters, and keeping a team pointed at the same thing for long enough to finish it.
That part has barely been automated at all.
For years, the slowness of building did something useful. It forced teams to think. You couldn't afford to build the wrong thing, so you argued, sketched, killed bad ideas before they got expensive. The friction was the discipline.
Take the friction away and the discipline goes with it.
When building was slow, sloppy decisions hid behind sprint timelines. The team was clearly doing things. Now, with implementation nearly free, the quality of the decisions is the only thing left to evaluate.
Most teams aren't ready for that.

The cost of changing your mind has dropped. The cost of not knowing your mind has gone up. You can pivot the build in an afternoon. You cannot pivot a confused team in an afternoon.
Strategy ages in days, not quarters. Some of the tools that made building cheap are themselves the clearest example. They won the implementation race and arrived in a world where implementation doesn't matter anymore. The deck written on Monday describes a product that has materially changed by Friday. Operating cadences haven't caught up.
The scarce skill isn't can you build it. It's can you hold a coherent thesis for longer than the build cycle. That's a different skill. The last decade hasn't selected for it.
Everyone is going to have great agents. The teams that win the next few years won't be the ones that built fastest. They'll be the ones that could still tell the difference between motion and progress when motion was free.
The most clarifying version of this problem isn't a startup that couldn't build. It's a product that built perfectly, shipped on time, and arrived in a world where the decision to build it was already wrong.
That's the expensive part. It's where the work is now.
What it actually looks like, choosing which problems to carry, building for the right people, removing what shouldn't be there, is the next thing I want to write about.