AI boosts value of experienced developers, says Lancom CTO

Lancom Technology Chief Technology Officer Aimee Lin said artificial intelligence has increased the value of experienced software developers, as many organisations still lack the skills to turn AI-built prototypes into production systems.

AI tools can now generate working applications quickly, she said, but those outputs often stop at the prototype stage and lack the controls needed for live deployment. As a result, the developer's role is shifting from writing code to making judgements on security, governance and system resilience.

“You absolutely can create applications with AI now. It's real, and pretending otherwise is a losing argument,” said Aimee Lin, Chief Technology Officer at Lancom Technology.

She drew a distinction between software that works in a demonstration and software that is ready for production use.

“But what you get is a prototype. It works and demos well, but doesn't have the essential features necessary for production.”

Access to AI tools has widened software development to a broader group of users, Lin said, while also giving attackers faster ways to find weaknesses. In her view, that has increased the importance of security controls and governance processes around software releases.

“Security and governance, in other words, have never been more important. You're shipping at machine speed into an environment that attacks at machine speed,” she said.

Three layers

Organisations should think about finished software as more than functionality alone, Lin said. She described three layers: a system that works, a system that is safe to deploy, and a system that is ready for external scrutiny.

“Even if a prototype works as it should, functionality is just one of three essential layers. The others are 'safe to deploy', which includes security, reliability and hardening controls, and 'ready for scrutiny', which includes threat modelling, environment separation and audit evidence.”

AI-generated code often lacks the latter two layers, she said, and many companies are not well placed to recognise what is missing. That creates risk when applications move from development into real-world use.

“'Finished software' isn't one layer, it's all three. With most vibe-coded or AI-coded applications, you only see the surface. The application works. It tells you nothing about whether authentication holds up under attack, whether the application scales appropriately, or whether it is fault-tolerant. It tells you nothing about threat modelling, environment separation and the audit evidence necessary when a regulator or an acquirer asks how your application was assessed.”

Beyond code

Lin said the risks around application development extend beyond software itself to the infrastructure on which it runs. Hosting arrangements, identity controls, backup systems and patching all form part of application security, she said.

“There's also the infrastructure where it's hosted. And that is often overlooked.”

Weak hosting decisions can undermine even well-written software, she said.

“Environment separation, identity, backup and recovery, patching: what happens when the underlying platform has a bad day? This is part and parcel of application security. Hosted poorly, even the best programmed application is at risk.”

Changing roles

These demands explain why experienced developers remain central even as AI automates parts of coding, Lin said. In her view, the work is not disappearing but becoming more concentrated in fewer hands.

“That's precisely the work of experienced developers. It's building the checks that run automatically and knowing which checks matter most. Anyone can wire up a test suite. Knowing what it fails to cover is the part that takes a decade.”

While AI may reduce the cost of creating a prototype, it does not reduce the cost of failure once software reaches production, she said. Operational outages, cyber breaches and costly rebuilds remain likely consequences of flawed systems.

“AI has made prototypes cheap. It hasn't made production mistakes any cheaper. The economics haven't changed. Get a prototype wrong and it costs an afternoon. Get it wrong in production and you're looking at an outage, a breach or a rebuild.”

Software development jobs are now being compressed as activities once divided among analysts, designers, engineers and testers are combined, Lin said. That means developers need stronger judgement rather than narrower technical specialisation.

“The handoffs are compressing, not the work. Requirements and design still happen, but they're no longer separate phases. Automated testing sits inside the build loop rather than at a stage gate. What once moved between analyst, designer, engineer and tester now sits with fewer people holding more of the picture.”

That places a premium on people who can spot when AI systems are wrong, she said.

“You need people who can be trusted to judge when agents get it wrong.”

At the same time, Lin said AI tools have changed her own work by making it easier to combine system design with hands-on building.

“I had to choose between designing systems and building them; there wasn't enough time for both. Agentic tools have given me that time back. I'm building again.”

The technology may expand what developers can do, but not replace the judgement that comes from experience, she said.

“They can build architectural judgement without stepping away from the craft. You can buy agentic capacity right now. Good judgement comes with practice, and it will always be scarce. While AI has certainly made coding fast, it's also moved the hard part of software development to a place where most organisations aren't looking. That's exactly where experience earns its keep, because the developers who matter are those who know how to deliver production applications that leave nothing to chance.”