A blog post called The Real Cost of AI

The Real Cost of AI That No One Is Talking About

As Artificial Intelligence (AI) moves from drafting content to agents carrying out entire business processes, the cost of getting work done is also rising sharply and going largely undetected.

Your monthly AI subscription is now only part of the cost. With the latest agentic capabilities running in the background of all major AI tools, the total cost of ownership becomes more complex.

A subscription that supported an initial pilot may not cover the demands of daily operations because more capable workflows can consume extra tokens, require more storage, need greater human oversight and introduce costs the original business case never considered.

In August 2026, Gartner predicted that the cost of running AI models per agentic workflow would increase more than fivefold through 2028. [1]

This pressure is already affecting adoption, as McKinsey’s August 2026 survey found that around 20% of respondents said AI operating costs, including token costs, had constrained their organisation’s use of AI. [2]

On the flip side, when an employee understands the new agentic capabilities and redesigns workflows, they can cancel some traditional SaaS subscriptions and realise cost savings.

Understanding both sides of that calculation is critical before leaders decide where to adopt and scale responsibly.

“I’ve been delivering AI training since 2023, and the change in how we work with AI over recent weeks is substantial. We can now delegate work across connected systems, which changes the skills, controls and commercial decisions organisations need. Our approach to adopting and scaling AI must also keep pace.”  Emma Shepherd, Founder, The AI Advantage Academy

How Much Do AI Tokens Cost?

Generative AI processes information in small units called tokens. These represent pieces of text or other content. In usage-based services, charges depend partly on how much information the model processes and generates.

Agentic AI uses models and connected tools to work towards a goal across multiple steps. It might search documents, compare findings, update a system and check its output. Each stage can consume additional tokens and storage.

Although the unit price of AI tokens has fallen, the work we ask AI to do is getting more complex, which requires more tokens and ultimately negates any substantial savings. Even if AI models continue to lower their token prices, the cost to complete a task may increase.  [1]

Choosing a less powerful AI model only helps if it delivers the required quality, and a faster workflow helps only if it improves the overall process.

Token usage is also unpredictable, and amounts remain unclear, making budgeting a minefield. 

Does Prompting Impact the Cost of AI?

Even the way we prompt AI has changed. One short prompt can trigger a complex workflow, so a short prompt doesn’t necessarily produce a small cost.

For example, “Prepare a supplier review” could trigger contract retrieval, performance analysis, external searches, comparisons and report preparation. The system may check its conclusions or repeat a failed step before returning the result.

An agent can also delegate work to other agents, a process that OpenAI explicitly notes can increase token consumption. [4]

In June 2025, Anthropic reported in its own data that agents typically used around four times as many tokens as chat interactions, while systems using multiple agents consumed around 15 times as many. [5]

The implication is clear: match the effort to the value of the task. Using AI to undertake extensive research might justify its cost for a major contract. However, the same approach could be excessive for a routine internal update.

Why Is the Cost of AI Rising?

AI costs can even continue after it produces an answer because searchable knowledge bases need storage and connected tools can charge for each use. [6]

Organisations therefore need to budget for integration, maintenance, monitoring and people checking results, because these less visible costs contribute to the overall cost of agentic AI.

These costs depend on the deployment and contract. They won’t all appear as separate charges on every subscription, but leaders need to know where they sit in the overall budget.

EY’s June 2026 analysis reinforces this point, identifying infrastructure, governance, organisational change, and recovery from failures as part of the total cost of agentic AI, beyond the token charges themselves. [7]

What Makes a Good AI Use Case?

AI use cases identified during pilots should be reevaluated to see whether they’ve expanded from a simple assistant into an autonomous workflow, because some original opportunities could now be unaffordable.

In June 2025, Gartner predicted that by the end of 2027, more than 40% of agentic AI projects would be cancelled because of escalating costs, unclear business value or inadequate risk controls. [8]

Consider this illustrative example. A task saves £2 of staff time but costs £2.50 in AI usage and human checking. At 10,000 tasks a month, that creates a £5,000 monthly shortfall before setup costs. Better quality or faster service might still justify the investment, but the business must explicitly value those benefits.

However, there are cost savings to be made on traditional SaaS providers. McKinsey’s August 2026 report adds evidence of this shift: 32% of respondents said their organisations decided against buying at least one software product or feature because they could build the functionality internally using agentic coding tools. This finding concerns avoided purchases, rather than confirmed subscription cancellations or net savings. [2]

For example, retiring a service that costs £1,500 per month and replacing it with an automation that costs £500 a month in tokens would save £12,000 annually in recurring costs.

Time saved through AI becomes a financial benefit only when the organisation uses that capacity productively or reduces expenditure.

How To Budget for AI

Sustainable AI adoption starts with understanding how new Agentic AI capabilities are changing the foundation of how we work: processes, people, information, systems and decisions. That requires an AI integration plan that approaches the entire business architecture.

Now, use cases need to define:

  • The outcome: What measurable business results should improve?
  • The boundaries: What permissions does AI have and what sits outside its remit?
  • The economics: What is the acceptable cost per successful task at expected volumes?
  • The accountability: Who owns the process, approves exceptions and reviews performance?

However, governance has yet to catch up with these ambitions, as Boston Consulting Group’s September 2026 Applied AI Index found that 42% of companies expected to grant agents autonomy by 2030, while only 5% had the full set of critical controls in place at the time of the research. [9]

How Do We Scale AI Responsibly?

Working with Agentic AI requires a different approach to traditional AI. Understanding its new capabilities and the risks it presents are critical. A competent AI Champion or AI Leader who understands the new capabilities can transform a business, but putting the same technology in the hands of an inexperienced employee could cost the organisation its future and reputation.

This also strengthens the case for organisational capability development, as McKinsey’s March 2026 AI Trust Maturity Survey identified knowledge and training gaps as the leading barrier to implementing responsible AI, while organisations with explicit accountability achieved higher maturity scores. [10]

Scale AI Responsibly

The AI Advantage Academy’s Enterprise Pathway helps organisations adopt a structured approach to managing AI responsibly through governance, operational capability and assurance. [11]

As part of the pathway, Senior Leaders can start their journey with one of the following:

C-Suite Accelerator – £1,500 – excluding VAT

A half-day workshop to establish a shared understanding of AI opportunities within your organisation and leave with an agreed scope for AI adoption and governance.

Up to 10 participants. Delivered in person or virtually.

C-suite Focus Day – £2,500 – excluding VAT

A full day to agree the objectives for AI adoption, manage AI use cases and risks, assess capability gaps and agree an initial six-month roadmap.

Blog Sources

Sources checked on 1 October 2026. Monetary examples use pounds; illustrative calculations are not customer results.

  1. Gartner: AI inference costs per agentic workflow forecast to increase more than fivefold through 2028. 17 August 2026.
  2. McKinsey: The state of AI in 2026: On the road to ROI. 25 August 2026. Survey of 1,719 participants in 97 countries; fieldwork 4 May–8 June 2026. Supports the 20% cost-constraint and 32% avoided-purchase findings.
  3. Stanford: AI Index 2025 — State of AI in 10 Charts. 7 April 2025. Historical comparison of model-query prices at a specified benchmark performance level, November 2022–October 2024.
  4. OpenAI: Multi-agent documentation. Accessed 1 October 2026. Supports the point that adding subagents can increase token use.
  5. Anthropic: How we built our multi-agent research system. 13 June 2025. The fourfold and 15-fold token comparisons reflect Anthropic’s own data, not universal billing multipliers.
  6. OpenAI: Developer pricing. Accessed 1 October 2026. Cited for separate storage and tool-call charges; no sterling tariff inferred.
  7. EY: Unlocking agentic value: a new investment discipline for the agentic era. 1 June 2026. Analysis of total agent costs, including infrastructure, governance, organisational change and failure recovery.
  8. Gartner: Over 40% of agentic AI projects forecast to be cancelled by the end of 2027. 25 June 2025. A forecast covering costs, unclear value and inadequate risk controls, not a measured cancellation rate.
  9. Boston Consulting Group: The Formula for Agentic AI Value — Applied AI Index 2026. 30 September 2026. More than 1,300 senior executives across over 20 sectors. The 5% finding concerns the full set of critical controls assessed, not whether companies have any governance at all.
  10. McKinsey: State of AI trust in 2026: Shifting to the agentic era. 25 March 2026. Approximately 500 organisations; fieldwork December 2025–January 2026. The link between accountability and maturity is an association.
  11. The AI Advantage Academy: The Enterprise Pathway. Accessed 1 October 2026. Supports the pathway description. Workshop names, prices, VAT treatment and delivery details come from Emma Shepherd’s supplied final copy.

 

Scroll to Top