More organisations are investing in generative AI but struggle to scale agentic systems, says Capgemini

A new Capgemini study reveals that while companies are increasing investments in generative AI, few are successfully deploying agentic AI at scale, hindered by trust, skills, and governance challenges amid sustainability concerns.

Capgemini’s new study, “Leadership perspectives: AI realities,” paints a picture of organisations caught between rapid investment in generative and agentic AI and persistent barriers to turning pilots into scaled value. Drawing o...

n interviews with chief data and AI officers and a broad industry survey, the report finds that 80% of organisations are increasing investment in generative AI strategies even as only a small fraction have moved to enterprise‑wide deployments.

According to the Capgemini report, 67% of mature organisations now deploy hybrid AI strategies that combine generative and traditional models, while 91% of executives expect generative AI to enhance process efficiency. The study also highlights the human dimension of scaling: 85% of respondents identified training and acculturation as essential to moving from experimentation to transformation. The report cautions that successful pilots frequently fail to scale, return on investment is often poorly tracked, and many organisations underestimate the hidden costs of change management.

Industry commentators and follow‑up analyses amplify both the promise and the limits Capgemini describes. Capgemini itself projects that agentic AI, systems that autonomously execute tasks and decisions, could deliver up to $450 billion of economic value over the next three years through revenue growth and cost savings. TechRadar Pro and IT Pro pick up that figure, noting agentic AI’s potential to convert passive generative tools into agents that execute workflows, reduce errors and realise measurable ROI. TechRadar Pro argues that agentic systems can close the gap between widespread generative use and concrete business outcomes by integrating with stacks and automating continuous workflows.

But the gulf between potential and practice is stark. The Capgemini findings, echoed in reporting by IT Pro and TechRadar, show only around 2% of organisations have fully scaled AI‑agent deployments; another minority are piloting or implementing, while most remain in early planning stages. The shortfall is especially acute in the UK, where TechRadar reports that only about 1% of firms have fully scaled agents and many risk missing out on the projected economic gains.

Trust, skills and governance are recurring obstacles. Capgemini’s marketing leaders survey found nearly 70% of marketing executives believe agentic AI will be transformative, yet just a small share, 7% in that sample, strongly agree that AI has improved marketing effectiveness. The company said skills shortages and data privacy concerns are central to that disappointment. IT Pro highlights lack of trust in AI agents as a principal reason deployments lag, noting leaders see competitive advantage in rapid scaling but remain cautious.

Security experts warn of further complications. Haider Pasha, EMEA CISO at Palo Alto Networks, told industry audiences that agentic AI presents major security challenges and predicted a high failure rate for projects lacking strong governance and cyber controls. He attributed potential project failures to unclear business objectives, weak governance and insufficient security measures, a view that underscores Capgemini’s call for new forms of governance as agentic systems proliferate.

The environmental footprint of large language models also features in Capgemini’s analysis: the report states that some LLMs may consume as much as 4,600 times the energy of traditional natural language processing models. That calculation frames sustainability as a material consideration for organisations weighing broader deployment, and strengthens the argument for hybrid strategies and model optimisation to balance capability with carbon cost.

Taken together, the research and commentary sketch a pragmatic agenda for executives seeking to convert AI investment into durable value. Industry data shows blended model strategies, clearer ROI metrics, stronger training and cultural change programmes, tighter security and governance frameworks, and attention to model efficiency and sustainability are all necessary to scale agentic and generative AI responsibly. Capgemini’s report offers practical guidance on these fronts but also serves as a reminder that the technical promise of agentic AI will be realised only where organisations build trust, skills and controls around its use.

Source: Noah Wire Services