I am blown away by generative AI.
I have spent the last three years building with it, and I am convinced it can create massive value.
But I also see teams using it in ways that create more work, not less.
The Old Proxy for Productivity Is Broken
Before generative AI, knowledge work was often measured by output volume.
It was never a great metric, but it was visible: more slides, more pages, more documents.
That is why people joked that you could measure consulting value by the weight of the presentation.
Today, content is cheap.
Anyone can generate large volumes of text in minutes.
So volume is no longer evidence of progress.
The Slop Trap
A common pattern now looks like this:
- Take six half-formed ideas.
- Ask an LLM to turn them into a long document.
- Paste huge amounts of context into a model.
- Forward the output as if more words mean more thinking.
The result is often AI slop: verbose content with low signal.
It looks productive, but it pushes reading and interpretation costs onto everyone else.
New Standard: Signal per Word
In the AI age, the goal is not to produce more content.
The goal is to produce clearer thinking.
Use AI to:
- compress,
- clarify,
- structure,
- and surface the essential points.
If a document is longer after AI touched it, that should require justification.
Trust Requires Human Judgment
To use AI well, you need an independent point of view.
You still need to read the source material, reason about it, and form your own conclusions.
Then use the model to test and refine your thinking.
If the AI output does not match your reasoning, investigate why.
Do not outsource judgment.
My Mental Model
LLMs are the most eager-to-please, productive, and capable interns we have ever had.
But they are still interns.
They need direction, constraints, and review.
When supervised well, they accelerate high-quality work.
When unsupervised, they generate confident noise.
That is the shift:
In the AI age, value comes from better decisions, not bigger documents.
