Companies curb AI agent spending as usage-based bills outpace budgets

On October 15, 404 Media reported that Accenture told many employees it 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, and are now, in the article's words, scrambling to keep the same level of usage. Joseph Cox's post about it went past two million impressions on X, which tells you how many people looked at their own bill and felt something. The article also points out the irony: Accenture has spent years telling clients to invest heavily in AI. Fine. Irony is free. Tokens aren't. I'm not going to argue about whether $29 is the right number for a consultancy. The useful question is the boring one: what does Claude Code actually cost per month, and how do you budget it so the cap is a decision instead of a reaction? What Claude Code costs in October 2026 Claude Code doesn't have its own price tag. It comes with Anthropic's paid plans, and it shares the usage pool with regular Claude chat. From Anthropic's pricing page, checked on October 16, 2026: Option Price What you're really buying Free $0 No Claude Code Pro $20/month, or $17/month billed annually Claude Code with usage limits Max From $100/month 5x or 20x Pro's usage per session, plus monthly API credits Team Per seat Admin, central billing, user and org spend controls Enterprise $20/seat/month + usage at API rates Everything in Team, billed annually Console (API) Per token No plan limits, no ceiling unless you set one Two details matter more than the headline prices. First, subscription limits reset on a rolling five-hour window, with weekly limits on top, and Anthropic doesn't publish a fixed message count. How far a plan goes depends on the model, the length of the context and what the agent is doing. So "Pro is $20" is true and also only half the answer. Second, the API has no ceiling by default. For reference, the same page lists Opus 5.5 at $4 input / $20 output per million tokens, Sonnet 5.5 at $2 / $10, and Haiku 5.5 at $0.10 / $0.50 for prompts up to 100K tokens ($0.50 / $2.50 above that). Claude Code's fast mode for Opus 5.5 is listed at $8 / $40. Those look like small numbers until an agent rereads your repo forty times in an afternoon. That second point is probably how somebody gets to $1,000. Why coding agents burn more than chat A chat costs what you type plus what you read. An agent costs what it reads to decide what to do, over and over, and you don't see most of it. I wrote about the same mechanics on the API side in Keeping Claude's token bill from ruining your month: LLMs are stateless, so every turn resends the whole conversation. A coding agent makes that worse in three ways: Context is big by default. Files, diffs, test output, tool definitions. The agent pulls them in because that's its job. Loops. Run the tests, read the failure, edit, run again. Each lap resends the growing context. Model choice is sticky. If the default is the most capable model, every trivial rename gets frontier pricing. None of that is a bug. It's the product working as designed, which is exactly why it's easy to stop noticing the meter. How to budget it: start from the work, not the price list A per-seat number like $29 is a budget built from the price list. It's simple to administer and it will be wrong for almost everyone: too much for people who open Claude twice a week, far too little for the person whose whole job is now reviewing what an agent wrote. A better order: Measure two normal weeks. Not a demo week. Normal. On Team and Enterprise there's usage analytics; on the API every response tells you the tokens it used. Group people by how they work, not by title. Occasional users, daily users, "the agent is my pair programmer" users. You'll usually find a long tail and a small heavy group. Give each group the cheapest option that doesn't interrupt them. Occasional users on Pro. Heavy users on Max or a premium seat, because a fixed monthly price is easier to defend than an open-ended API bill. Automated pipelines on the API, with a hard limit. Re-check monthly. Models and prices move. A budget you set once is a budget you'll get surprised by. If you're solo, the same logic applies with smaller numbers: if you keep hitting the five-hour limit on Pro, that's the signal to try Max for a month and compare, not to quietly add API credits on top and forget about them. How to measure it You can't cap what you don't measure, and "the invoice" is not a measurement, it's a post-mortem. For agents I build on the API, the cost comes straight off the response: input tokens, output tokens, cache reads and writes. Log those per call, tag them with the user or tenant and the task, and "who is expensive and why" stops being a guess and becomes a GROUP BY. That's the setup I described in the token bill post, and it works just as well for internal tools as for products. The useful metric isn't cost per user. It's cost per finished task: per merged PR, per fixed bug, per qualified lead. $1,000 a month is expensive if it buys a lot of rerun test loops. It's cheap if it replaces a week of someone's time. Accenture's cap treats both cases the same, which is the part I'd question if I worked there. How to cap it without killing the useful part Caps work best in layers, each set where the money is actually spent: Plan limits. Subscriptions already cap you; the five-hour window is a budget whether you like it or not. Admin spend controls. Team and Enterprise plans list user and organization-level spend controls. Use them per group, not one flat number for everyone. API hard limits. For anything on Console keys, set a workspace limit. An automation with an uncapped key is how a typo becomes an invoice. In-code limits for your own agents. A daily budget check before each call and an iteration cap on tool loops. In my WhatsApp bot, an iteration cap of five on the tool loop is what stops a runaway loop from turning into a runaway bill. I wrote that up in the Claude token post as "a kill switch". Building an agent and want the cost controls in from day one instead of after the first scary invoice? The cheap wins before anyone caps anything Most of the savings come from not paying for the same tokens twice: Prompt caching for the static part of the context. Cache reads are a small fraction of the input price on every model on Anthropic's list. Right model per task. Classification, renames and boilerplate don't need the top model. Route them to Haiku or Sonnet. Shorter context. Point the agent at the files that matter instead of letting it rediscover the repo every session. Batch what isn't urgent. Anthropic's Batch tier exists for exactly the work nobody's waiting on. If you're estimating the full cost of an agent you're building, not just the coding tool you use, I broke the build and run costs down in AI agent development cost. So, is $29 a month enough? For someone who uses Claude occasionally to draft or explain things, probably yes. Pro alone is $20. For someone running Claude Code as a daily coding agent, no, and the 404 Media piece suggests the people affected already know that. The lesson isn't "AI is too expensive" or "companies are cheap". It's that usage-based tools need a budget built from usage. Measure for two weeks, split people by how they actually work, cap in layers, and track cost per finished task. That's less dramatic than a memo, and it doesn't leave your best people scrambling for tokens on a Tuesday. FAQ How much does Claude Code cost per month? As of October 2026, Claude Code has no standalone price. It's included in Anthropic's paid plans: Pro at $20/month ($17/month billed annually), Max from $100/month (5x Pro usage) up to the 20x tier, plus Team and Enterprise seats. Enterprise is listed at $20 per seat per month plus usage at API rates. You can also run Claude Code with a Console account and pay per token at API prices, with no monthly ceiling unless you set one. Is a subscription or the API cheaper for Claude Code? For one person coding most days, a subscription is usually the predictable option: a fixed price with usage limits that reset on a rolling five-hour window plus weekly caps. The API is cheaper for light or bursty use and is the only option where the ceiling is entirely yours to set, which also means nobody sets it for you. Measure a couple of weeks of real usage before deciding. Why did Accenture cap Claude at $29 a month? According to 404 Media (October 15, 2026), Accenture told many employees it would now spend only $29 a month per person on Claude, after some had been spending as much as $1,000 a month. The article frames it as part of a wider trend of companies curbing AI tool spend once usage outgrew budgets. It doesn't publish an internal rationale beyond that. How do I stop an AI coding agent from blowing the budget? Set the cap where the money is spent, not in a spreadsheet: per-user or per-org spend limits in the Team/Enterprise admin, a hard limit on the API workspace, and in your own agents a per-day budget check and an iteration cap on tool loops. Then log cost per task so you know which work is worth the tokens. Originally published on joancomadran.dev. I'm Joan, an AI engineer in Barcelona — work with me.

Microsoft is spending the next month running webinars for business customers on controlling AI costs and proving that Copilot agents pay for themselves, and the invitation puts the problem in one sentence: “AI costs are easy to start and hard to forecast.” When the company selling the product says so, it is worth looking at your own AI line. For most businesses it started as one subscription on one card. It is now more likely to be several: ChatGPT Plus for two or three people, a Claude plan for whoever does the modelling, Copilot switched on inside Microsoft 365, and, since last month’s agent launches, at least one thing billed by use rather than by seat. The amounts are small. The total is rarely on a budget line, and the return is rarely measured. The surveys say this is normal. U.S. Bank polled 1,000 senior finance leaders in the United States and reported on 29 September that 51 per cent said spending on AI tools and platforms had exceeded budget over the past year, while 69 per cent said the investment had delivered benefits. A June survey of 260 finance executives by the cost-management firm CloudZero found that only 22 per cent could tie AI spend to business outcomes, and 60 per cent agreed with the statement that they were “spending more on AI than we can currently justify with measurable outcomes.” Both are vendor-run surveys, mostly of large US companies, so hold the exact figures loosely. The direction is not in doubt. Per seat: ChatGPT Plus US$20, Claude Pro US$20, Microsoft 365 Copilot US$30 per user amonth on an annual commitment. Per use: tokens, or Copilot Credits at one US cent each, 25,000 forUS$200 a month prepaid. Included: a starting price, not always the final price. (Branded graphic by PGH Consulting, LLC) Three ways you are now charged AI is priced in three shapes, and the confusion comes from mixing them up. The first is per seat. ChatGPT Plus is US$20 a month, Claude Pro is US$20, and Microsoft 365 Copilot is US$30 per user a month on an annual commitment, or US$21 on the small-business edition for companies under 300 users. A seat is predictable: you know the number in January. The risk is sprawl, not surprise. The second is per use, and this is where the forecasting breaks. Under the hood every AI model charges by the token, roughly three-quarters of a word, counted on what you send and on what comes back. Business products wrap that in friendlier units. Microsoft bills agents built in Copilot Studio in Copilot Credits at one US cent each, pay as you go or as a prepaid pack of 25,000 credits for US$200 a month, and one complicated question can cost a dozen credits. On Claude, the newest model is a separate purchase of prepaid usage credits on the US$20 plan, as Week 15 covered. Usage pricing is not a trick. It is a bill that depends on behaviour, and behaviour is the thing a budget cannot see. The third shape is “included”. Copilot agents, OpenAI’s Dots, Claude in Google Sheets: each arrived with the message that it comes with a plan you already pay for. Read the terms and “included” has edges. Microsoft’s licence covers an agent used by your own licensed staff, within fair-use limits; point the same agent at a customer-facing web page and it draws credits regardless of the licences your staff hold. The first Dot is included, but on the Pro tier or a Business Premium seat, not on Plus. Included is a starting price. Long chats resend the full context. Agents make many model calls. Expensive models get usedfor simple jobs. Nobody owns the line. And the surveys: 51 per cent over budget, 22 per cent can tiespend to outcomes, 60 per cent spending more than they can justify. (Branded graphic by PGH Consulting, LLC) Why the bill creeps Four things push the number up, and none of them is anyone deciding to spend more. The first is that the model forgets. Microsoft’s own explanation, from an August post by its Foundry team, is that models are stateless, so “the full context is sent with every request.” A long conversation re-sends its whole history with every follow-up, so a question in the tenth turn costs several times what it cost in the first. The second is agents. One instruction to an agent is not one model call. In Microsoft’s words, “A single user request can generate many model calls,” and each retry, lookup and step is metered. An agent that fails halfway and tries again has billed you twice for one outcome. The third is using the most expensive model for everything. Microsoft’s line is blunt: “Most requests do not need a frontier model.” Summarising an email or cleaning a date column does not need the model that tops the reasoning benchmarks, and the two-hour bake-off from Week 15 will tell you which cheaper model scores the same on your work. The fourth is that nobody owns the line. CloudZero’s survey found AI is driven outside finance in 74 per cent of companies but finance owns the bill in 60 per cent. Seats get added by whoever has the card, and the first person to see the total is whoever reconciles the statement. Keeping it predictable The fix is the same discipline you apply to telephone and fuel. Put every AI subscription, seat and usage account on one register with an owner, a monthly cost and what it is for. The permissions register from Week 16 is the natural place; add a cost column. Prefer a cap to an open meter. Where a vendor offers a prepaid pack and a pay-as-you-go option, the pack is the one with a ceiling. A pay-as-you-go agent facing the public has no ceiling unless you set one. Set the alert before you set the agent loose. Route by job, not by habit: the frontier model for the work that needs it, the cheaper one for everything else. Measure cost per finished job, not per message. A second Microsoft post, two weeks later, calls this “cost per completed outcome,” what the business actually paid across every turn and retry to get one usable result. A cheaper model that needs three attempts is not cheaper. Line 1: monthly cost equals seat price plus usage. Line 2: value created equals hours savedtimes hourly cost, plus work you would otherwise pay for. Measure one job, tally minutes saved, thenrenew or remove. (Branded graphic by PGH Consulting, LLC) The two-line test for payback Proving return on an AI tool needs two lines, written before you renew. Line one is what it cost: the seat price plus any usage, per month, from the register. Line two is what it saved or earned: the hours it removed, multiplied by what those hours cost you, plus anything it produced that you would otherwise have paid for. Line two is the hard one, and the only way to fill it is to count. Keep the ten-cell scorecard from the bake-off, add a column for minutes saved, and tally it for a month. A US$30 Copilot seat is US$360 a year; if the tally shows it saving an hour a month for someone whose time is worth more than US$30 an hour, it pays. If the tally shows nothing, the seat goes, however good the demo was. The sequence matters. Measure first on one job, then expand. Microsoft’s ROI session is pitched the same way: prove the return on the agents already in the licence before buying more. CloudZero found that three-quarters of the companies that cannot measure outcomes have held back investment, and 35 per cent have killed an initiative outright. Not measuring is the expensive option. What this does not settle The survey figures come from a bank and from vendors who sell cost-management software, and their samples are mostly large US companies. A business with eight staff will not see the same percentages; it will see the same pattern on a smaller bill. Prices here are the published US figures. ChatGPT bills local subscribers in Jamaican dollars at a localised price, and Microsoft licensing runs through resellers, so the number on your statement will differ, and several of these prices changed last month. Hours saved are self-reported unless you time them, and people are generous to tools they like. Have someone other than the enthusiast keep the tally. And a tool that passes the two-line test on cost can still fail on accuracy. The survey in Week 19 found finance teams spending about a quarter of the week checking AI output. Checking time belongs on line one. List every AI subscription, seat and usage account with an owner and cost. Set a spending alertor cap on each usage-priced tool. Write the two lines for one tool. Run one routine job on a cheapermodel. Put the AI total on the management accounts as its own line. (Branded graphic by PGH Consulting, LLC) What to try this week 1. List every AI subscription, seat and usage account the business pays for, with an owner and a monthly cost. Most owners are surprised by the total. 2. Open the billing page of each usage-priced tool and set a spending alert or cap. If the tool has no cap, note that on the register. 3. For one tool, write the two lines: monthly cost, and hours saved times the cost of those hours. If you cannot fill line two, start a tally this week. 4. Run one routine job on the cheaper model and compare the result. If it scores the same, change the default. 5. Put the AI total on next month’s management accounts as its own line. A number that appears every month gets managed. Survey figures are as published by U.S. Bank (29 September 2026), CloudZero (24 June 2026) and Datarails (6 October 2026) and are vendor-commissioned; prices are the published United States figures as at 8 October 2026 and may change. An AI bill you cannot forecast is still a bill you can measure. Peta-Gaye Hardy is the founder of PGH Consulting, LLC, where she helps finance and operations teams adopt AI in practical, low-risk ways. She writes the weekly AI in Finance & Business column and is based between Jamaica and the United States. Learn more at www.pghconsultinggroup.com. Follow on Instagram and YouTube @pghconsultinggroup, and connect on LinkedIn at linkedin.com/in/peta-gaye-hardy. Disclosures: This article is informational and does not constitute investment, tax, legal, or accounting advice. Readers should consult a qualified professional before acting. Prices, plan inclusions and billing mechanics are as published by OpenAI, Anthropic and Microsoft at the time of writing and are subject to change; readers should confirm the current position on the vendor’s own pages before subscribing. Survey findings are reported as published by their sponsors and have not been independently verified. AI tools can produce errors, and every figure, clause, or claim they produce should be verified against a source before being shared or acted upon. The author has no commercial relationship with OpenAI, Anthropic, Microsoft, U.S. Bank, CloudZero, Datarails or any product mentioned and was not compensated by them. All product names, logos, and trademarks are the property of their respective owners and are referenced for editorial purposes. The examples described are illustrative and do not depict any real business.