Earlier pieces in this series looked at how AI is changing the engineer’s role, where it’s delivering value, and what it means to be ahead of the curve or behind it. There are two practical questions that need to be considered in tandem with the value of AI:
What does AI Actually Cost? And Where is that Cost Coming From?
The assumption is that AI gets cheaper over time, the way all technology eventually does. Broadly speaking, that is true. Model costs are falling, infrastructure is becoming more efficient, and access to increasingly powerful tools is becoming easier. In reality, that is not quite what organisations are experiencing.
A lot of businesses are finding their AI bills going up, not down, even as the underlying technology gets cheaper. The reason is simple: usage is growing faster than prices are falling.
A team introduces an AI coding assistant. It delivers value. More developers adopt it. Another team starts experimenting with AI agents. More workflows become automated. Within a short period of time, AI usage can spread across an organisation faster than anyone anticipated.
The businesses seeing the strongest returns are not necessarily the ones spending the most on AI tools. They are the ones that understand where the spend is going, where value is being created, and that treat AI investment with the same discipline as any other significant cost line.
The Cost of AI is More Than What is on the Invoice
When businesses talk about the cost of AI, the conversation usually starts with the obvious numbers: licence fees, API usage, model pricing and infrastructure costs. These costs matter, but they are only part of the picture.
The harder costs to measure are the ones that happen around the technology: the time spent experimenting, evaluating outputs, redesigning workflows, training teams and putting governance structures in place. For software engineering teams specifically, AI introduces a new category of work.
Someone needs to decide which tools teams should use. Someone needs to manage access, monitor usage, assess security implications and establish standards around what should and should not be automated. The companies getting value from AI are increasingly recognising that adoption shifts from a software purchase to an operating model change.
An AI assistant does not automatically make an engineer more productive. The benefit comes from knowing where AI performs well, where human judgement remains essential, and how engineering processes need to evolve around it.
What Happens when AI Usage Grows Faster than Expected?
The challenge many organisations are discovering is that AI costs do not behave like traditional software costs. A SaaS licence typically gives a business a predictable monthly cost per user. AI introduces a consumption model where costs increase based on activity – prompts, context size, model selection, agent usage and the complexity of workflows being automated.
As adoption spreads, small experiments can quickly become significant spending lines.
A developer creates an AI agent to automate one workflow. Another team sees the benefit and creates their own. Soon multiple teams are running AI workloads, often without a clear view of total consumption, ownership or the value being generated.
This challenge is already becoming visible across the industry.
TechCrunch: the rise of AI cost management
- Recent reporting from TechCrunch highlighted how organisations are beginning to reassess AI spending as usage scales. While the cost of individual AI interactions continues to fall, increased adoption of AI tools and more complex agent-based workflows are driving overall consumption higher.
- The conversation is shifting from simply adopting AI tools to understanding usage, setting controls and measuring whether the value created justifies the spend. The Token Bill Comes Due: Inside the Industry Scramble to Manage AI’s Runaway Costs – TechCrunch
McKinsey: moving from AI pilots to production
- Research from McKinsey has highlighted that many organisations underestimate the cost of moving AI from experimentation into production.
- The investment is not just in model access or licences. Scaling AI requires infrastructure, governance, integration work and operational processes to ensure solutions deliver measurable value. McKinsey QuantumBlack AI Insights
FinOps Foundation: the rise of AI cost management
- The FinOps community, which originally focused on managing cloud expenditure, is now expanding its thinking into AI consumption.
- The emergence of concepts such as “AI FinOps” and “token economics” reflects a broader shift in how organisations think about AI spend: not just controlling costs but understanding whether usage is creating enough value to justify the investment. FinOps Foundation – Token Economics: The Atomic Unit of AI Value
The Talent Cost of AI
There is another cost discussion happening alongside the technology conversation: what AI means for engineering teams themselves.
AI is changing how teams are structured. Experienced engineers equipped with the right tools can often deliver more, reducing some of the routine development work that previously helped create entry-level opportunities. This creates a longer-term question for the industry: If businesses reduce investment in early-career engineers today, where do tomorrow’s experienced engineers come from?
AI may reduce some short-term labour costs while creating a future challenge around developing technical capability. The organisations that benefit most from AI will be those that automate more tasks AND continue building strong engineering talent alongside those tools.
Takeaway
Bringing us back to our original question – what does AI actually cost? Well, the answer is more complicated than a licence fee or an API bill. AI costs money to run. It costs time to implement properly. It requires investment in governance, training and new ways of working. However, the biggest cost may come from using AI without a clear strategy.
The cost per token will almost certainly continue to fall. The technology will become more accessible. The tools will continue to improve. But cheaper AI does not automatically mean cheaper engineering. The real benefit will be tangible in organisations that treat AI like every other engineering investment: measured, governed and linked to real outcomes.
I guess this means that the question is no longer “how much does AI cost?”. It is “are we creating enough value from AI to justify what we are investing?”
Connect with Rachel at rachel.mcguckian@barden.ie or on LinkedIn>>>
