Annex II: Some drivers impacting the use of agentic AI
In our scenarios section, we briefly identify some of the drivers behind the emergence of agentic AI. In this annex, we provide more detail on these drivers.
Agentic AI drivers
Model training costs
Analysts project that keeping up with AI applications will demand compute power requiring USD $5.2 trillion investment by 2030. 34 The cost of training large AI models has escalated over time, with researchers predicting that by 2027 they will require billion-dollar investments. This could potentially concentrate power and control into only the larger technology providers who can invest at that level. 35
A drop in compute prices and increased processing power driving accessibility
Research from Epoch.ai shows that the amount of physical compute required to achieve a given performance in computer vision models is decreasing at a rate of three times per year. This is driven by efficiency gains and systemic improvements. 36 There have been improvements in factors such as computational performance, compute required to achieve a given performance and the cost of training frontier AI models. Compute budgets are effectively doubled by the introduction of better algorithms every nine months. 37
Increasingly large, high-quality datasets are available
Training datasets have grown over time, driven by demand and improvements in the creation and use of synthetic datasets. 38
Venture capital funding and the AI bubble
In the first half of 2025, agentic AI startups worldwide received approximately USD $2.8 billion of venture capital funding. 39 Analysts predict that 10% of all AI funding in 2025 will focus on agentic applications.
In the UK, while the number of AI investment deals has decreased slightly, the average deal size has increased. Analysts expect this increase to create 6,500 jobs. 40
‘Fear of missing out’ driven by the hype cycle and marketing
Gartner reports that organisations are ignoring the real cost and complexity of deploying agentic systems. This could ultimately lead to misapplication of the technology, driven by hype. They predict that over 40% of agentic AI projects will be cancelled by 2027 because of unclear business value and inadequate risk controls. 41
Further contributing to the issue is the practice of ‘agent washing’. This refers to vendors rebranding other technologies as agentic to take advantage of interest in the technology.
Highly intersectional technology
One in six UK organisations is using at least one type of AI in the workplace. Performance increases in compute power, data and algorithms used to process that data to complete tasks, as well as greater awareness of AI applicability, are driving this. 42
Previous government information shows the IT and telecommunications sector has the highest AI adoption rate at 29.5%, with the legal sector at 29.2% and hospitality, health and retail sectors at around 11.5%. 43 The majority of use is data management and analysis (9%). Other applications included natural language processing and generation (8%) and computer vision and image processing and generation (5%). This indicates a range of applications for analysis as well as content generation. People increasingly interact with AI at work, and on the devices and in the applications they use at home.
Cost savings from reduced staff costs and labour
While technologies such as generative AI have seen widespread adoption, they seem to have had little impact in terms of value generation. 44 Some argue that agentic workflow automation will be far more impactful, bringing personalisation, adaptability and resilience to operations. 45 Proposed frameworks for measuring return on investment in the implementation of agentic AI look at productivity gains, cost savings and increased customer satisfaction derived from ADM.
A push on AI from national governments
In January 2025, the UK government released its AI Opportunities Action Plan, which detailed its commitment to using AI for the economic benefit of the nation and the social benefit of citizens. The USA followed in July 2025 with its own AI Action Plan. In August 2025, China issued a guideline on its AI Plus initiative with the aim that the ‘intelligent economy’ will become a significant growth driver for the Chinese economy. All the action plans share a focus on innovation and growth.
34 McKinsey article on the cost of compute power
35 Time article on the cost of Building AI
36 Epoch AI article on Machine Learning Trends
37 Epoch AI article on Revisiting Algorithmic Progress
38 Epoch AI article on Trends in Training Dataset Sizes
39 Prosus report about the rise of AI agents in the workplace pg.14
40 UK Government Artificial Intelligence sector study 2024
41 Gartner article on predicted Agentic AI project failure rate
42 Office for National Statistics report on understanding AI uptake and sentiment
43 Capital Economics report on AI Activity in UK Business