There are technology events you attend, and there are technology events that stay with you on the drive home. The Ikigai Industry Nights panel I had the pleasure of moderating on the evening of the 2nd July 2026 — provocatively titled CTRL + ALT + HUMAN — was firmly in the second category, and I have been turning it over in my head ever since.

I opened the evening with a number I find genuinely difficult to get past. ChatGPT reached 100 million users in two months — the fastest consumer technology adoption in human history. Spotify, for context, took eight years to do the same. Whatever this moment is, it is not business as usual. And when I asked the packed room at Ikigai who among them uses AI, virtually every hand went up. That, right there, told me everything about where this conversation needed to go: not another recitation of the hype, but what I like to call practical AI — how stuff actually gets done, right here in Kenya. 

The Panel

I could not have asked for a better trio to unpack it with, because each brought a genuinely different vantage point:

• Shikoli Makatiani, CTO of Akili AI, is one of the ‘OGs’ of Kenyan technology — close to three decades building and implementing enterprise systems across East Africa, with a deep sideline in cybersecurity and forensics, now building AI systems for banks, SACCOs and enterprises.
• Victor Ambuyo heads growth at Madavi and operates at the cutting edge of how organizations actually adopt — or quietly fail to adopt — new technology. 
• Marvin Oyoo, a management systems and information security consultant at Panoramic Synergy, brought the standards-and-frameworks discipline that AI conversations usually lack. 

What the Hype Gets Wrong

I put the same opening question to all three: what does the AI hype get most wrong, and what is the one shift every organization in this room must make in the next 12 to 18 months? The three answers, laid side by side, essentially wrote the evening’s thesis.

Victor went first, and went straight at the industry’s obsession with tools. Everyone, he observed, is talking about Claude and Gemini — and almost nobody is talking about the organizations and the people who are actually supposed to use them. Inside real companies, people don’t know which tool they need, or how to implement it — the conversation orbits the technology and neglects the humans it is meant to serve. Marvin took aim at the replacement narrative:

“AI is not here to replace people — it’s here to replace repeatable tasks. The real question is: are we, as leaders, prepared? Do we have a strategy to embed these tools into our workflows so they actually fulfill a business need?”

Shikoli delivered an eye-opening perspective drawing on over twenty years of walking into organisations to implement businesss technoology systems:

“When it comes to actually getting AI into organizations, only 10% of the work is AI. Ninety per cent of the work is everything around it.”

His example was a client who asked Akili to automate their loan processing. The team’s first request was simple: fix one specific part of the process. The act of mapping it revealed that the AI could automate the whole process end-to-end and create an impact much wider than the one the client first envisioned. 

In Shikoli’s words, building without understanding the wider context means the AI would do its job but it probably wouldn’t address the underlying problem. Take the time to see what is actually needed first; then, and only then, point the technology at it.

Why Adoption Fails: A People Problem Wearing a Technology Costume

The stretch of the evening that had heads nodding hardest was Victor’s anatomy of a failed AI rollout — a story he says Madavi has watched repeat for two years. An executive attends an AI event, much like this one. Gets inspired. Signs up for subscriptions on the spot. Comes back and announces to the team: we have this new tool, everyone should use it. Six months later, adoption is below 20%. His diagnosis?: leadership was sold the idea but never models the usage themselves; nobody is given ownership of making it work; and — the part almost nobody addresses — the workforce is quietly afraid, because the tool appears to do the very thing they are paid to do, so adopting it feels like signing their own redundancy letter. Layer on wildly varying technical abilities across the team, and you get inconsistent usage, pretend usage, and eventually no usage. His conclusion is a simple explanation: it is a people problem wearing a technology costume.

Marvin gave the same disease its clinical name — the reality gap, the distance between how excited an organization is about AI and how ready it actually is — and then demonstrated it live. He asked the room, by show of hands, who was conversant with terms like SLM, LLM and context windows. In a room where every hand had gone up for ‘who uses AI’, only a scattering went up now. Most organizations, he argued, are extracting perhaps 5% of what these tools can do, because they never established what exactly they have. His test for AI maturity is candid:

“I ask organizations: do you have an AI strategy? And they tell me — we have Copilot Enterprise, we have Claude for Enterprise. That’s not a strategy. That’s a subscription.”

A strategy, in Marvin’s telling, is leadership direction: objectives, the business needs you are solving, the outcomes you expect — and clarity that AI is not an IT project but everyone’s responsibility. He added a warning from his information security practice that before the governance discussion: without that direction, people ‘abuse’ AI daily — uploading sensitive company information into free tools as their only interaction with the technology.

The Jobs Question, Answered Three Ways

Of course, the question everyone came to ask is the jobs one, and I put it to the panel directly: are we automating jobs or augmenting people — and which skills gain or lose their premium? What made the answers memorable is that all three refused the comfortable script, each in a different way.

Victor’s answer centred on what these tools actually are: pattern machines, powerful in proportion to the volume of data behind them. Which is why the human skill that appreciates most is judgement — the discipline of not taking output at face value, of being the person who decides whether what came back is actually what was needed. In Kenya specifically, he argued, one deeply local advantage endures: trust. Ours is a relationship-based market — the ‘I know a guy’ economy — and no model replaces the human relationships deals are actually built on. When I asked him whether he was endorsing the cliché that ‘AI won’t take your job — someone using AI will’, his reply was to the point:

“It’s not the AI itself. It’s someone more fluent with AI — because they’ll do the job faster and better using the tools of the time.”

Shikoli, to his enormous credit, refused to pretend nothing disappears:

“There used to be a job called lamplighting — someone went around the streets switching on the lights. Then electricity came. Let’s not cheat ourselves: AI will erase some jobs.”

Some work is pure labour, and pure labour is what this technology consumes — he expects driving to be a dying activity within five years. But citing Wharton’s Ethan Mollick, whose book Co-Intelligence is being followed by one titled Coexistence, he offered the perspective I found most useful all night: we are moving from collaborating with AI to coexisting with it — machines taking the tasks, humans owning the judgement. However, when it comes to decisions touching one human lives, stay with humans…. Indeed, he conceded that if he were dying alone in a forest with nothing but his phone and his AI told him to eat a particular leaf — he would eat the leaf. The room laughed!

Keeping the Machines Honest: Synergy, Checkers and a Council of AIs

If Victor owned the people question and Shikoli the builder’s view, the governance part belonged to Marvin — it was quite possibly the most important part of the night. He rejects the ‘human versus AI’ argument outright in favour of human–AI synergy: the machine gets you there faster; a human checker validates that the result is real, because — as he put it — our AI is only as good as the data we feed it, and hallucination and data poisoning are not theoretical risks. From his own field he offered the perfect illustration: the security operations centre, where analysts drown in alert fatigue, and a well-trained AI agent triages the flood so humans investigate only what matters — machine speed, human judgement, working one problem together.

On transparency, consent and bias, his standards-practitioner instincts showed. You cannot, he argued — nodding to Shikoli’s earlier point about policies written for humans — hand a fifty-page loan policy to a model and expect it to navigate ethics, morality and inclusivity unaided: a machine thinks the way its platform allows it to think. His practical prescription was almost mischievous: borrow from the YouTuber Felix Kjellberg (PewDiePie), who runs an ‘AI council’ — multiple models answering the same question and judging one another, with a human reading the disagreement before deciding. Models checking models, and a human above them all. Somewhere in that anecdote is a governance framework waiting to be formalized.

Shikoli’s hospital example completed the picture of what mature deployment looks like: a Kenyan hospital where AI now interprets a patient’s insurance cover in real time — collapsing a three-to-five-hour back-and-forth into an instant answer — while the doctor, and only the doctor, decides what to prescribe. Speed where the machine excels; a named human wherever judgement carries consequence.

What’s Actually Possible: Tea Drones, ATM Footage and a Physics Exam

The eye-opening stretch was Shikoli’s insight on AI’s most underexploited superpower in Kenya — not text generation, but vision. “AI could stop developing today,” he said, “and we would still never fully use what it can already do.” Tea buyers are using footage from the drones that spray smallholder farms to assess crop quality — and buying the best tea three to four months before it reaches auction. From his forensics days: where a bank fraud investigation once meant a human combing through three hundred videos of ATM footage, AI now flags the moment that matters in minutes. He then brought it home: his daughter was able to pass a physics exam using AI as a patient tutor through the Feynman technique and the Socratic method. An organization, he reminded us, is just a collection of workflows; workflows are a collection of tasks; and AI comes for tasks — his parting provocation being that the next billion-dollar company might well be built by a solopreneur.

My Key Takeaways

• The 10/90 rule is the whole game (Shikoli). The technology is the easy part; process, people, policy and ownership are the work — and if the process is broken, AI merely accelerates your problems.
• Adoption is a people problem wearing a technology costume (Victor). Leadership must model usage, assign ownership, and address the fear directly — or watch adoption stall below 20%.
• A subscription is not a strategy (Marvin). AI is not an IT project; it needs leadership direction, objectives and outcomes — and most organisations are using 5% of what they already pay for.
• Fluency beats fear (Victor). It’s not the AI that takes your job — it’s the person more fluent with it. Judgement and trust are the skills that appreciate; in relationship-driven Kenya, the human premium is real.
• Be honest about jobs — then talk coexistence (Shikoli). The lamplighters are gone; the useful conversation is which tasks go to machines and which judgements stay human.
• Keep a human checker on everything that matters (Marvin). Hallucinations and data poisoning are real; synergy means machine speed with human verification — up to and including a council of AIs judging each other.
• You cannot govern — or benefit from — what you cannot see (the whole panel). Every thread of the night led back to visibility: of your processes, your decisions, your data, and what AI is already doing inside your organization.

My sincere thanks to the Ikigai team for a superbly curated evening, and to Shikoli, Victor and Marvin for showing up with substance rather than slides — three genuinely complementary perspectives that kept a packed room engaged deep into the night. This is exactly the calibre of conversation Kenya’s technology ecosystem needs more of when it comes to all things AI, and I suspect a part two is inevitable!

Listen to the Podcast: