Rewiring workflows and forecasting capabilities are critical for AI

Business leaders need to consider how the technology may alter their business model a decade from now to make the right decisions today, experts say

November 17, 2026
A man working at a laptop.

Artificial intelligence has never been more in the news, dominating business and political headlines. Yet four years after the launch of ChatGPT, most companies are still struggling to generate significant returns from the technology. We gathered senior leaders recently at the Royal Institution in London for a conversation with Louis Mosley, CEO of Palantir UK, and Krzysztof Bar, CEO of the London Initiative for Safe AI, about how organizations can capture the benefits of AI while containing the risks.

Companies need to focus on rewiring workflows rather than automating jobs to create value with AI. It’s not surprising that only 12% of CEOs surveyed for the Oliver Wyman Forum’s CEO Agenda 2026 report cost savings or revenue uplift of more than 10% from AI. Giving AI tools to workers can make individuals more productive, but leaders need to reorganize their businesses around AI to capture its full value. It took decades for industry to redesign factories around electrification in the early 20th century, Mosley pointed out. He contended that the step change will come from new companies built around AI and from established enterprises when they fundamentally reshape the way they do business. By contrast, simply shedding jobs is difficult to do and risks losing valuable human intelligence. “The fundamental insight I would want business to really think about is workflow, not worker,” Mosley said.

Organizations should forecast how AI capabilities are likely to improve and build that into their rewiring plans. The AI frontier is jagged, with large language models having made dramatic gains in software development even as they remain less capable in other areas. Bar said CEOs should make explicit forecasts of where they expect AI capabilities to be in 12 months’ time and let that guide when, where, and how they rewire workflows. That can help take AI transformation outside of departmental silos and push it across the organization. Leaders also need to be looking further ahead and asking themselves how the business will make money in 10 years’ time and where its sustainable moat lies when intelligence becomes ubiquitous. “That forces everyone into a completely different mindset,” said Mosley.

Europe needs to move quickly on AI or risk a widening competitiveness gap. Both speakers expressed concern about what they perceive as a cautious, wait-and-see attitude by many European businesses toward AI, in contrast to the more rapid deployment in the United States. European companies shouldn’t let the risks of AI adoption blind them to the opportunity the technology offers, Bar said. The bigger risks for Europe are likely to arise from external AI developments, such as cyber incidents like the Hugging Face hack, but that’s all the more reason for European companies to experiment more with the technology, he said. As for the risks posed by AI agents, Mosley said the answer is to treat them like people. Humans are fallible and may lie and cheat, yet companies mitigate the risks through supervision, access controls, and other safeguards. They should take a similar approach with AI agents, he added.  

Sovereignty concerns may give a boost to open-weight models. Sovereignty is fundamentally a question of who owns the intelligence layer, but where people stand on the issue depends on where they sit. In the US, there are growing concerns that embedding frontier models more deeply into operations could enable hyperscalers to become direct competitors, Mosley said, citing Anthropic’s recent launch of Claude Design, seen by some as a rival to Figma. European business and government leaders, by contrast, worry more about excessive reliance on American AI models and infrastructure, and the risk that access could be turned off. Open-weight models can help address both sets of concerns. They’re good enough for most purposes, cheaper than frontier models, and companies can build their own control layer on top. “It’s more secure, much cheaper, and wholly sovereign,” said Mosley.