AI economics tool · 2 minutes

The Attention Bottleneck

AI drove the cost of a first draft to near zero. It didn't make organisations infinitely productive — it just moved the bottleneck to whoever has to read, verify, and approve the result. Find out how much of your AI output actually gets used, and what's really capping it.

01 — Why the bottleneck moved

Four shifts that rewrote the economics of work.

The constraint didn't move by accident. It's the inevitable result of four changes in how value gets generated. Tap a card for the longer version.

02 — Calculate

How much AI output can you actually use?

Machine output scales exponentially with compute. Human review doesn't scale at all. Enter rough daily numbers — drafts, PRs, analyses, agent actions, anything an AI hands you to check.

AI-generated outputs per day

140200

Outputs you can genuinely review, verify, and approve per day

1850

Other forces at play

What else is slowing deployment down? Check any that apply.

Someone has to take the liabilityLegal, financial, or reputational risk means a human must own the outcome before it ships — regardless of review capacity. Compute or budget is the real capGPU access, cloud spend, or energy limits mean you can't even generate as much as you'd like. Your team mostly grades now, rarely createsMost attention goes to evaluating AI output, with little time left to produce original work or data.

03 — Counter-arguments

Attention isn't the only candidate for the ceiling.

Human attention is the most immediate, felt constraint right now — but several other systemic limits could end up defining the real ceiling on AI-driven output.

04 — Strategic implications

Whichever constraint wins, the response looks similar.

Regardless of whether human attention turns out to be a permanent limit or a transitional one, it's the defining operational constraint of this decade.