Agent-washing and the cost and value of agent projects

In June 2025 Gartner predicted that over 40 per cent of agent projects would be cancelled by the end of 2027. The reasons the firm gives: costs are rising and the business value remains unclear. Risk controls are often missing too. According to Gartner, most projects are early experiments driven by hype and often aimed at the wrong problem. In a January 2025 poll of 3,412 webinar attendees, 19 per cent said their organisation had invested heavily in agents and 42 per cent cautiously; 31 per cent were waiting to see. Gartner also names the phenomenon this briefing is named after: agent-washing, the rebranding of existing products such as assistants and chatbots as agents. Of the thousands of vendors that call themselves agentic, Gartner reckons about 130 really are, and many applications sold as agents do not need one.

Meanwhile use is growing fast, especially at large companies. Microsoft’s own telemetry shows fifteen times as many active agents as a year earlier, and eighteen times as many at large enterprises. Of the 20,000 AI users surveyed in ten countries, 16 per cent use agents for multi-step work; in those teams people are almost twice as likely to share AI tips and mistakes with each other (61 against 36 per cent). McKinsey finds that 40 per cent of organisations with more than a billion in revenue are scaling agents, up from 27 per cent a year earlier, while smaller organisations are stuck at 22 per cent. One in five organisations says the running costs of AI are already holding back use.

BCG surveyed over 1,300 executives in September 2026 and finds the dividing line Gartner predicted. Agents’ share of total AI value rose in a year from 17 to 22 per cent, and is expected to reach 39 per cent in 2030. But 44 per cent of front-runners get value from agents today, against 2 per cent of laggards. What the front-runners do differently: 61 per cent have moved from separate pilots to a single multi-year programme for the whole organisation, and 95 per cent manage against clear KPIs or track the value directly in the profit and loss account. Responsibility sits at the top. Companies with all six controls in place get three times as much value from agents as companies with only one.

Capgemini Research Institute sees the same distinction in automotive product development, in a July 2026 survey of 800 executives from twelve countries. Of these, 68 per cent use or are testing agentic AI, while 45 per cent still have no strategy or roadmap for AI in product development. The group getting the most value develops 17 per cent faster. Capgemini attributes this mainly to a better data foundation, clear business cases and disciplined governance.

In manufacturing, Deloitte Switzerland surveyed over 140 companies of different sectors and sizes; the results were published on 30 June 2026. Four in ten use generative AI or agents. Of all AI applications, about one in five has been scaled across sites or the whole company. High implementation costs are the barrier cited most often, at 43 per cent, and 79 per cent see a risk that AI will disrupt ongoing production. Deloitte recommends cutting fragmented pilot portfolios back to fewer applications that deliver more value.

How solid is this? So far the clear-out is a prediction by a single firm, and the growth figures come from the first year of measurement. Nobody has yet counted the cancelled projects; that only starts in 2027.