The Validation Surface: Why AI Software Development Speed Depends on Trust

Two kinds of work happen inside our studio, often in the same week. One person builds a useful internal tool in a few days, working through a chat interface. If it breaks, the impact is limited and the change can be easily reversed. At the same time, another team might spend weeks making a small change to a live product with millions of users and critical transactions running through it.

The code itself might not take long to write, but the work around it does: protecting payment flows, migrating data, coordinating across services, testing edge cases and proving the change is safe before it reaches production.

It's tempting to ask how much faster AI makes software development. But that question misses the bigger picture. The more useful question is: what type of work are you asking AI to accelerate?

The difference between these two examples isn't how quickly a first version can be created. It's the level of confidence needed before that change can be trusted in the real world. When generating code is cheap, what you're really paying for, in time and in budget, is the work of making a change safe to trust.

Where the time actually goes

‍Payment systems and early experiments have always been treated differently. What's changed is the balance between them. Producing a working version of something got dramatically cheaper. Being sure that version is correct, safe and fit to run did not.

When building was slow, the work of establishing that confidence used to hide inside the development cycle. Nobody noticed it as a separate cost because it was absorbed into how long everything took anyway. As building speeds up, that assurance work becomes the largest part of what's left.

We think of this as the validation surface: the evidence, testing and control needed before a change can be trusted in production. When generating code becomes faster and cheaper, understanding this surface becomes increasingly important.

Why AI speed is not the same everywhere

‍The validation surface changes depending on the risk involved.

A small internal experiment with limited users may need minimal validation. A change affecting customer data, financial transactions or a core workflow requires much more.

The surface grows when more users are affected, more systems are connected, behaviour is harder to predict, changes are more difficult to reverse or failures carry greater consequences.

This is why there is no single answer to "how much faster does AI make development?" The speed available to a project depends on the confidence needed before it can go live. AI can accelerate creation, but production still requires judgement, testing and engineering discipline.

Reducing risk without slowing down

‍The validation surface is not fixed. The right product, design and engineering decisions can reduce uncertainty and help teams move faster with confidence.

Feature flags, staged rollouts, monitoring, automated testing, controlled environments and clear permissions all help teams identify issues earlier, limit impact and recover quickly when something changes. They don't remove risk, but they change how it's managed.

This is why we increasingly think about risk before thinking about speed. The first question is not simply "how quickly can we build this?" It is "what can we do to make this safer to change?”

That thinking applies across the entire product team.

Product teams need to know which opportunities are worth pursuing and how quickly they can learn. Designers need to think about how users interact with increasingly intelligent products. Engineers need to ensure faster delivery doesn't cost reliability.

None of these are separate stages passed from one discipline to another. They are connected decisions about how teams create confidence while moving faster.

Building faster is only part of the opportunity

‍AI has made creating software faster. What has not changed is the need to know whether something is ready to be trusted.

The biggest opportunity is not just building faster. It is creating the conditions that allow teams to learn, validate and improve with confidence.

AI has shifted software development from a generation problem to a validation problem.

The real value is not just in what AI can generate. It is in creating something a team can confidently stand behind.