Accenture's $29 Claude cap puts the AI bill against your name

Last week we covered Microsoft cutting its own Claude bill by more than a third, with monthly caps in one division reportedly falling from $100,000 per person to roughly $10,000. That was a story about a technology company managing a supplier. This week the cap has reached the individual consultant, and the number is a lot smaller.

On 15 October, Joseph Cox at 404 Media reported that Accenture has told many employees the firm will now spend only $29 a month per person on Claude. Some of those people had been spending as much as $1,000 a month. According to the internal announcement 404 Media saw, as relayed by The Next Web, the change took effect on 1 October and anything above the $29 has to be charged to a Work Breakdown Structure code, which is Accenture's budget code for a specific project or team.

If you work in a consultancy, a contractor or a client-side project team, you already know what a WBS code is. It is the thing that decides whether a cost sits with the firm or with a job. Accenture has decided that AI usage above a token amount sits with the job, which means it sits with a fee, a client and a named person. This is separate from a central IT license and is becoming increasingly relevant to the built environment, regardless of whether your firm uses Claude.

According to The Next Web’s account of 404 Media’s reporting, an internal Accenture memo titled ‘Claude Enterprise Users: Important Updates to Licensing and Usage Costs’ outlines changes to how the firm allocates and pays for Claude usage.

The memo reportedly states that:

The changes have reportedly raised concerns about project costs and access to the tool. Internal messages reviewed by 404 Media indicate that:

One employee asked how to expedite a Premium account upgrade, saying the change “obviously has a significant impact on project costs”.

Another employee, who uses Claude Code extensively to build an AI accelerator for several clients, estimated their usual monthly usage at $500 to $1,000.

A third employee reported that their access had been revoked altogether.

An Accenture spokesperson told 404 Media that the firm’s approach is “to provide the right tool for the right work” and that it continues to invest in AI capabilities, skills and enablement.

The issue predates the October memo. In June, 404 Media published leaked audio in which Justice Kwak, Accenture’s agentic AI strategy lead, told colleagues that “it’s actually not our engineers that are driving the token consumption. It’s a lot of the non-engineers that are doing some of those behaviors”.

The October memo appears to formalise the firm’s response to a usage pattern it had been aware of for months: AI costs are increasingly being tied to individual projects, with licence access and usage becoming subject to tighter controls.

If a company cuts someone's AI budget from $1,000 a month to $29, they are unlikely to stop using it. They may simply switch to a personal account, pay for it themselves, and upload company documents to get the work done.

Deloitte put a number on how common that already is. In a 12 October Deloitte Insights article, Gregory Vert, Brad Kreit and Abha Kulkarni cite the firm's May 2026 TrustID workforce study: among workers using AI at work, 49% reported using unapproved AI solutions, and senior leaders were 32% more likely than staff to do so. The underlying report, based on roughly 16,000 workers surveyed between February and April 2026, notes that 49% is down from 52% the previous quarter, and that the gap widens in technology companies, where senior leaders are 70% more likely than staff to use shadow AI. The people most likely to go round the policy are the people who approve the policy.

The Deloitte authors ask the question Accenture's memo answers by default: 'If a worker uses a personally developed agent for work, who bears the cost—the organization, or the individual?' They also report, citing Business Insider, that Walmart limited token use to avoid duplicative requests and vibe-coded projects, and that 'other companies could follow next'. The governance side is not keeping pace. A separate Deloitte survey of 3,235 IT and business leaders across 24 countries found only 21% say their organisation has a mature governance model for agentic AI, whilst 74% expect at least moderate use of agents by 2027.

There is a specific risk in the free-tier route that most people have not priced in. On the same day as the Accenture story, Cox reported that OpenAI lets advertisers inject ads into ChatGPT users' sessions, targeting groups of people by email address, phone number or a unique phone identifier, and that the targeting is also based on the prompts users enter. A consultant who pastes a client's cost plan into a free consumer chatbot because the firm's $29 ran out on the 9th of the month is now a target segment. That is a data protection conversation your client did not agree to have.

The cap only makes sense if token spend is a meaningful measure of anything. Stanford research published on 7 October suggests it is not, in either direction.

Arvind Karunakaran, an assistant professor in Stanford's School of Engineering and a faculty affiliate at the Stanford Institute for Human-Centered AI, studied AI deployments at a law firm, an advertising agency and an IT services firm with colleagues from the University of Virginia and Dartmouth. In the law firm, two near-identical divisions were given the same tool.

The division whose manager framed it as a way to enrich jobs, backed by training and set-aside time, used it 58% more and experimented with it 70% more than the division told it was a productivity tool.

The article also addresses firms that track token consumption and treat heavy use as a proxy for value. Stanford's write-up says that strategy encourages workers to use AI for its own sake, 'generating cat memes to hit a quota', as Karunakaran puts it. The inverse holds too. Capping tokens at a flat $29 says nothing about whether the work behind them was a cost plan check or a PDF converted to slides, which is exactly the behaviour Kwak described in June.

For a project director, that is the useful frame. A per-head allowance is a procurement decision dressed up as a people decision. Karunakaran adds that without resources behind the stated purpose, 'employees recognize it as little more than cheap talk'. A memo that says access is not guaranteed, with no account of which work the firm actually wants done with the tool, is the cheap-talk version.

Our industry runs on WBS codes, fee budgets and recoverable costs, so Accenture's model will translate directly. Expect the next framework tender or professional services appointment to ask whether AI tooling is included in the fee or recharged as a disbursement. Expect a quantity surveyor on a fixed-fee commission to be asked why their Claude usage was $400 in a month when the allowance was $29, and to have no good answer because nobody agreed what the tool was for.

The alternative is worse. If the consultant's answer to the cap is a personal account, the client's drawings, cost data and commercial correspondence are leaving the firm's tenant with no audit trail and, on a free tier, possibly feeding an advertising profile. The firm saves $371 a month and takes on an uninsured data risk.

Deloitte's finding that workers using only unapproved AI report much lower trust in their employer's AI tools is the other half of this: once people go round the system, they stop believing in the system.

Accenture has made AI usage a project cost with a person's name on it. That is defensible as accounting and weak as management, because it measures tokens rather than work and it pushes the overflow somewhere the firm cannot see. If you lead a team or run a commission, the move is to get ahead of the cap rather than wait for it.

Project Flux will keep tracking how per-head AI caps, project recharging and shadow accounts play out in consultancies and contractors, and what the fee letters start to say about it. Subscribe at projectflux.ai to see it before your next appointment is drafted.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.