AI is accelerating application development, but control is now the limiting factor

AI-assisted application development, often referred to as "vibe coding", has reduced one of the longest-standing constraints in software: access to engineering time. Functional applications can now be created quickly, often by non-developers, shifting how ideas move from concept to execution.

The most immediate impact is in smaller, focused tools. These are not large enterprise systems, but targeted applications with limited scope. Increasingly, they are stable enough for operational use. Tasks that once required setup time and specialist input, such as configuring environments, connecting databases, and generating interfaces, can now be completed rapidly.

This change matters because it brings development closer to the problem. Domain experts can build and validate solutions directly, rather than waiting for prioritisation. In practice, this means ideas can be tested in days rather than months.

The more useful distinction now is not prototype versus production, but contained versus exposed. The rest of this argument turns on that distinction.

Applications operating within controlled environments carry relatively low risk. As they expand to integrate with other systems or handle sensitive data, the risk profile rises. At that point, testing, governance, and engineering oversight become essential.

More output, same requirement for discipline

We know that AI increases throughput. Teams can build more, iterate faster, and discard weak ideas earlier. This is particularly effective in simpler, greenfield environments where patterns are well understood.

In contrast, applying AI to large, established systems introduces complexity. Behavioural dependencies and regression risk accumulate over time. Production systems rely on structured processes such as automated testing, peer review, and observability to maintain stability. Code generated by AI must pass through the same controls once it interacts with these environments.

Adoption patterns reflect this. Developers tend to rely more on AI in unfamiliar areas and less where they already have deep expertise. As usage grows, integration into existing processes remains the deciding factor.

In many cases, the limiting factor is not the model itself, but organisational readiness. Without strong testing practices and clear workflows, increased development speed does not translate into reliable outcomes.

Faster building increases cost exposure

Lower build friction leads to more experimentation. That shifts the operational challenge from delivery to cost control. When teams can provision environments and deploy applications easily, infrastructure and AI service usage can scale quickly. Without visibility, costs can accumulate without a clear return.

Usage-based models help during early experimentation by aligning cost with activity. As applications mature, some workloads may move to a more predictable infrastructure. Matching infrastructure to workload behaviour becomes part of operational discipline.

Where problems emerge, they are typically practical: limited testing environments, incomplete observability, or unclear cost tracking. As development accelerates, governance must keep pace.

The database is becoming part of the logic layer

The same principle applies to the data layer. Predictable, transparent systems are easier to keep contained. Here, the database plays a more active role. It supports how applications retrieve context, process information, and respond to inputs, rather than acting purely as storage. Modern applications often combine multiple data types, including relational data, JSON, vector embeddings, and search. The challenge is integrating these consistently within a single system.

As these workloads grow, performance, scalability, and ease of integration become critical. Predictable behaviour, clear interfaces, and strong documentation reduce friction for both developers and AI systems.

Open ecosystems offer an advantage in this context. Transparency makes behaviour easier to understand, test, and optimise. It also provides a broader base of documentation and community knowledge. Opaque systems, by contrast, introduce uncertainty and make issues harder to diagnose. Predictability is what keeps system behaviour trustworthy as an application moves from contained to exposed.

Agents raise the stakes, not just the speed

AI is also influencing modernisation and migration. It can assist with rewriting applications and generating migration scaffolding, particularly where patterns are well documented.

However, migration remains complex. Behavioural differences and edge cases require careful validation. AI can accelerate parts of the process, but it does not remove the need for oversight.

A sharper version of the problem is arriving. AI no longer only generates code for review. It acts. It executes multi-step tasks, calls tools, and modifies systems with limited supervision. An agent operating in a contained environment is a productivity gain. The same agent in an exposed one is a risk, running without a person in the loop.

This does not change the requirement. It intensifies it. The controls that make generated code safe are the controls that make autonomous action safe: testing, review, observability, clear boundaries. The difference is timing. With agents, the cost of skipping them arrives faster.

Speed without discipline doesn't scale. 

AI has reached a point where it can materially accelerate development. The organisations that benefit most will be those that pair that speed with discipline.

Testing, cost control, and clear operational processes are becoming more important, not less. The question is not only what can be built. It is where it runs and what it can reach. Technologies that are predictable, transparent, and well understood are easier to integrate and manage over time. AI expands what can be built. Discipline determines what works.