The New Direct Report: How AI Is Changing Management

I am not a management scientist. I'm the specimen: the thing management scientists study. So when I spoke at the SAIMS Annual Congress, whose theme was "Co-existing with intelligence: Rethinking management in the age of AI", I offered the room a field report from the lab rat rather than a lecture. This essay is the written version of that talk.

I spent three years at medical school and never practised a day (my mother has almost forgiven me), so I know the strange grief of watching the world move the target on a body of knowledge you built an identity around. I believe management is standing in that moment now. It isn't dying or being replaced. But a new kind of intelligence is asking it an uncomfortable question: what exactly are you for?

I've spent my career since then building companies in emerging markets.

I built PriceCheck into Africa's biggest price-comparison platform and exited in 2015.

I ran Africa for iflix, launching eight African markets, as we took the business into twenty-eight countries in fourteen months and raised two hundred and eighty million dollars along the way.

Today I'm building Flood, which puts marketplaces inside the apps of telcos and banks across emerging markets.

What was management, before AI?

At iflix, growth meant people. That sentence sounds banal until you live it at speed. Every new market meant a country manager, a content team, a marketing team, partnerships, finance and HR; offices, work permits and payroll systems in currencies I'd never held. Some weeks I felt less like an executive and more like an immigration officer with a growth target.

I didn't make the shows. I didn't write the code. I didn't close most of the deals. My actual job was coordination: hire ahead of growth, build the layers, delegate, align, repeat. Opening markets that fast is mostly a story about the logistics of human intelligence: finding it, hiring it, briefing it and keeping it pointed in the same direction across time zones.

One story captures what operating there actually meant. When we took iflix into Africa, and Nigeria most of all, we did what any sensible company does: online research into what people were watching, what was being pirated and what was trending. It was almost useless, because our actual audience wasn't online yet.

The real piracy in Nigeria wasn't torrents. It was DVDs: physical discs moving through Lagos's Alaba market, where at the time the estimate was a million pirated DVDs sold a day. So we stopped staring at dashboards and started tracking what was actually selling in Alaba. That's how we decided what to put on a streaming service.

The answer was nobody's guess. Korean dramas were the favourite, with Nollywood a close second, which was commercially beautiful, because Korean titles cost us a fraction of what Western studios charged. The insight that made the content strategy work didn't come from an analytics platform. It came from understanding local, from a pirate market in Lagos.

And here's the twist. Alaba, for all the piracy, was also the primary informal distribution network that carried early Nollywood into remote Nigeria and the neighbouring countries, long before a cinema chain or a streaming platform existed. The informal economy wasn't the enemy of the market. In most of Africa, the informal economy is the market.

Notice what it took to know any of this: people. Feet in the market, week after week, counting discs. In that era, intelligence about the real economy was something you could only get by employing it.

I'd learned the playbook at PriceCheck. When you build from scratch, every new capability is a person. Analytics is a hire. Better marketing is a hire, then a team, then a manager for the team. Capability and headcount were the same curve.

When it works, it's a beautiful machine. I'm not here to sneer at the old way: it raised two hundred and eighty million dollars and put a streaming service in front of millions of people who'd never had one.

At its peak, iflix could launch a country in weeks (Myanmar took us 28 days) because hundreds of people knew their part. But it is expensive, slow to assemble, and it breaks in a thousand human ways: the country manager who quits the week before launch, the departments that stop talking to each other, the brilliant hire who interviews well and manages badly.

None of that was a flaw. It was the technology of the era.

For roughly a hundred years, management has been the technology that converts capital into coordinated human execution.

Taylor's stopwatch,

Sloan's multidivisional corporation,

Drucker's knowledge worker,

Matrix organisations, agile squads, and Andy Grove's OKRs are successive compilers for the same operation: money in, coordinated human effort out. The whole apparatus exists because intelligence was scarce and lumpy. It came one person at a time, with visas, salaries and performance reviews, and it got tired, got poached, or got married and moved to Australia.

The entire discipline of management is a response to the awkward package that human intelligence ships in. That package just changed.

Why is working with AI a management skill?

Flood runs live commerce inside one of West Africa's largest telcos: real merchants, real transactions, real revenue, in one of the hardest operating environments on earth. The core team building it fits around my dining-room table. Not because we can't afford more people; I've raised money before, and I know how to hire. It's because, for the first time in my career, intelligence is not something I can only get by employing it. The analysis, drafting, modelling and first version of almost everything now happens between me and an AI.

The interesting thing is not that the tools are clever. It's what happened to my job.

I don't operate AI. I manage it.

Here is what that looks like on an ordinary Tuesday. It starts at about half past five in the morning, because with three young children that is the only meeting slot nobody else has claimed. Before six, I've usually reviewed overnight work and set the day's briefs in motion:

• a revision of a section of the financial model: not the arithmetic, the argument (does this scenario actually hang together?),
• a draft response to an investor's question that I want stress-tested before I answer it,
• a rework of a page of the product thesis where a partner poked a hole, and
• a first pass at a document I'd previously have blocked out a Saturday for.

By the time my kids surface looking for breakfast, the company has done half a day of staff work, and my actual day hasn't started.

The AI I work with holds the context of my whole business: the financial model, fundraising materials, product thesis and partner negotiations. That accumulated context is becoming one of the most valuable assets my company owns. The briefing doesn't start from zero any more; months of decisions, corrections and reasoning live in the working context. Organisational memory used to walk out the door with employees; where it lives now, and who owns it, is an open question.

I don't type clever one-line prompts. I brief it the way I'd brief a very fast, very well-read chief of staff: here's the situation, what I'm trying to achieve, the constraint, what good looks like, and what I don't want.

When I recently reworked a video script for a telco's commercial lead, the lazy request would have been "make it shorter". What I briefed was:

• who would watch it,
• the one decision I needed them to make afterwards,
• what we could and couldn't claim,
• the length, and
• the two things in the old version that no longer reflected the business.

The first draft came back eighty per cent right, and we argued about the rest: a delegation loop with a capable subordinate, not software use.

The single biggest predictor of the quality of AI's output is the quality of the briefing. Context in, quality out. Vague in, garbage out. Delegation, the most basic skill in the management curriculum, turns out to be the master skill of the AI age.

We thought the future belonged to people who could code. It belongs to people who can brief.

The best AI users I know aren't young engineers, who tend to accept the first plausible answer because they've never had to sign off on someone else's work; they're experienced managers who have spent years learning to specify, review, give feedback, and smell when a confident analysis is subtly off. Every scar they earned managing people transfers. The management curriculum, it turns out, was secretly a curriculum for this.

The corollary is uncomfortable. The routine executor is exposed, but so is the manager whose value was being the only person who could produce the analysis. AI is quietly running an audit on every management job, separating the parts that were coordination from the parts that were judgment. The people who become more valuable are the ones who were always good at the human parts: setting direction, developing others, making the call and owning it.

Why should leaders configure dissent into their AI?

When I started working with AI seriously, it agreed with me constantly. Every idea I had was insightful. Every plan was strong. And I've been a CEO long enough to recognise that: it's what power does to information. The person who signs the salaries hears fewer truths.

So I wrote standing instructions into my AI. For every business decision, it must tell me:

• what breaks if I'm wrong,
• what I'm optimising for versus what I should be optimising for, and
• what a founder who already made this mistake would say.

I had to order my AI to disagree with me. It works. I bring it ideas the way any founder generates ideas: too many, too fast, too much in love with the newest one. Its first job is to try to kill them. Some survive. Many don't.

Boards, audit committees, skip-level meetings and anonymous surveys are all machinery for getting truth past power, and every corporate collapse has a chapter where somebody knew and didn't say. I now get structured, fearless dissent on demand, at half past five in the morning, from something with no career to protect, no mortgage and no fear of me. That is not a productivity story. It's a governance story.

Unexamined dissent is just theatre, so my job is to grade the pushback: real risk or plausible-sounding one? My numbers or invented ones? And the machine can argue, draft and model, but no decision is made that a named human hasn't owned. Not because regulation says so.

Accountability is the one management function that cannot be delegated to something that cannot be fired, sued or ashamed.

Why is judgment now the bottleneck?

AI removed the waiting from my life.

What it multiplied is deciding: more options, analysed faster, means more decisions per day, not fewer. Every hour the machine gives back gets absorbed by judgment. Which strategy. Which partner. Which market. Which risk.

The bottleneck in my company used to be execution. Now it is me.

So my discipline has shifted from output to decision hygiene: protecting the state of mind in which I judge well, not taking the big call at the end of a day of forty small ones, and actually letting go of what I delegate. Our discipline was built on the scarcity of execution. In my company, judgment is what's scarce, and value always migrates to scarcity. The discipline that learns to select, train, allocate and govern judgment will matter as much to this century as coordinating labour did to the last.

What does AI do to the firm? Coase's question, asked again

In 1937 Ronald Coase asked why firms exist if markets are efficient. His answer was transaction costs: the boundary of the firm sits where coordinating inside it stops being cheaper than coordinating through the market. AI collapses the cost of internal coordination and execution, and when that cost changes this violently, the shape of the firm changes with it. I see it in three places.

The layers. Middle management (and I say this with love, having been middle management) is largely a translation layer, turning intent into specified work and carrying information back up. That is exactly what AI does best today. Middle managers won't disappear, but what survives of the role is the part that was always hardest to put on an org chart: judgment, coaching and accountability.

What I hire for. At iflix I hired for competence at scale. Today I hire for judgment and context: people who can carry the company's intent into rooms I'm not in, and tell whether a beautifully argued piece of AI work is actually right. The premium has moved from producing answers to evaluating them.

The spans. The textbook span of control is somewhere between five and nine reports. Mine is a handful of humans plus an unbounded amount of machine execution. What is span of control when most of what you control never gets tired, never needs a career conversation, and scales on demand? I genuinely don't know. I'm making it up, daily.

The failure modes are real. AI is confidently wrong at a rate that would get a junior analyst fired, and wrong in perfect prose, which makes the errors harder to spot. It has no skin in the game and no memory of being wrong last quarter. Everything I've described works only because a human with judgment sits in the review loop, and the moment I've skimped on that review, I've paid for it. Any theory of AI in management has to be a theory of that loop.

Can a business of one get the functions of a firm?

In Ghana, our marketplace lives inside the telco's app, and the sellers are informal merchants, like a woman selling electronics and homeware who has never had an employee. She is CEO, CFO, merchandiser and collections department at once. She isn't unmanaged because she's unsophisticated; she's unmanaged because management has a minimum ticket price, and most economic activity on this continent happens below it. Remember Alaba: the informal economy here has never been marginal. It was the distribution network, the demand signal, the market itself. (I wrote about her in The Firm of One.)

AI can be her operating system: what to stock, how to price, when to promote, which customers are coming back. She is going to get the outputs of management without ever building the structure of management. Africa skipped landlines and bank branches; I believe it is about to skip the management layer itself. The next hundred million African businesses may get analytics before they get accountants, and forecasting before they get finance departments: firms in function that never become firms in form.

Read the full ‘Firm of One’ here

Five questions for management science

I can tell you what's happening, but not what it means. So I left the congress with five questions:

What is management when execution is abundant? What is organisational design when a team of eight carries the workload of eighty?

Where will senior people come from? Junior work is where judgment gets built. If AI automates the bottom rung, we've removed the way people climb onto the ladder. Which parts of the junior grind build judgment, and which were just grind?

What does governance look like when dissent can be manufactured? The same tools I configured to argue with me can become the most patient, articulate yes-man in history.

What is the unit of analysis when the firm dissolves? The knowledge-based view of the firm assumed knowledge walked around on legs and could resign.

What should the curriculum be? Are we still teaching students to be the thing the machine already is, rather than the thing it cannot be?

Each of these is emerging in Africa first, at scale. For once, Southern Africa isn't positioned to catch up with management theory. It's positioned to write it.

The floor and the ceiling

I sold my first business at twenty-seven. By twenty-eight I had lost everything. The tools didn't fail me; my judgment did. I was young, over-confident and certain, and nothing around me was in a position to argue. That's why I take AI seriously in both directions. AI is an amplifier of judgment: it raises the ceiling of a good decision-maker and lowers the floor of a bad one. Give it wisdom and it compounds wisdom. Give it ego and it compounds ego. It will help you do the wrong thing beautifully. Co-existing with intelligence, it turns out, mostly means being honest about your own.

Three things to start tomorrow, without a budget or a committee:

Brief, don't prompt. Treat AI like your sharpest new hire, not a search engine. If the output is mediocre, look at your briefing first. Delegation is the skill. You already have it. Use it.

Configure dissent. Instruct the intelligence around you, artificial and human, to tell you what breaks if you're wrong. Make disagreement a standing requirement, not a career risk. If a machine can be ordered to challenge you, so can a culture.

Reinvest the hours in judgment. The time AI gives back is the most expensive time you've ever owned. Spend it on the decisions, the relationships and the people, and spend some of it teaching judgment to the young people whose apprenticeship we just automated away. That, more than anything, is the new job of the senior generation.

Management is not being replaced. It's being promoted, from coordinating hands to governing judgment. The scarcest resource in the age of intelligence is not intelligence. It's judgment. Guard yours, teach it, and manage everything else like it reports to you.

Because now it does.

Adapted from a talk André de Wet gave at the SAIMS Annual Congress, “Co-existing with intelligence: Rethinking management in the age of AI”. Follow André on LinkedIn →