The ceiling is your imagination

Fable is not as good as you thought. But maybe the problem is the user.

The pattern with AI seems to be better models, lower prices, but output that still comes out the same. Are the models hitting a wall? I think the real issue is your imagination.

Most of the basic "build this feature" work is already solved. Not much new there. If you still struggle with it, maybe your codebase is not set up well for agents, or it is a skill issue (the human kind, not the agentic one).

Right now the real ceiling on what AI is worth to you is not the prompting or the magic weights in the model. It is what you think to ask it to do.

Are you doing the same work with a bit more automation bolted on here and there? Has your product development flow actually changed? Are you ready for agentic work that merges thousands of file changes into your codebase, and are you still confident it will not be a disaster?

For most engineers and teams, the real shortage is imagination about where to point automation. I think that is why AI companies now even hire forward deployed engineers. Someone sits inside your domain and wires the existing tools into it so it moves the needle, because you could not figure it out yourself.

It will not always work. Big companies are hard to change, and imagination is not just knowing the tool. It is domain, history, and customers. But if you are doing the same thing you always did at 1.2x speed, how much are you leaving on the table? Your competitors all get the same small speed bump. The real question is how you build something from the ground up to work in this brave new world.

What can you ask AI for today that was impossible a year ago? And the harder question is whether we are ready for what we will be able to ask for in twelve months. Not just from the perspective of your infrastructure, but the people and their imagination to use it.