The wrong AI budget question: why capping employee tokens backfires

Companies are capping employee AI use because the bills arrived and nobody can guess what an afternoon of AI actually costs. On October 31, 2026, Fortune published a commentary by Joe Mohen, chairman of Neocortex Robotics, and Robert B. Cohen, chief executive of AI Economics, arguing that the question every manager is asking, how many tokens should we give each employee, is the wrong question. The right question, they write, is what a finished piece of work costs, before AI and after. The evidence they assemble, alongside Accenture's own guidance and a study of 12,000 developers, suggests per-seat caps punish precisely the people creating the most value.

The Accenture shift, and what the company actually says

The trigger for the debate was a Wall Street giant quietly changing how it pays for AI. According to an internal announcement obtained by 404 Media and described in the October 31 Fortune piece, on October 1, 2026 Accenture began centrally funding 29 dollars per person per month toward Claude, with anything above that charged to the project concerned. Internal messages described employees spending 500 to 1,000 dollars a month, and one employee whose access was revoked entirely. Accenture says reporting that it has put a firm limit on its people's access to AI tools is incorrect.

The company's own economics explain the sensitivity. Fixed-price contracts represent more than 65 percent of Accenture's bookings, so token costs land on its own margin. And Accenture's July 2026 guidance, cited by Fortune, already records the pressure: in some environments, requests to lift individual spending limits arrive by the hundreds each day. That same guidance urges against blanket cuts and recommends routing work to the cheapest model that does the job, noting that in nearly every deployment examined, including its own, fewer than 10 percent of users and workflows drive the vast majority of the bill.

What the numbers actually say about token costs

Two honest caveats travel with that table. The Jellyfish study is observational, so it shows association rather than cause. And the comparison "320 times more" covers an extreme range; the direction, not the ratio, is what matters.

The 60,000-dollar website: capability, not productivity

The Fortune authors compare notes from their own portfolio. One company needed a new website. The agency quote was 60,000 dollars. Instead, three people from the business team, none of whom had ever built a website, did it in three days with AI agents, for a token cost of a few hundred dollars. The result, they write, outperformed the agency proposal: it worked properly on Windows and Mac computers, iPhones, Android handsets and Chinese phones running HarmonyOS, none of which the team knew to ask for. The agents also produced high-resolution logos, made marketing material consistent for the first time, and surfaced customer testimonials and awards nobody knew existed.

A second anecdote is more unsettling. Preparing a board presentation for an acquisition, they asked AI only to reformat slides. It fact-checked them, found several material errors, then reached the slide with the offer price and suggested cutting it by 40 percent. They had not asked for its opinion. They checked the reasoning and lowered the bid.

The pattern connecting both stories is capability, not speed: work that used to require a specialist can increasingly be done by whoever understands the business problem. On hiring, the authors scrapped several entry-level marketing roles at around 60,000 dollars each and hired one person at a considerably higher salary, who now designs collateral and ships analytics software.

Why caps backfire

The arithmetic against per-seat caps has three legs. First, the visible bill is a fraction of the real cost: Cohen's four-industry modeling, covering bank fraud detection, pharmaceutical research, chemical discovery and chemical manufacturing, found the token bill was about a quarter of total deployment cost once retries, idle capacity, monitoring and review time are counted. Cut the bill and you cut only the quarter you can see. Second, a cap punishes your heaviest user, who in the Jellyfish data is likely your most productive one; the marginal token is worth less than it used to be, and still worth far more than it costs. Third, rules written in 2026 will still bind in 2028, when the facts have moved. The authors reach for a warning from another era: New York's heavy 2015 licensing regime for cryptocurrency firms, written before anyone understood the industry, drove exchanges away for good.

A better framework for small-business marketing teams

1. Route, do not ration. Send each task to the cheapest model that handles it, as Accenture's own guidance recommends, instead of capping seats.
2. Budget by outcome. Price projects, not people: a website, a campaign, a quarterly report.
3. Track cost per finished deliverable. A few hundred dollars of tokens replacing a 60,000-dollar quote is a number your CFO understands.
4. Review quarterly, and sunset rules fast. The universities that banned AI in 2024 now require it; Harvard Business School hands MBAs ChatGPT and Claude, and Pace University gives every student a Gemini subscription. Institutions flip faster than rules do.

If your marketing budget is still organized around seats instead of outcomes, that reframing is one of the highest-leverage conversations you can have this quarter. We help teams build it. Start at The Vivid Company.

Read next

• One line in robots.txt can now opt your site out of AI training, everywhere
• Samaya Sterling beat Claude on a finance benchmark at about one-eleventh the cost
• From rules to reasoning: the new marketing automation wave