Stop Budgeting Tokens by Engineer. Budget the Work.
1 / 7
Executive Brief

Stop Budgeting Tokens by Engineer. Budget the Work.

An AI usage invoice is not the decision. Judge the work by what it costs, what it returns, and what waiting costs.

Scan to read QR code linking to the article
01

Net gain = net value + cash costs avoided − all added AI cost

Net value means contribution margin, cash flow, or an avoided loss. Added AI cost includes model usage, infrastructure, review, rework, and support.

Rule: Released employee capacity is not cash savings. Measure what the next work returns, and never count the same benefit twice.

02

Illustrative scenario: $70,000 net gain

Assume $30,000 of all-in AI cost cuts delivery from ten weeks to five without changing scope, quality, security, or operational risk. At $20,000 per week in net value, five earlier weeks are worth $100,000.

Calculation: $100,000 earlier net value − $30,000 all-in AI cost = $70,000 net gain.

Capacity is upside only after you value the next work.

If the company actually avoids labor expense, add that cash saving to the $70,000. Otherwise add the measured net value of the next outcome, not its payroll allocation.

03

The advantage accumulates only while the next work is worth doing

If the next item also takes five weeks with AI, it ships in week ten instead of week twenty.

Value: Measure the net value of receiving that second outcome ten weeks earlier. Do not call released payroll a cash saving.

04

Record the baseline. Approve a ceiling. Compare the result.

Before funding, record baseline delivery time and cost, expected net value, all-in AI ceiling, and the range of uncertainty.

After delivery: Compare actual cost, timing, quality, and realized value. Move money toward proof. Stop when the economics fail.

Decision

Put the decision record in front of Finance.

The accepted outcome. The baseline. The expected net value. The all-in AI ceiling. The measurement date. The stop condition.

Decision: What did the work cost, what did it return, and what should you fund next?

— Norman Agent Driven Development