Andrew McMahon

Part of the IT Leaders 100 - a list of the most influential IT leaders in the UK in 2026.

Andy McMahon is a Principal AI and MLOps Engineer at Barclays, specialising in building production-grade AI systems at scale. He previously led MLOps initiatives at NatWest and has delivered AI platforms and products across logistics, energy and financial services.

Andy is the author of Machine Learning Engineering with Python and a guest lecturer on AI and MLOps at the University of Oxford. His work focuses on turning experimental machine learning into reliable, secure and high-value production systems.

What's the most exciting part of your current role?

Turning cutting-edge AI research into production systems that actually deliver value. In large organisations there's often a gap between experimentation and reliable deployment, and closing that gap is where the real engineering challenge lies.

I work at the intersection of AI, software engineering and platform architecture - designing systems that allow teams to safely build, evaluate and deploy AI at scale. That includes everything from model evaluation and observability to governance and secure deployment in regulated environments.

What motivates me most is enabling others: building platforms, communities of practice and technical standards that allow hundreds of engineers and data scientists to move faster with AI. When that works well, you're not just delivering one solution, you're creating an ecosystem where innovation compounds across the organisation.

What do you do in week one of a new project to avoid pain in month six?

Week one is about clarity and constraints.

I focus on three things early: the problem definition, the operating model, and the non-functional requirements. Too many projects jump straight to building before aligning on what success looks like, who owns which decisions, and how the system will actually run in production.

Practically this means stakeholder mapping, defining success metrics, agreeing architecture guardrails and identifying operational concerns like monitoring, evaluation, security and cost.

I also push teams to prototype quickly but deliberately. Early experiments help expose hidden complexity in data, integrations or infrastructure before the system becomes difficult to change.

In my experience most "month six problems" aren't technical; they come from unclear ownership, missing evaluation frameworks, or systems that were never designed for production. If those foundations are set early, the engineering becomes much easier.

What’s the part of your estate you most want to simplify in 2026?

The biggest opportunity for simplification across most organisations is the AI platform and tooling landscape.

Over the last few years many teams have independently adopted different tools for model development, experimentation, deployment and monitoring. While this can accelerate innovation early on, it eventually creates fragmentation and operational overhead.

In 2026 I'm particularly focused on simplifying the end-to-end lifecycle for AI systems -bringing together model management, evaluation, observability and deployment into a coherent platform experience.

The goal isn't fewer tools for the sake of it, but a clearer developer journey: predictable pipelines, consistent governance, and shared infrastructure that allows teams to focus on solving business problems rather than stitching platforms together.

Well-designed platforms reduce cognitive load, improve reliability and make responsible AI practices far easier to implement at scale.

Excluding smartphones, what piece of technology could you not live without and why?

An AI coding and research assistant.

Tools like modern AI copilots have become a genuine thinking partner when designing systems. They accelerate everything from exploring ideas and prototyping code to summarising documentation or stress-testing architectural approaches.

What's most interesting isn't just speed, it's how they change the way you think. They make it much easier to iterate on ideas quickly and explore alternative approaches before committing to a design.

Used well, they don't replace engineering judgement; they amplify it.

What excites you most about the future of IT?

The transition from software that follows instructions to systems that can reason, learn and collaborate with humans.

We're moving into a world where AI systems become active participants in the engineering process - helping design software, analyse data, automate operations and accelerate discovery.

The challenge now isn't just building models; it's building reliable AI systems. That means new approaches to evaluation, monitoring, governance and platform design.

The organisations that succeed will be those that combine strong engineering discipline with the ability to experiment quickly. If we get that balance right AI won't just improve existing processes, it will fundamentally expand what small teams are capable of building.

What talent are you quietly proud of that would never show up on LinkedIn?

I'm good at translating complex technical ideas into language that different audiences can understand.

Whether I'm speaking to engineers, executives, regulators or students, I enjoy finding the explanation that makes a difficult concept suddenly click.

That skill isn't something you usually list as a technical competency, but it's incredibly valuable when building AI systems in large organisations. A lot of the work is bringing people along the journey, aligning teams, building confidence in new technology, and helping stakeholders understand both the opportunity and the risks.

What’s your most-used keyboard shortcut or command that you wish everyone knew?

On my macbook -> Cmd+shift+4 gives me a nice resizable screenshot window.