EP32: The Biggest Risk in Business: Falling Behind in the Agentic AI Revolution with Simon Torrance

Reframing AI Risk: The Danger of Standing Still

After running an AI think tank for a group of insurance companies, Simon founded AI Risk specifically to help traditional businesses harness the most advanced forms of AI. While risk can be broken down into technical, compliance, operational, and financial categories, Simon's central argument is that the single biggest risk today is not adopting agentic AI quickly enough. He frames agentic AI not as a tool, but as the foundation of a new kind of digital workforce — one capable of scaling at near-zero marginal cost. A competitor with an AI-driven workforce several times the size of yours, within just a few years, represents a genuinely existential competitive threat.

Why Most Companies Are Still at "Table Stakes"

Simon describes an AI maturity model, moving from rigid, pre-GenAI robotic process automation, through today's widespread adoption of generative AI as a personal productivity tool — Copilot, ChatGPT access — which he's blunt about: this is now simply table stakes, offering no real competitive advantage, since virtually every competitor has access to the same tools. True differentiation, he argues, comes from agentic AI: autonomous agents capable of completing end-to-end tasks that previously required human judgement, at a fraction of the cost of hiring people, and with the ability to be replicated almost infinitely.

Case Study: An Insurance Team Run Almost Entirely by AI Agents

The standout example Simon shares is a publicly listed company that launched an entirely new insurance product line — staffed almost entirely by AI agents, rather than human hires. Structured around a theory called "artificial life" (the idea that systems built from many small, specialised "brains" often outperform one large, centralised system, much like a swarm of bees or flock of birds), the team consisted of distinct agents — a customer service agent, a GDPR specialist, a pricing and underwriting expert — coordinated by a facilitator agent, all collaborating on Slack, just as human colleagues would.

After roughly a year of refinement, including quality thresholds before any decision was actioned, human oversight was no longer needed. Over 18 months, the AI-run team achieved Net Promoter Scores around 25% higher than comparable businesses, record underwriting profit, and significant hiring cost savings — while operating 24/7 with no breaks. Notably, the agents occasionally proposed profit-maximising strategies that were ethically or legally questionable; the fix wasn't complex retraining, but simply providing the team with a clear values document, which they adapted to quickly.

Digital Twins: A New, Complicated Frontier

One of the most thought-provoking elements of the case study is the creation of "digital twins" — AI agents modelled on specific human employees' prior emails and Slack contributions, built to let those employees' expertise keep pace with agent-speed collaboration without requiring their constant, real-time presence. Employees who agreed to be "twinned" were paid extra for the added value their digital counterpart contributed. Simon is candid about the unresolved questions this raises: who owns a digital twin if the employee leaves the company? Should an individual continue to be compensated for a twin's ongoing use after they've departed or retired?

A Strategic Blueprint, Not an IT Experiment

Simon is emphatic that agentic AI transformation must be led from the top, not delegated to IT as an isolated experiment. His recommended approach starts with a strategic question: what could the business do with, say, four times its current operational capacity within three years? From there, leaders identify genuine pain points where agentic capability could deliver material short-term impact, building a coordinated portfolio of activity — rather than the scattered, disconnected pilots Simon says are common, and which rarely scale or meaningfully affect the bottom line.

He notes that leadership attention varies significantly by company size and individual disposition — smaller, visionary-led scale-ups tend to move fast once shown what's possible, while larger corporations depend heavily on whether leadership is thinking in genuinely long-term strategic terms, versus managing toward a shorter-term exit or handover.

Realistic Timelines and the Danger of Disconnected Pilots

Simon suggests meaningful, tangible value can often be delivered within around six months — citing a current client target of responding to 100% of customer emails within an hour using AI agents, up from a baseline of roughly 5%. However, he cautions strongly against random, disconnected pilots that never reach scale — a common pattern behind CEOs complaining that AI spend isn't affecting the bottom line, which Simon attributes to a combination of not using genuinely agentic AI, and lacking a coherent overarching strategy that includes change management, reskilling, new metrics, and legacy IT integration planning.

Managing Genuine Risk Without Being Paralysed by It

On cybersecurity and compliance risk specific to agentic AI, Simon advocates building a formal "controls layer" within enterprise architecture — ensuring compliance, security, and privacy are properly addressed, and that systems remain auditable. He highlights that because the insurance case study's agent interactions ran entirely through Slack, the entire decision-making process was fully logged and auditable — arguably making the system easier to demonstrate compliance for than a purely human-run equivalent.

The Future of Work: Augmentation, Not Just Automation

Asked about the broader societal impact on jobs, Simon is candid that significant white-collar disruption is likely, drawing a direct parallel to the earlier hollowing-out of manufacturing employment through automation. He argues government urgently needs a clearer understanding of agentic AI's potential workforce impact, and to invest proactively in retraining and education — rather than reacting after disruption has already occurred.

He offers a memorable reframing of future workforce skills: rather than simply being competent with AI tools — now itself a baseline expectation — real competitive advantage will come from individuals who can bring their own team of AI agents to the table, effectively multiplying their own capability and value.