Agentic AI: A Once-in-a-Generation Opportunity to Solve the UK's Productivity Puzzle
Britain's central economic problem is not a shortage of ideas, rather, a shortage of execution.
UK productivity growth collapsed after the 2008 financial crisis, averaging roughly 0.3% a year over the following decade against around 2% before it.1
The ONS calls this the "productivity puzzle", and describes the slowdown as the slowest sustained recovery since the Second World War.
Had the pre-2007 trend held, productivity today would be about 16% higher 2,and with it wages and living standards.
The recovery since the pandemic has not closed the gap. Output per hour worked stood only 3.5% above its 2019 level in the first quarter of 2026.3
With public debt and spending pressures rising across health, pensions and welfare, the UK's path to sustainable growth lies in productivity, not expenditure alone. And productivity, ultimately, depends on how efficiently work is executed across systems and teams daily.
Breaking the Fiscal Deadlock
Much of the government's room to invest in growth has been absorbed by the rising cost of day-to-day commitments. When budgets tighten, capital investment, the spending that raises future capacity, is the first thing cut, because its cost is felt now and its return arrives years later. It is a familiar trap: the country needs to invest to grow, but feels it has no fiscal space to invest.
Agentic AI offers a way out, because it asks for comparatively little public capital. Unlike a new rail line or a rebuilt energy grid, productivity gains from agents draw mostly on systems and data organisations already own, which makes them well suited to going first. The sequence is straightforward: higher productivity raises output, higher output raises GDP, and faster growth both widens the tax base and lowers the debt-to-GDP ratio, creating fiscal headroom. Roughly a third of additional national output reaches the Exchequer over time. Low-capital gains, in other words, can fund the high-capital investment that has stalled.
One caveat keeps the argument honest: productivity gains take years to register in the national figures, and reinvesting the proceeds in infrastructure rather than elsewhere is a political choice, not an economic certainty. But the direction is sound and self-reinforcing. Done well, it turns a vicious circle into a virtuous one.
The Constraint is Deployment, Not Capability
Most technologies improve individual tasks; Agentic AI transforms entire workflows, fundamentally changing how work happens. Instead of helping one person draft one email more quickly, an agent can coordinate a process end to end: reading from several systems, taking a sequence of actions, and escalating to a human where judgement is required.
That distinction is the whole opportunity.
Back in 2017, PwC predicted that AI could contribute up to £232 billion to the UK economy by 2030, equivalent to a 10.3% boost to GDP. 4 With the current advancements, the economic gains are likely to be higher. But to realise this potential, what's needed is deployment at scale, not increasingly sophisticated AI models.
Where the Gains are Largest
In the NHS, clinicians lose substantial time to administration and coordination that never touches patient care. Agents can manage referrals, scheduling, discharge and patient communication across disconnected systems, recovering thousands of clinical hours a week. 5
In manufacturing, which produced roughly $279 billion of UK output in 2024, 6 agents can coordinate production schedules, optimize inventory and orchestrate predictive maintenance, lifting output and resilience without a proportional rise in labour cost.
In financial services, which contributed more than £208.2 billion to the UK economy, accounting for 8.8% of total economic output, 7 much of the operating cost sits in compliance, risk and fraud detection: evidence-heavy, rules-bound work that agents handle well, escalating anomalies to the people whose judgement matters.
Across the public sector, case management, planning, licensing and procurement are coordination problems before they are anything else, where agentic AI could provide significant support.
The pattern is the same in each. The value is not in replacing the expert. It is in clearing the operational debris that keeps the expert from doing expert work.
Why Most of this will Fail, and What Separates the Exceptions
Honesty about the obstacles is what gives optimism its credibility. A widely cited Harvard Business Review Analytic Services research found that while 94% of organisations recognise connected data, processes and applications as essential for AI success, only 27% have achieved such integration.8 That single statistic explains most of the failed pilots already accumulating across British industry.
An agent is only as capable as the systems it can reach and the data it can trust. Data quality, interoperability and integration are not chores to be rushed before the interesting work begins. They are the work, alongside the monitoring, security and auditability that make an autonomous system trustworthy. The organisational requirements are just as real: the firms that succeed treat agents as augmentation rather than replacement, build AI literacy across the workforce, and keep humans in the loop on decisions affecting patients, customers and citizens.
A Narrow Window
The UK's opportunity lies not in building smarter AI, but in deploying Agentic AI responsibly to boost productivity.
The economies that move fastest from experimentation to operational transformation will define the next decade of competitiveness. Britain can be one of them. The work is not glamorous and it is not theoretical. It is the patient business of connecting systems, cleaning data and rebuilding workflows so that capable technology can finally do what it has long promised. That is the puzzle. It is also, at last, solvable.
References:
