In software development, AI has already become part of the everyday workflow for most teams. Studies estimate developer adoption rates at remarkably high levels, with some placing them close to or even above 90%. And this level of adoption would hardly be possible without a widespread perception that AI helps people work faster, even if the evidence of its actual impact remains highly inconsistent.
Today, tasks that once took hours, such as documenting, analysing information, generating code, or preparing test cases, can often be completed in minutes. Yet when we look at the software development lifecycle as a whole, one fundamental question remains unanswered: are we actually becoming more efficient?
Developing software faster does not necessarily mean developing it better. And it certainly does not mean developing it more cost-effectively.
Perhaps this is one of AI’s greatest invisible risks in software engineering: the perception of productivity can grow much faster than our ability to measure it.
And even if we can measure productivity, there is an even more strategic question we still need to answer: how does this acceleration affect the financial control of software development?
The challenge: knowing what your software really costs
When software development is managed using metrics that do not reflect the product being built, such as hours worked, velocity, or billing rates, organisations often face paradoxical situations. You’ve probably seen two teams delivering completely different outcomes while investing the same amount of effort, or two vendors charging dramatically different amounts for delivering equivalent functionality.
The arrival of AI has made these differences even more visible. As the apparent cost of producing software decreases and development speed increases, traditional metrics become even less useful for explaining what is really happening.
Measuring only visible activity, such as prompts submitted, tokens consumed, lines of code generated, user stories completed, or AI tool usage, provides useful signals, but it does not prove that productivity, quality, or cost efficiency have actually improved.
It is also important to remember that even if AI generates a significant proportion of the code, it does not mean the software being delivered is any smaller. The functionality remains exactly the same. What changes is the effort required to build it, the delivery time, or the associated cost.
And today, that cost includes far more than human effort alone. Organisations must also account for software licences, model consumption, tokens, infrastructure, AI agents, integrations, training, governance, and human oversight. Improvements in coding productivity can quickly lose part of their value if they simply shift effort towards reviewing or correcting AI-generated outputs, or if they introduce new technology costs that were never part of the original equation.
That is why organisations should not measure how much AI they use. They should measure whether AI enables them to deliver more valuable software, at a lower unit cost, with the same or better quality, and without compromising the overall software development lifecycle.
Software needs objective benchmarks
As AI adoption grows, organisations are being forced to answer questions that many believed could be postponed. Questions about the real productivity of their teams and vendors. Questions about AI’s impact on software quality. And, of course, questions about its financial impact on software projects. Wouldn’t you like to know whether the savings generated by AI are actually reaching your organisation or simply remaining with your vendor?
Answering these questions requires objective metrics, contextual analysis, benchmarking, and a consistent way to relate delivered functionality to effort, cost, and quality.
No matter how much AI changes the way software is built, it does not change the need to measure how much software is actually delivered. If anything, it makes it even more important to shift the focus away from activity and towards the product itself.
AI creates value only when that value can be demonstrated
Ultimately, the real question is not how much AI you use to develop software. It is how much business value that software ultimately delivers.
For more than two decades, LedaMC has helped leading organisations measure the productivity, quality, and efficiency of their software development initiatives. Today, we apply that same expertise to the era of Artificial Intelligence, objectively measuring its real impact, comparing it against market benchmarks, and turning that insight into better management decisions.
There is little doubt that AI will continue to make software development easier than ever. But the organisations that can answer how much their software costs, what quality it delivers, and how much value it creates will always stay one step ahead.
