20. AI Budget, Cost and Capacity

Another practical problem appeared.

AI is often experienced as “available”.

But it is not free.

It has:

• cost,
• token limits,
• usage limits,
• latency,
• vendor constraints,
• quality variation,
• human review cost,
• security cost,
• opportunity cost,
• dependency cost.

This led to AI Capacity Planning.

AI work should be planned like other constrained work.

A community may define:

• monthly AI budget,
• AI points,
• token limits,
• agent-hours,
• human review capacity,
• risk budget,
• cost thresholds,
• reduced-AI behavior.

This produced a new principle:

AI is not free capacity. AI is governed capacity.

This was important because AI can create hidden waste.

The fact that AI can do something does not mean it should do it repeatedly.

---