AI · Sep 2, 2026 · 7 min read
The New Economics of AI Leverage
The advantage is no longer access to models. It's knowing which decisions are worth automating and which are worth improving.
For most of the last decade, technology advantage came from access. The company with better tooling moved faster. That gap has closed. Capable models are a subscription away, which means the tooling itself is no longer the differentiator.
What replaces it is judgement: knowing which decisions in a business are worth improving, and which repetitive work is worth removing entirely.
Leverage has three inputs
Volume — how often does this task or decision happen? Cost of error — what breaks when the output is wrong? Verifiability — can a person confirm the answer quickly?
Work that scores well on all three is where AI produces immediate, defensible leverage: drafting, summarising, classifying, extracting, monitoring. Work that fails on verifiability tends to create hidden labour instead of removing it.
Price the decision, not the tool
A useful exercise: pick the five decisions your business makes most often, and write down what information is missing when each one is made. That list is your roadmap. It usually has very little to do with which model you chose.
The companies compounding fastest right now are not the ones with the largest AI budgets. They are the ones who identified their most repeated decisions and made those decisions better, one at a time.