Stop chasing models: The real key to AI adoption

On 16 September 2026, Dr Mark Bloomfield, founder of Turbulence and Fellow at Cambridge Judge Business School, addressed Luxembourg AI Factory to outline why organisations must look beyond the underlying AI model to achieve successful AI adoption. Dr Bloomfield argued that the rapid commoditisation of frontier models is shifting the battleground of business innovation.

"Value itself never disappears when something commoditises," Dr Bloomfield noted. "However, it migrates to whatever is still scarce." To illustrate how humans and AI interact, he shared how he uses a voice agent programmed to disagree with his ideas to force him to defend and refine them. He also shared a lighthearted dinner-table experiment where his children built "Hermes", an offline agent designed to listen to family conversations and log parent hypocrisy. "Today's AI is the worst AI you will ever use," Dr Bloomfield jokingly said, urging organisations to focus on the customisable layers of technology rather than chasing the latest model release.

From simple chatbots to agentic commerce

Dr Bloomfield outlined the evolution of AI across four distinct categories: predictive, generative, agentic and physical. He defined predictive AI as using historical data to forecast outcomes, generative AI as creating new content from learned patterns, agentic AI as planning and executing multi-step actions autonomously, and physical AI as intelligence interacting with the real world through robotics or hardware. "AI is greater than generative AI," he told the audience. "Pick the right tool at the right time, including automation.”

Dr Bloomfield then explained that the industry is experiencing a profound economic shift as systems move from making simple recommendations to automating decisions, orchestrating systems and ultimately reshaping markets. He cited Google Maps rerouting traffic in Paris, which inadvertently triggered a housing market crash in quiet residential areas, as an early example of how orchestrating systems can reshape real-world economies.

In this new era of "agentic commerce", where a business's next customer might be an autonomous consumer agent rather than a human, Dr Bloomfield represented the architecture of an agent using a simple equation:

Agent = Model + Harness + Tools + Trigger

Within this framework, Dr Bloomfield identified the harness — which integrates instructions, skills, context and memory — as the most critical layer for proprietary differentiation. Unlike the model, which is selected from an increasingly commoditised market, the harness is where a business encodes its proprietary knowledge and decision logic. As frontier models converge in performance, he argued that competitive advantage is migrating to five scarce sources of value: proprietary data, governance and trust, coordination of people and agents, simulation of decisions before committing capital, and deliberate human-AI collaboration.

On coordination specifically, Dr Bloomfield drew a sharp distinction between automation logic, which speeds up tasks inside an existing workflow, and coordination logic, which removes friction between decisions and handoffs to redesign the system itself. "Control shifts to whoever resolves the system's coordination gaps," he said, "not whoever completes tasks the fastest.”

Governance and sovereign infrastructure as strategic assets

As agents gain autonomy, governance becomes a primary source of competitive advantage rather than a compliance burden. Dr Bloomfield highlighted Singapore’s IMDA Agentic AI Framework, which links the digital identities of autonomous agents to specific employee numbers to ensure legal accountability.

He recommended that organisations adopt the Cambridge "evaluation, deployment and monitoring" (EDAM) framework to manage risk across three levels:

Adopt — building inventories, mapping use cases and deploying responsibly

Defend — securing AI systems, managing access and continuous testing

Govern — maintaining human oversight, ensuring compliance and producing audit trails

Addressing digital sovereignty, Dr Bloomfield emphasised that "sovereignty does not mean self-sufficiency; it means choice." He advised organisations to redeploy software licensing budgets into sovereign infrastructure, data residency and sovereign harnesses to maintain control over their data assets.

How Luxembourg AI Factory supports AI adoption

For local enterprises, Luxembourg AI Factory serves as a vital business enabler to accelerate successful AI adoption. Through its tailored Data Exploration & Storytelling service, the initiative helps small and medium-sized enterprises (SMEs) clean and structure their raw data to ensure quality before model training. To help businesses select and validate their technology, Luxembourg AI Factory offers AI Solution Scouting and Assessment, alongside the AI Assessment Sandbox to test solutions for compliance with the EU AI Act.

Rather than rushing into complex, customer-facing deployments, Dr Bloomfield advised Luxembourgish SMEs to start by identifying specific business pain points, such as food forecasting in restaurants, to bypass internal resistance and build momentum.