Skip to content
, ,

Stop Budgeting Tokens by Engineer. Budget the Work.

Per-engineer token caps optimize the most visible cost while hiding labor, delay, and business value. Budget AI production by accepted outcome, not individual contributor.

·

Executive Deck

Let your agent read this

Executive briefClick to expand

Suppose a principal engineer spends $1,200 on AI usage and closes a revenue-blocking defect in one day instead of two weeks. Finance sees the $1,200. The thirteen days recovered are harder to see.

Now suppose another engineer spends $180 and delivers nothing the business accepts. A per-engineer token cap makes the second engineer look cheaper because it measures the invoice instead of the outcome.

Budget the work, not the person.

Use one decision rule

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

Net value means contribution margin, cash flow, or an avoided loss, not automatically gross revenue. Added AI cost includes the full incremental cost of model usage, infrastructure, review, rework, and support.

Released employee capacity is not cash savings. Count the value of the next work only after measuring it, and never count the same benefit twice.

Start with the defensible case

Consider one illustrative scenario. Four engineers need ten weeks to deliver a roadmap item. At a fully loaded cost of $3,750 per engineer per week, the labor assigned to the item is $150,000.

Suppose $30,000 of all-in added AI cost cuts delivery to five weeks without changing scope, quality, security, or operational risk. Labor assigned to the item falls to $75,000, but company payroll may not fall.

Assume shipping five weeks earlier produces $20,000 per week in net value. Earlier delivery is worth $100,000.

> $100,000 earlier net value − $30,000 added AI cost = $70,000 net gain

If the company also avoids $75,000 of actual labor expense, the gain becomes $145,000. If the same employees move to another priority, add the measured net value of that work instead. Do not treat $75,000 of payroll as the value of the capacity.

If the next item also takes five weeks with AI, it ships in week ten instead of week twenty. The additional benefit is the net value of receiving that outcome ten weeks earlier. The advantage accumulates only while the next work is worth doing.

Measure, then move the money

Before funding an item, give Finance the accepted outcome, baseline cost and time, all-in AI ceiling, expected net value, and measurement date. Use a range when value is uncertain, and agree on the point where spending stops.

Set a ceiling for the portfolio, then move money toward outcomes that prove their value. Track individual usage for security and misuse, not as a performance score.

The budget question is simple: what did the work cost, what did it return, and what should the company fund next?

Read the full executive package →

Pen doodle illustration for stop-budgeting-tokens-by-engineer-budget-the-work

3 min read

Suppose a principal engineer spends $1,200 on AI usage and closes a revenue-blocking defect in one day instead of two weeks. Finance sees the $1,200. The thirteen days recovered are harder to see.

Now suppose another engineer spends $180 and delivers nothing the business accepts. A per-engineer token cap makes the second engineer look cheaper because it measures the invoice instead of the outcome.

That is the mistake.

Budget the work, not the person.

Use one decision rule:

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

Net value means contribution margin, cash flow, or an avoided loss. It does not automatically mean gross revenue. Added AI cost means the full incremental cost of model usage, infrastructure, review, rework, and support.

Released employee capacity is not cash savings. Count what the next work returns only after you measure it, and do not count the same benefit twice.


Use one illustrative scenario. Replace these assumptions with your actual numbers.

Four engineers need ten weeks to deliver one roadmap item. At a fully loaded cost of $3,750 per engineer per week, the labor assigned to the item is $150,000.

Suppose $30,000 of all-in added AI cost cuts delivery from ten weeks to five without changing scope, quality, security, or operational risk. The labor assigned to the item falls to $75,000, but the company may still pay the same salaries.

Now assume shipping five weeks earlier produces $20,000 per week in net value. Earlier delivery is worth $100,000.

Apply the rule:

$100,000 earlier net value − $30,000 added AI cost = $70,000 net gain

That is the defensible base case.

If the company also avoids $75,000 of actual labor expense, the gain becomes $145,000. If the same employees move to the next priority, add the measured net value of that work instead. Do not use $75,000 of payroll as a substitute for its value.

Suppose the next item also takes five weeks with AI. It ships in week ten instead of week twenty. The additional benefit is the net value of receiving that second outcome ten weeks earlier. This is how the advantage accumulates across the portfolio. It only accumulates while the next work is worth doing.


Before funding a roadmap item, put the accepted outcome, baseline delivery cost and time, all-in AI ceiling, expected net value, and measurement date in front of Finance. Use a range when the value is uncertain. Agree on the point where you stop spending.

Some work will justify almost no AI cost. Other work may justify more AI cost than labor. There is no useful universal allowance per engineer because the value and risk belong to the work.

Set a ceiling for the portfolio, then move money toward outcomes that prove their value. Track individual usage for security and misuse, not as a performance score.

The budget question is not, “How many tokens did this engineer use?”

It is, “What did this work cost, what did it return, and what should we fund next?”

Companion

Written by

The views and opinions expressed in this article are the author’s own and do not represent the positions of any employer, client, or affiliated organization.

Every article, narrated. Listen while you ship.
From the Author

Corporate fiction

Three books. One operating problem. No clean hero.

Read 2028, Meridian, and AgentDrivenDevelopment.com’s Survive free online.

Read free online →

One useful note a week

Get one good email a week.

Short notes on AI-native software leadership. No launch sequence. No funnel theater.