Speaking in front of a room filled with a host of Scottish tech leaders hosted by DLA Piper, and organised by ScotlandIS, Marshall posited what he believes will be this year’s keyword: Uncertainty.

It’s a fitting choice as we stand on the precipice of great technological change, with AI hype driving huge disruption in everything from tech funding and business operations, down to a societal level with regards to shifting policy agenda and the wholesale reshaping of educational and cultural frameworks – for better or worse.

As a result, AI unsurprisingly dominates large swaths of Marshall’s predictions.

After warming the audience up to the backdrop in which Marshall’s top ten tech trends will take place, he started to list them off. 

1 – Bridging the Gap Between Strategy and Execution

Marshall began by addressing a critical disconnect between different layers of organisations.

He said: “At the board level, you have people looking at high-level strategic issues, while on the ground, enthusiastic individuals focus on what cool new tools they can use.”

However, the middle layer of management – people responsible for operational execution – feels stuck.

“They’re not lost because they’re stupid,” he said.

“They’re lost because their imperatives are aligned with the past, not the future,” he explained. This misalignment, he warned, could hinder organisations from adopting transformative technologies like AI.

Essentially, the pace of development with new technologies – while certainly exciting to those in the ‘top and bottom layers’ – are causing bottlenecks for those in the middle that have to deal with implementation.

2 – The Convergence of IT and OT

Historically, information technology (IT) and operational technology (OT) have operated in silos, but Marshall sees this changing rapidly.

“IT and OT have always been enemies,” he noted.

“OT people have been running the factories very successfully, IT people have been running offices, and they’re almost separate worlds. Nobody really liked each other. They didn’t talk to each other. IoT [Internet of Things] was the sort of beginning of things, but it was only a small part of it.”

Marshall highlighted both positive and negative drivers for this shift, such as the need for unified security frameworks and integrated building management systems.

“With a single identity domain, you can unify your whole network, ensuring better resilience and security for the organisation,” he emphasised.

Marshall says this convergence of IT and OT will help build the “digital DNA” of organisations.

3 – Smart Glasses: Finally Cool?

Smart glasses, once dismissed as clunky and impractical, are experiencing a renaissance, according to Marshall.

“Remember Google Glass? It was uncomfortable, hot, and lacked a clear use case,” he recalled. However, with newer, more stylish designs from brands like Ray-Ban and innovative industrial applications, the narrative is shifting.

Marshall says: “Products like BMW’s head-up display glasses for motorcyclists are pushing boundaries, but they also raise safety concerns,” he added. While consumer adoption remains limited, niche applications are proving their worth.

4 – Compliance Gets Complicated

Where is the data? What regulations apply where? Who can read it? How has it been processed? What risk treatment is defined? Who can change it?

As new sub-sets of nascent technologies become adopted across myriad industries, compliance is becoming every more layered, fragmented and obtuse – Marshall sees that continuing this year.

He said: “AI compliance standards, like ISO 42001, are making things more complicated,” Marshall explained. He highlighted the challenge of tracing AI-generated outputs back to their source data, a key requirement for regulated industries.

“As AI models evolve, demonstrating compliance will be a huge challenge, especially in regulated industries like finance, medicine, and insurance.” He also noted the fragmented landscape of AI regulation, which is adding complexity for global organisations.

5 – Bimodal AI

[bimodal AI i don’t think is commonplace enough, but be good to introduce it again in the text to explain what it means]

Marshall reflects on his time at Gartner where he first came across the idea of Bimodal AI (though he concedes its something that they’ve “stopped talking about.”)

“It was called Bimodal, as well as ‘Mode 1’ and Mode 2 for the simple reason that no one could agree what the two different Modes involved.”

Essentially, Bimodal had one side that was “entirely dedicated to your existing infrastructure – ensuring that your legacy work systems work.

Mode 2, as Gartner puts it, is ‘exploratory, experimenting to solve new problems and optimised for areas of uncertainty.’

While general-purpose AI like ChatGPT dominates headlines, Marshall sees a growing focus on small, specialised models.

“We need to move away from trying to create systems that do everything,” he said.

Instead, the emphasis should be on constrained solution spaces that deliver specific outcomes efficiently.

“Smaller models are more energy-efficient, accurate, and easier to validate,” he argued. He pointed to legal and medical fields as areas where specialised models can prevent misinterpretations of precise terminology.

6 – Synthetic Personas

Synthetic personas have the potential to help decision-making and market research by enabling businesses to simulate insights from thousands of virtual individuals.

These personas can range from “conservative to liberal critics,” providing insights on everything from fashion trends to medical prescriptions.

“Imagine creating 10 synthetic reviewers to analyse a report, each with their own unique perspective.”

However, the potential of synthetic personas hinges on data quality. Poorly trained personas could misrepresent trends, leading to inaccurate conclusions.

“If the training data isn’t up to scratch, the outputs will reflect those inaccuracies.” Which brings us to…

7 – Data Quality as a Major AI Blocker

Data quality is a critical barrier to AI development, particularly with unstructured data like PDFs, text files, and presentations, which constitute “over 90% of enterprise data” – according to IDC report that Marshall cites.

Unfortunately, the majority of this data is rife with issues as Marshall points out using data from shelf.io: “22% of files are outdated, 82% have major inaccuracies, and 33% of files contain duplicate or redundant information.”

One challenge with this kind of unstructured data that Marshall points out is “data entropy,” where information decays over time due to poor tagging, misclassification, or duplication. SharePoint and Dropbox repositories often become “chaotic loft spaces,” exacerbating the problem.

To address these challenges, companies are emerging with solutions for unstructured data quality management – an area Marshall predicts proliferation in 2025.

8 – AI-Enhanced Software Development

Marshall is quick to differentiate between software development and engineering, to preempt any concerns around AI replacing the need for human input.

He says: “Software engineering is not about coding, it’s about understanding that there’s a problem. It’s understanding the requirements, it’s understanding what you’re building, it’s about architecture, it’s about performance, and a lot of that stuff is still secret sauce for humans.”

Marshall states that AI tools are “amazing at creating small chunks of code. They’re very good at getting to an 80% solution and then completely destroying it.”

Marshall references Benedict Evans, who once compared AI in development to having an infinite number of interns.

While AI excels at repetitive tasks like writing tests or analysing legacy codebases, it struggles with nuanced tasks like defining architectures or solving complex problems.

Ultimately, software engineers and human oversight remain critical to the process.

Expertise in translating requirements, building scalable architectures, and testing solutions means that human oversight isn’t going anywhere soon, according to Marshall. However, AI will play a huge role in removing some of the more onerous aspects of coding.

9 – Cleantech

AI adoption risks further widening the gap between tech-rich and tech-poor organisations.

Companies with modern, AI-ready systems enjoy greater productivity, while those with “10-year-old PCs and poor performance” are being left behind.

SMEs are particularly vulnerable, as many lack the resources to upgrade systems or implement robust cybersecurity measures. Marshall noted that “old, slow systems limit productivity, scalability, and overall competitiveness.”

On an individual level, the divide is just as pronounced.

“Those with access to cutting-edge tools and training will surge ahead, leaving others excluded from the AI-driven economy.”

Without intervention, the gap between tech haves and have-nots will continue to grow, exacerbating societal and economic inequality.