Leadership in an Exponential Age

Herman Singh, CEO of Future Advisory and former group chief digital officer at MTN, Vodacom, and Standard Bank, joined Colin Iles for this Henley Leadership Series conversation on leading through agentic AI. Singh's sharpest point: agentic AI is not proof that machines have caught up to general intelligence, it is an admission that they haven't, and the boards and CEOs who understand that distinction will manage the risk far better than the ones chasing hype.

The iceberg and the sun: why change is hitting some parts of the business and not others

Singh frames organisations as icebergs and technology as the sun: the edges melt fast, the centre melts at glacial speed. Customer-facing functions, fraud detection, credit scoring, sales, and marketing are changing rapidly, while internal audit, compliance, regulatory reporting, and HR are moving far more slowly. He calls it a two-speed digital transformation, and argues most of what corporates currently brand as AI is really older analytical engines, credit scoring, fraud detection, anti-money-laundering systems, relabelled. Genuine agentic AI, he says, is the shift from rules-based logic, where business rules already exist and can be turned into software, to data-driven logic, where an agent has to infer the rules itself within a narrow domain because no flowchart or decision tree exists yet.

Klarna's 700 jobs: the corporate downside of moving too fast

Singh points to Klarna's rollback of roughly 700 customer service roles after replacing them with agentic AI as the cautionary tale corporates should study. Klarna, preparing for a listing valued at $30 to $40 billion, took a large customer and execution backlash from the move. His broader argument is about asymmetric risk: a startup's downside from a failed AI rollout is low, while a corporate's can be existential, which is why large organisations are sticking to sandboxes and point solutions with a human always in the loop. He cites the big four auditing firms investing hundreds of millions of dollars into building AI-auditing capability as the clearest signal of where the next profit pool sits, since AI systems hallucinate, carry bias, and are prone to jailbreaking, and every implementation needs an audit trail for when something goes wrong.

The three-horizon board model: where the investment should actually go

Singh lays out a framework he is encouraging boards worldwide to adopt for allocating AI investment across three horizons. Horizon one covers technology that is mature but new to the organisation, roughly 80% of board attention, governed by established good and best practice. Horizon two covers technology new to the industry and problems new to the market, about 15% of attention, where there are no textbooks yet, only emerging case studies of what other players are doing. Horizon three is technology and problems that are new to the world, 3 to 5% of attention, where there is no precedent to copy, and as Singh puts it, you are the one who has to write the book. He argues most organisations are right to weight their spend this way, but that horizon three, once a hypothetical, is now arriving inside a year or two.

Why fintech is already beating banks, and what boards should do about it

Drawing on Nubank's more than 100 million customers, Singh argues fintech has already overtaken traditional banking on every measure except assets under management, and that agentic fintech, leaner still, is next. He is explicit that the barrier to adoption is rarely technology itself: it is social engineering, the willingness of customers to change. He points to Standard Bank's internet banking launch, which reached only 50,000 mostly young, technical customers after a 12 million rand investment, and South Africa's first BlackBerry rollout in 2005, which produced almost no remote work uptake for 20 years until COVID forced 90% of employees home within a month. His conclusion: people change not because something becomes possible, but because it becomes mandatory, convenient, and urgent. He expects incumbent banks to eventually be broken up, since their retained customer base skews older and more risk-averse, while newer entrants will need to merge into universal banks to combine corporate funding access with retail deposit bases, something no digital-only bank has yet built.

Why judgment and creativity are the two words that matter for the next generation

Asked what young people entering the workforce should focus on, Singh gives two words: judgment and creativity. He illustrates judgment using aviation autopilot: planes fly under computer control roughly 90% of the time because every routine variable has been mapped into a digital twin, but control is handed back to a human the moment an unmodelled event occurs, such as engine failure, which is also why no plane is allowed to take off on autopilot. His advice is to go into roles where human judgment is either constantly needed or needed rarely but at the highest stakes, the person the AI escalates to. He also points to professions protected by law and professional bodies, such as law, auditing, and engineering, as durable because they require human sentiment and value judgments that legislation deliberately keeps out of automated systems. On creativity, he distinguishes generative AI, which recombines existing patterns (he compares it to a Monte Carlo simulation he once ran to find new configurations on a 50-year-old steel slitting line) from genuine creative emergence, which he says remains difficult for algorithms to produce.

Why is agentic AI different from earlier automation like RPA?

Robotic process automation solved formulaic, rules-based work, logging into systems and following defined workflows, and millions of RPA instances now run globally. Agentic AI is aimed at data-driven logic instead, where business rules do not exist yet and an agent has to infer them by analysing data within a narrow domain, rather than following a pre-built decision tree.

Will smartphones become obsolete?

Singh argues yes, over the long term. He points to Sam Altman's $6.5 billion acquisition of io, the hardware startup co-founded by former Apple design chief Jony Ive, as evidence of a coming post-smartphone device built around agentic AI that needs less bandwidth, storage, and processing power because agents bypass apps and app stores entirely, going directly to the source.

What should corporate boards actually be doing about AI risk right now?

Singh's advice is to build, test, and monitor rather than wait. Organisations should run experiments across the three-horizon model, keep humans in the loop on every implementation, and watch what peers, startups, and leading-edge companies are doing before committing capital, much like Formula 1 teams waiting to see who changes tyres first in changing weather before the rest follow at scale.