A Fortune commentary argues companies should judge AI spending by the work it produces, not impose blanket limits on token use. Its practical case is that typical office use may be cheap, while heavier use can cost more and still pay off if it replaces costly work.
Accenture AI Watch analysis
What happened
The commentary says a typical office worker using AI for 30 questions a day consumes about 1.3 million tokens a month, costing roughly $6. It contrasts that with agent-heavy tasks, which it says can use 50,000 to 500,000 tokens each. Its advice to businesses: compare the total cost of a finished job before and after AI, rather than treating each employee’s usage as the result to manage.
It offers examples, including a website that an agency had quoted at $60,000, which three non-technical staff reportedly built in three days for a few hundred dollars in tokens. The essay also cites a study of 12,000 developers at 200 companies, where the heaviest users spent far more per completed job than the lightest. The study is observational, so it does not establish that heavier AI use caused higher output.
Why it matters
The argument lands in a real budgeting dilemma: a per-person cap can make spending look tidy while obscuring whether the work itself got cheaper or better. The commentary says Accenture began centrally funding $29 a month per person towards Claude on 1 October, with additional costs charged to projects. Accenture says it is incorrect to describe this as a firm limit on employees’ access. Those are materially different descriptions, and the distinction matters to staff as well as finance teams.
The useful test is not whether AI use is high or low in isolation. It is whether the complete job, including checking and oversight, costs less or delivers something the organisation could not otherwise do.
Our read
The strongest point is also the least glamorous: measure the outcome, not the token meter. But the essay’s boldest examples are anecdotes, and the developer study it cites shows association rather than cause. Businesses should test the economics on real work before turning either a dramatic success story or a tidy-looking usage cap into company policy.
What to watch
- Whether Accenture clarifies how its Claude funding and project charges work in practice.
- Whether businesses publish comparisons of total job costs, including review and staff time.
- Whether larger studies find that heavier AI use reliably improves output, rather than merely accompanying it.
Discussion spark: Should employers cap AI spending per person, or judge the bill against the work produced, even when the heaviest users cost much more?
Sources and evidence
- Our AI agent experience: we scrapped $60,000 marketing jobs and hired one person for far more – Fortune (31 October 2026, 11:31 UTC)
not affiliated with or endorsed by Accenture AI