There\u2019s a point in almost every AI programme where progress stalls. The technology is impressive. Investment is flowing. Teams can point to a handful of wins. And yet, soon after early success, momentum fades. Not because ambition has dried up. But because organisations are managing AI like a change programme. It isn\u2019t one. They\u2019re managing the next Industrial Revolution.\u00a0\u00a0<\/p>\n
Agentic AI is already delivering value across UK sectors\u00a0<\/p>\n
Almost all UK organisations (90%) are now engaging with forms of agentic AI, though maturity varies. In practice, this means small numbers of semi-autonomous agents embedded in specific processes – enough to deliver value, but also enough to introduce challenges around accountability, control and risk.\u00a0<\/p>\n
In finance, legal and risk, agentic systems are handling complex, repeatable processes end-to-end: analysing data, applying rules and producing outputs, while humans focus on judgement. In customer and product workflows, development cycles are getting shorter and service is improving.\u00a0<\/p>\n
The pattern is consistent: where work is structured, repeatable and data-driven, agentic AI can take on a significant share. Most organisations can point to these early wins. But far fewer are seeing enterprise-wide transformation. Six in ten UK leaders say AI is delivering value; only 5% say it\u2019s delivering returns at scale.\u00a0<\/p>\n
At a certain point, AI stops being about making existing work easier or faster. Beyond 10\u201315% automation, it starts to change how work is produced. That\u2019s a much bigger shift, closer to building a new production system than rolling out a new tool.\u00a0<\/p>\n
The three forces that stall AI programmes\u00a0<\/p>\n
As organisations move towards scale, progress slows. Not because of the technology, but because three pressures emerge: fear, focus and friction.\u00a0<\/p>\n
Fear: the adoption problem is usually a clarity problem\u202f\u00a0<\/p>\n
Fear is often described as resistance to AI, but it\u2019s usually about uncertainty. It starts with job security and becomes something deeper that relates to identity. Teams accept AI on the edges of their work but draw a line around the parts that define their expertise. The part that says, \u201cthis is what I do.\u201d\u00a0<\/p>\n
Only 1% of UK leaders report strong resistance. Most see mixed or uneven adoption. That\u2019s what you\u2019d expect as AI moves into higher-value work. Early on, you can work around this. At scale, you have to deal with it directly.\u00a0<\/p>\n
One way to address it is with a Time-Saved Contract. This explicitly says that time saved by AI is reinvested in higher-value work: better analysis, stronger outcomes, improved customer experience and upskilling. Roles evolve toward editor, supervisor and decision owner. This tells people what success looks like for them, not just for the technology.\u00a0<\/p>\n
Another lever is to\u202fdeliberately\u202fstate\u202fyour value proposition. What are you optimising for? Without a clear answer, initiatives drift toward cost reduction, which fuels anxiety and slows adoption. Progress comes when organisations are explicit about value, roles and outcomes.\u00a0<\/p>\n
Focus: improving BAU becomes replacing BAU\u202f\u00a0<\/p>\n
Once you move beyond 50% automation in a workflow, you are no longer improving business-as-usual. You are replacing it. Organisations end up running two systems at once: the old model delivering today\u2019s results, and a new production system taking shape alongside it.\u00a0<\/p>\n
This is a capacity challenge, and businesses recognise it. 62% say hiring and upskilling constraints could slow AI implementation, and skills gaps are the biggest barrier to moving beyond pilots. In response, they\u2019re investing in reskilling, hiring and role redesign, with 91% willing to pay a premium for AI expertise.\u00a0<\/p>\n
Scaling AI means redirecting\u202flimited skills and capacity\u202ftowards the hardest problems.\u202fUltimately,\u202fit\u2019s\u202fa leadership choice\u202f\u2013 are you willing to put\u202fyour best people on the hardest\u202fbottlenecks?\u202f\u202f\u00a0<\/p>\n
Friction: the organisational immune system\u00a0\u00a0<\/p>\n
Friction is the governance controls, risk management, and the need for accountability as AI moves into core business processes.\u202fIt\u2019s\u202fyour organisational immune system in action.\u202f\u202f\u00a0<\/p>\n
Data security, privacy and risk are the biggest factors shaping AI strategy in the next six months, with 81% of UK organisations saying these concerns could slow or pause implementation.\u00a0<\/p>\n
The problem isn\u2019t the presence of controls, but how they operate. Traditional governance is built for stable, step-by-step processes. AI systems are dynamic: outputs vary, outliers appear, and decisions happen continuously. When controls are slow or unclear, governance becomes the bottleneck.\u00a0<\/p>\n
A common response is \u201chuman-in-the-loop\u201d at the end of a process to check outputs after the fact. By the time problems are detected, work has already gone off course, creating duplication of effort and delays.\u202f\u00a0<\/p>\n
A more effective approach is to place humans at key decision points throughout the workflow: setting direction, choosing between options and intervening where judgement is required, while AI does the heavy lifting in between. This is humans in the lead, not just the loop.\u00a0<\/p>\n
The same principle applies to governance more broadly. The goal isn\u2019t to remove friction, but to make it intentional, with clear ownership, continuous monitoring and controls that match the pace of AI-driven processes. Done well, governance becomes an enabler of scale, not a brake.\u00a0<\/p>\n
Making AI scale\u00a0<\/p>\n
If you want to deploy AI at higher levels of automation, you need to be clear about what you\u2019re building. This isn\u2019t a tool rollout. It\u2019s a new production system \u2013 with new roles, new controls and new ways of creating value.\u00a0<\/p>\n
The question is no longer whether AI can deliver impact. It already is. The real question is whether you\u2019re prepared to redesign how your organisation works to unlock it. Because beyond a certain point, scaling AI is not a technology decision. It\u2019s an organisational one.\u00a0<\/p>\n