So, where does AI actually accelerate enterprise software development in ways a CIO or CTO can trust? 

From AI theatre to production value

Martin sees a broad spectrum of outcomes across enterprises: some genuinely moving 30% or more of their AI initiatives into production, others still trapped in endless proofs of concept.

The biggest difference isn’t the model they picked. It’s how clearly they define the problem.

Too many organisations start with “We should do something with AI” and then launch pilots with fuzzy goals. They play with agents, prompts, and prototypes but struggle to answer a basic question: Which metric are we trying to move?

The organisations that are breaking out of pilot mode do something more old-fashioned – and more disciplined. Before they build:

• They decide which business KPI to optimise: cycle time, cost, capacity, risk, revenue, NPS, or a specific productivity metric.
  
• They design the AI initiative around that measure.
  
• They treat iteration as a first-class citizen: try, measure, adapt, repeat.
  

In other words, AI doesn’t replace product thinking or basic governance. It makes both more important. Without clear outcomes and feedback loops, you just get faster at generating… more experiments.

Vibe coding: intoxicating first date, not the marriage

One of the most visible ways AI is accelerating development is “vibe coding”: describing what you want in natural language and watching an AI agent generate an application.

Martin calls it an “intoxicating” experience – in a good way. Instead of weeks of specs and wireframes, business stakeholders can sit with an AI, explain what they’re trying to do, and see a working prototype emerge. For IT leaders, this can radically compress the slowest part of the lifecycle: figuring out what to build.

But there’s a catch. If vibe coding is the first date – fast, exciting, full of possibility – enterprise software is the marriage: long-term, demanding, and unforgiving of shortcuts. Under the surface of that instant app, the code may be messy, inconsistent and undocumented. Security assumptions may be unclear. Dependencies may be brittle. Governance is usually nonexistent.

Martin cites a friend who built a three-tier web app over the weekend using AI and cloud services to share neighbourhood security camera footage. The app worked. The problem? He’d created “a mountain of tech debt” and knew that as soon as neighbours asked for changes, it would become unmanageable.

For CIOs and CTOs, that story is playing out inside the enterprise today: everyone can generate software; not everyone is generating something you want near your core systems.

The implication is not “don’t vibe code.” It’s: treat vibe coding as a powerful front-end to a disciplined software development platform and process. Use it to explore, prototype and converge on requirements – but don’t confuse “it runs” with “it’s production-ready.”

Blending non-deterministic AI with deterministic platforms

Underneath the excitement sits a tension every enterprise leader feels: large language models are inherently non-deterministic. Ask the same question twice, and you may get different answers. That’s wonderful for creativity – and dangerous when you’re in a regulated industry or dealing with money, safety, or compliance.

Martin’s view is that the way forward isn’t to try to make AI behave like traditional software, but to pair non-deterministic reasoning with deterministic execution.

He describes a pattern many enterprises are adopting:

• Use AI agents for the parts of a workflow that benefit from reasoning, interpretation, summarisation and judgment – for example, reading documents, interpreting unstructured input, or proposing decisions. 
• Use deterministic components for the parts that must be consistent, auditable, and tightly controlled – for example, which email template is approved for a customer, how a policy is applied, or how data is written back into core systems.
  

In OutSystems’ world, this manifests as “agentic workflows”: orchestration that explicitly defines where agents are allowed to “think” and where they must choose from predefined, governed options.

This architectural separation matters. It gives enterprises levers:

• You can change the model or the prompts without rewriting your entire app.
  
• You can prove to auditors exactly which steps are controlled and which are probabilistic.
  
• You can let AI explore possibilities while ensuring the last mile of customer interaction, record-keeping, and policy enforcement remains predictable.
  

The broader lesson: AI accelerates development not by replacing traditional architecture, but by sitting alongside it. Platforms that bake in observability, lifecycle management, security, and repeatable code generation give you a chassis sturdy enough to carry the ambiguity that AI introduces.

The enduring role of humans

As AI adoption grows, many IT leaders are asking the same question: Is AI designed to replace humans or work alongside them?

Martin is unequivocal. He uses AI every day – to research customers, understand industries, and prepare for meetings. But the part of his job he values most – building relationships, understanding context, earning trust – isn’t up for automation. The same will be true for many roles in IT and the business.

For software teams, AI will reshape developers’ work, not eliminate it. Some skills will become less central (handwriting boilerplate, re-implementing common patterns), while others will become vital: system design, security, data governance, agent orchestration, and the ability to connect technology to business outcomes.

Platforms like OutSystems matter not because they make AI possible, but because they make AI manageable. They give organisations:

• A deterministic backbone beneath non-deterministic models.
  
• A way to control how and where AI interacts with core systems.
  
• A lifecycle for change that matches enterprise expectations.
  
• A place where developers can encode guardrails, patterns and institutional knowledge that AI alone doesn’t have.
  
  

Who dares, wins

AI accelerates enterprise software development in all the ways the hype promises: faster prototypes, shorter cycles, more automation. But the organisations that will truly benefit aren’t the ones with the flashiest demos. They’re the ones that pair AI with disciplined platforms, clear goals and human orchestration and oversight.