The Hours Compression Model

Hours index, 2030

Hours index, 2035

Hours deleted by 2035

14%

Your scenario Reference scenarios Deleted hours

Run it yourself

AI exposure of knowledge work Share of tasks where AI is actually applied 50%

Task-hour reduction Time saved on exposed tasks 25%

Demand offset Share of the gain absorbed by new demand 30%

Demand offset erodes after 2030 The agriculture case. Demand absorbs productivity gains for a while — until markets saturate and it stops. The curve stays flat for years, then bends.

hours(t) = 100 × ( 1 − exposure × reduction × ( 1 − offset(t) ) × ramp(t) )

How to read it

The demand offset dominates everything. At a 75% offset, even aggressive adoption barely dents the curve. Exposure and task-hour reduction multiply each other, so compression only turns dramatic when both run high at once — neither alone produces the agriculture-like case.

The optimistic argument lives in that one slider: cheaper output creates more output, so hours hold. The agricultural argument is that demand does not expand infinitely. Once the market has enough output, productivity gains stop creating proportional labor demand. That is where the curve bends — and the erosion toggle above is what it looks like when the offset itself is a temporary condition rather than a law.

The sequence to watch is not employment. It is hours. Output per worker rises first. Hiring slows second. Aggregate hours flatten third. Headcount is the last variable to move, not the first.

The likely sequence

2024–2026

Adoption rises. Hours stay mostly flat. Job postings weaken in exposed roles while AI-mentioned postings grow.

2026–2028

Fewer junior hires, fewer backfills. “Do more with the same team” becomes the default pressure.

2028–2031

Aggregate hours flatten in exposed sectors even as output grows. Productivity and hours visibly decouple.

2031–2035

Potential headcount decline — if AI agents become reliable production systems. The model’s most speculative segment.