Zoe Webster has spent over 25 years in the field of AI, from developing systems as a computer scientist to shaping national strategy at Innovate UK, leading BT’s AI enablement centre and now working as an independent AI advisor. In this conversation with JP Doggett, she talks about the confusion around “agentic AI,” why many organisations feel burned by earlier promises and how leaders can identify where AI really fits in their supply chain.
JP Doggett: Zoe, you’ve had quite a career journey across the AI landscape. Could you give us a quick overview?
Zoe Webster: I started out classically trained in AI - computer science, a master’s and a PhD - and worked hands-on as a scientist developing and demonstrating systems. Later I moved to Innovate UK, where I spent 13 years looking at technology innovation through a macroeconomic and societal lens. Then I joined BT and set up the AI centre of enablement. That meant leading a team of engineers and scientists across dozens of use cases across B2B, B2C and internal operations, including procurement and supply chain.
Quite often I was asked to vet suppliers and ask difficult questions to see what was really under the bonnet. The hype was usually greater than the substance. A year ago I set up on my own to advise organisations on AI strategy, practice and governance and I also do talks on adoption, skills and where AI is real versus hype.
JP Doggett: The latest hype cycle seems to be around “agentic AI.” What’s your perspective?
Zoe Webster: There’s a lot of confusion. True agentic AI would mean autonomy: systems that reason across multiple goals, adapt in real time and orchestrate tasks without humans in the loop. We’re not there yet. The nearest thing in practice is self-driving cars and even they are limited more by regulation, insurance and management systems than by the technology.
What many companies call “agents” are really programs that perform a specific task, sometimes with AI elements like machine learning. That’s different from agentic AI. For example, an agent that answers customer queries is not the same as one that runs your entire marketing function. The difference is orchestration: if a human still decides who does what, it’s not agentic. If the system is adapting and making those calls itself, that’s closer to agentic AI.
JP Doggett: In our best practice-sharing network, people talk about AI as both a paradigm shift and a disappointment. How do you help clients reconcile that?
Zoe Webster: I don’t try to brush over the scepticism, I empathise with it. AI is not magic. If the data isn’t in good shape, AI won’t fix it. Many leaders have been sold the wrong tool for the wrong job. They’ve been told “AI will solve your problems” and then burned when it didn’t.
The first step is to put AI aside and ask: what is the organisation’s purpose, what are the goals and where are the inefficiencies? What would you love to know about your customer, suppliers or partners that you don’t know today? Once you frame it that way, you can work back to whether AI has a role and if so, what kind of AI approach might help. Sometimes the right answer isn’t AI at all...it might just be a better interface or better data management.
JP Doggett: It's a very dated reference but one worry I hear is what I call the “Betamax problem” i.e. investing in the wrong technology and being stuck when the standards shift. Do you think that’s valid?
Zoe Webster: Absolutely. It’s the right question to ask. Technology acts as an amplifier: it can make a good process more efficient or make a bad one worse. That’s why the focus should be on what task you’re trying to achieve, not on pinning your flag to a particular vendor or product.
To future-proof, you need to be as platform-focused and vendor-agnostic as possible. That way you can plug and play as things change. Otherwise you risk embedding inefficiencies and making it harder to adapt later.
JP Doggett: Many supply chain leaders are uncertain about what questions they should be asking vendors to test their claims around AI. What’s your advice?
Zoe Webster: Always ask how their models are tested. How well do they perform in cases like yours? How is the model governed? How does it adapt when your needs change? Too often suppliers can’t answer or haven’t tested in relevant contexts. That’s a red flag.
And don’t assume big equals safe. A large vendor might sell you a general-purpose system but you may need something bespoke to handle the specifics of your data or contracts. In one financial services example, a large language model couldn’t extract terms from contracts effectively so the company built a tailored solution instead.
JP Doggett: For organisations without massive resources, is building in-house realistic?
Zoe Webster: It’s more realistic than people think. A skilled data scientist with some engineering knowledge can prototype quickly. It may look rough but it can prove the concept. Students on master’s programmes are often keen to do real projects, sometimes for free, if you provide data. Freelancers can also help.
The trick is to start small, run an experiment and build trust in stages. That way you can decide whether to scale without making huge upfront commitments.
JP Doggett: From your experience, what’s most often missing for organisations struggling to use AI?
Zoe Webster: Translation. Business leaders need someone who can translate strategic goals into technical specifications and technical results back into business value. At BT, I made my team write use case specifications: what are we doing, why, what does good look like, how will we test it, how does it integrate?
That translation layer is critical. It’s not just about AI skills but also process expertise, analytics and communication. Without it, you either get business people buying into hype or data scientists building clever models that don’t solve the right problem.
JP Doggett: With your Innovate UK background, how do you see government’s role in AI adoption?
Zoe Webster: Innovate UK runs programmes like Bridge AI to connect people with ideas and needs and they still provide grants and loans to accelerate innovation. But government ambitions to be an “AI champion” often underestimate how long adoption takes. Scaling AI in organisations requires time, data readiness and governance. Policy can encourage it but delivery is always slower and messier in reality.
For Zoe, the message is simple but challenging: don’t start with AI, start with the problem. Be wary of hype, stay vendor-agnostic where you can and invest in translation skills that connect business goals with technical execution.
As she puts it, “AI is not magic. It won’t fix your data. Focus on what you want to achieve and then decide if AI has a role.”