AI Agents Let One Marketer Replace Several $60,000 Hires

Estimated reading time: 9 minutes · Last updated: 31 October 2026

Joe Mohen, chairman of Neocortex Robotics, and Robert B. Cohen, chief executive of AI Economics, say AI agents changed how they staff marketing. They had planned to hire several entry-level marketers at around $60,000 each. Instead they hired one non-technical person at a considerably higher salary. That person now designs collateral and builds analytics software that benchmarks customers against industry peers. Writing in Fortune on 31 October 2026, the pair argue that companies capping AI spending per employee are measuring the wrong thing. In their view, the number that matters is how much it costs to complete a unit of work, compared with and without AI. Their main example of a company getting this wrong is Accenture's new $29 monthly Claude allowance.

Why Mohen and Cohen Hired One Marketer Instead of Several

The plan was to bring on several junior marketing staff, each costing roughly $60,000. Mohen and Cohen hired a single person instead and paid that person considerably more. The hire has no technical background. Even so, this one person produces the company's collateral and has built analytics software that compares customers' operations with industry peers, work the authors say the planned group could not have done.

Their second example is a website. An agency quoted one of their portfolio companies $60,000 for a new site. Three people from the business team, none of whom had built a website before, did the job in three days with AI agents. The authors judge the result better than the agency's proposal. It worked on Windows and Mac, on iPhone and Android, and on Chinese phones running HarmonyOS, none of which the team had thought to request. The agents also produced high-resolution logos and made the marketing material consistent for the first time. They found customer testimonials and awards the company did not know it had. Tokens cost a few hundred dollars.

Both authors have a stake in AI adoption. Cohen's firm, AI Economics, advises companies on AI costs. Mohen chairs Neocortex Robotics, which uses physical AI and blockchain to automate warehouses and tankers. Neither the portfolio company nor the hire is named, so these accounts cannot be checked independently.

Is AI Changing What Staff Can Do, or Just How Fast?

A board presentation shows how far the tools went beyond their brief. The authors were preparing slides on a company they wanted to acquire and asked AI only to reformat them. It did that, and it also fact-checked the slides without being asked, turning up several material errors. When it reached the offer price, it called the figure too high and proposed a 40% cut. The authors checked its reasoning and lowered their bid.

Mohen and Cohen argue that the pattern is not speed. People who could not build a website built one. A marketer who does not write code is shipping a software product. They call this a change in capability rather than productivity: work that once needed a specialist is moving to whichever employee best grasps the business problem.

They accept that this costs jobs. The junior roles they skipped will not exist. A graphic designer who is a close friend of one of the authors was laid off with two colleagues after new owners decided AI could cover the graphics department's work. Their answer is that a company refusing the tools does not protect those roles. It loses them later and risks its own survival as well.

What Accenture's $29 Claude Allowance Reveals About AI Caps

In July, Accenture published guidance for companies worried about their AI bills. It advised against blanket usage cuts and against telling staff to use AI less. It recommended sending each task to the least expensive model able to handle it. Accenture also found that in nearly every deployment it examined, its own included, under 10% of users and workflows account for most of the spending. In some environments, it wrote, hundreds of requests a day arrive asking to raise individual spending limits.

Starting 1 October, Accenture has centrally funded $29 a month per person toward Claude, with anything beyond that charged to a project, as first reported by 404 Media, which obtained an internal announcement. According to staff messages, some employees had previously run up monthly bills of between $500 and $1,000. One said their access had been revoked entirely.

Accenture says it is incorrect to report that it has put a firm limit on staff access to AI tools. Mohen and Cohen accept that position but point to a real cost pressure. Fixed-price contracts make up more than 65% of Accenture's bookings, so token costs come out of its own margin. For an employee, they argue, the difference between a cap and a project recharge is slight.

How Much Does AI Actually Cost Per Employee?

AI is priced in tokens. A token is the billing unit for both what a user sends and what the system returns, and it is roughly three-quarters of a word. By the authors' figures, a million tokens equals about 750,000 words, or eight or nine novels, and costs between $2 and $8.

Thirty questions a day from a typical office worker add up to roughly 1.3 million tokens over a month, costing about $6. That, the authors note, is how software vendors sell $25 seats and still profit. Agents change the arithmetic. When agents read files, check numbers and retry failed steps, a single task can consume 50,000 to 500,000 tokens. Mohen and Cohen estimate about $12,000 a year for the most intensive user, matching the heaviest Accenture spenders described in the staff messages.

This year Jellyfish, which measures what software teams complete, examined 12,000 developers across 200 companies. For those who used AI least, each completed job came to 28 cents in tokens; for those who used it most, the bill per job reached $89, almost 320 times more. The authors liken that gap to Taylor Swift's Eras Tour against JoJo Siwa. The median developer spent about $52 a month and the 90th percentile $691. That is a gap of roughly $7,700 a year for twice the output. The study is observational, so it shows association, not cause.

Why the Visible Token Bill Is Only a Quarter of the Cost

One of the two authors modelled AI deployments in bank fraud detection, pharmaceutical research, chemical discovery and chemical manufacturing. The model counted repeated attempts, idle capacity, monitoring and the staff time spent checking output. In all four industries, the token bill came to about a quarter of the total. Cutting the bill trims only the visible quarter and leaves the larger share untouched.

Their alternative is to budget by outcome. They point to an Accenture example. A quarter-end deck that took three analysts 15 hours now takes one person two hours, at a token cost of a couple hundred dollars. A per-head budget shows the token line and hides the labour it replaced.

They also warn against early rule-making. When New York imposed heavy licensing on cryptocurrency firms in 2015, a wave of exchanges left and never returned. Approval workflows, restricted tool lists and per-person caps carry the same risk, they argue. A cap penalises the heaviest user, who is often the most productive, and a rule written in 2026 will still apply in 2028.

Universities have already reversed course. Harvard Business School gives MBA students ChatGPT and Claude. Darden encourages their use on assignments, and Pace University gives every student a Gemini subscription. Mohen and Cohen say fluency with foundation models is the most important skill they screen for in recent graduates.

The case for and against measuring AI by outcome

The bottom line

The Mohen and Cohen argument is strongest on measurement. A $6 monthly user is not the problem. A per-head budget hides the labour that AI replaces, such as a three-day website standing in for a $60,000 agency job. The evidence is thinner than the confidence. The companies in their examples are unnamed. The Jellyfish data shows association, not cause. The model finding that tokens are only a quarter of total cost comes from one of the authors, whose businesses depend on AI uptake. Their call to measure outcomes rather than consumption is still a sound test for any finance team. The cost they acknowledge, entry-level jobs that will not come back, deserves as much attention as the savings.

What to watch

• Watch whether the per-person AI caps companies adopt in 2026 are still in force in 2028, the year Mohen and Cohen say the facts will have moved past them.
• Watch for any change to Accenture's $29 monthly Claude allowance or its project-charging rule; no date has been set.
• Watch whether Jellyfish or other researchers publish causal evidence on whether heavier token use raises developer output; no date has been set.