AI has become part of the everyday workflow for many engineering teams. Developers are using it to create code, prototype features and complete routine development tasks more quickly than ever before. But businesses are measuring AI productivity at the wrong point in the software lifecycle. The value of AI is not realised when code is generated, but when a change reaches production safely and starts producing commercial results.
For many organisations, products aren’t reaching customers any sooner, release schedules remain slow, and critical fixes still take weeks or months.
The reason is that coding accounts for only a fraction of what it takes to get software into production. Everything else, including testing, governance, security, compliance and release management, remains largely unchanged in many organisations. This is the 80% problem: accelerating the coding stage achieves little if a feature then spends weeks moving through testing, approvals and release processes before it reaches a customer.
If anything, AI makes these underlying constraints more visible. This is AI as the Great Amplifier. It doesn't just increase developer productivity; it amplifies the strengths and weaknesses already present across the software delivery lifecycle.
The effect of this delivery friction is clear when you compare delivery times across organisations. While high-performing teams with highly automated pipelines can achieve lead times of 24 hours, median enterprise lead times remain around 30 to 45 days.
AI strategy needs a platform foundation
This changes how organisations need to think about AI ROI. Measures such as volume of code generated or time saved on individual development tasks can demonstrate productivity improvements, but they tell you very little about the value AI is creating for the wider business. If software still takes weeks to reach production, accelerating one stage of its journey only gets you so far.
The more useful question is how efficiently software moves through an organisation’s entire Route to Live: from initial idea through development, testing and governance and ultimately into customers’ hands.
That means looking at where work waits. How long does it take to test and validate a release? How quickly can changes be approved? How often can software be deployed safely into production? This is where platform engineering becomes critical. What began as a way to improve developer experience is increasingly providing the technical foundation that enterprises need to embed AI safely, consistently and at scale.
A well-designed internal platform gives engineering teams a consistent and governed Route to Live, bringing together automated testing, security controls, approved tooling and self-service environments. Rather than requiring every team to navigate fragmented processes independently, it creates repeatable pathways that allow software to move faster without weakening quality or control. For enterprises looking to scale AI, those same principles provide the infrastructure, guardrails and consistency needed to move from isolated experimentation to repeatable deployment.
This matters because the model is only one part of the AI ROI equation. The strategy may determine where and how an organisation wants to use AI, but the platform underneath it determines how effectively those ambitions can be put into production and scaled. Two organisations can give their developers access to the same AI capabilities and see very different results depending on the delivery systems around them. If one can test, govern and release software quickly while the other relies on manual hand-offs and lengthy approvals, faster code generation will only widen the difference between them.
At one large organisation, just 5% of a £2.76 million annual engineering investment was translating into software that reached customers. AI ROI ultimately depends on whether those productivity gains translate into software reaching production and creating value for the business. The question for leaders is not just which AI tools to buy, but whether the organisation has the delivery platform needed to turn coding speed into business value.
Fixing the Route to Live
Over the next 12 months, leaders should focus on what is stopping the technology they already have from delivering greater value.
The starting point is measurement. Leaders need an end-to-end view of how software moves through the Route to Live, rather than focusing on how much developer time AI saves. Lead time to production, deployment frequency and change failure rates can show whether software is genuinely moving faster, while measuring time spent waiting can expose where progress is being lost between stages.
Once that is visible, organisations can identify the constraint setting the pace for everyone else. If releases are repeatedly held up by manual approvals, automating more development work is unlikely to change the outcome. If testing is the constraint, the priority might instead be earlier automated testing and continuous validation. The point is to invest in the part of the Route to Live that is limiting overall performance, rather than making an already-fast stage faster.
From there, platform teams can create a consistent Route to Live, giving developers a simpler, self-service way to move software into production with testing, security and compliance controls built in. This means teams can move faster without having to work through the same manual checks for every release. Those investments should also be tied to a business outcome. That might mean reducing the cost of delivery, getting revenue-generating products to market sooner, improving resilience or reducing the risk of failed changes. Without that link, organisations risk improving productivity without seeing a meaningful return for the business.
This platform foundation will become even more important during the next stage of AI adoption. As agents begin interacting with delivery pipelines and enterprise systems, organisations will need clear controls around identity, access and observability. Without them, AI risks introducing another source of operational friction rather than removing it.
The next phase of AI adoption will be defined not by how much code organisations can generate, but by how reliably they can turn it into business change. The organisations that close the gap between creation and production will be the ones that convert AI productivity into measurable return.