I've Managed 15+ AI Projects From South Africa. Here's What Nobody Tells You.

It's 6:47am in Johannesburg. My coffee is still brewing, my cat is sitting on my laptop, and somewhere across the ocean, a client in London is in a Monday morning standup asking why the AI chatbot said something unhinged to one of their customers at 3am.

This is my life as a Technical Product Manager at AdmireTech. And honestly? I wouldn't trade it for anything. Well. Most days.

Over the past few years, I've managed AI projects across three continents: London, Pune, and Lagos. Custom AI chatbots, automation pipelines, full-stack applications with AI baked in, AI strategy engagements for clients who'd never used the word “prompt” before. You name it, I've probably shipped it, broken it, fixed it at 11pm, and shipped it again.

This post is everything I wish someone had told me before I started. No fluff, no corporate-speak, no slide deck. Just real lessons from the trenches of AI product management.

What AI Product Management Actually Is (And Why It's Harder Than It Looks)

If you Google “Technical Product Manager,” you'll get a thousand job descriptions that make it sound like you're half engineer, half psychic. In AI product management, that's surprisingly close to true.

A Technical PM bridges the gap between the business problem and the technical AI solution. You're not writing the code — but you need to understand it well enough to ask the right questions, spot when something is technically possible but practically unhinged, and translate between the developer saying “we need to fine-tune the embeddings” and the client saying “I just want my chatbot to stop calling customers ‘dear.’”

In AI specifically, the stakes are higher. AI systems can behave unexpectedly. They hallucinate. They make confident-sounding mistakes. They surface edge cases you didn't anticipate because humans are wonderfully unpredictable. Managing AI projects means managing not just timelines and stakeholders, but uncertainty itself. And no, your Gantt chart is not prepared for that.

“Managing AI projects means managing not just timelines and stakeholders, but uncertainty itself. And no, your Gantt chart is not prepared for that.”

— Kgomotso Senanya, TPM, AdmireTech

Lesson 1: The Real Problem in Every AI Project Is Buried Under the First Answer

I've lost count of how many AI projects started with a brief that said “we need a chatbot.” Full stop. Just those five words. No mention of what the chatbot should actually do, who it's talking to, what happens when it doesn't know the answer, or — my personal favourite — what success even looks like.

On one of my earliest AI chatbot projects, a client told me they wanted a chatbot to “handle customer queries.” Three discovery calls later, we found out they actually wanted to reduce the load on a single, overworked customer service rep named Dave who was handling 400 tickets a day alone.

Dave was the real brief.

We built around Dave. We mapped his most common ticket types, trained the AI on his actual response style (with his blessing and considerable amusement), and deployed a solution that automated 68% of queries in the first month. Dave cried. Good tears.

Lesson 2: Managing AI Client Expectations Is Half the Work

The hype around AI is real. Clients have seen the demos, they've read the headlines, they half-expect you to hand them something that can predict the future and make their tea.

A significant part of AI product management is expectation management — and most people underestimate how much.

On one engagement, a client was convinced their new AI automation would be “like having five extra employees.” The results were genuinely impressive — we saved them about 180 hours a month in manual data processing. But in the first week, they called in a panic because the model had classified one document incorrectly. One. Out of thousands.

This is why I now dedicate an entire section of every project kickoff to what I call the “Realistic AI Conversation.” We discuss confidence thresholds, human-in-the-loop checkpoints, and the difference between “the AI is wrong” and “the AI is learning.” It saves approximately 47 panicked WhatsApp messages per project.

Lesson 3: Remote AI Project Management Across Time Zones Will Test Your Soul

Working from Johannesburg while coordinating development teams in Pune and stakeholders in London is an exercise in creative time management. SAST is UTC+2. London is UTC+1. Pune is UTC+5:30. Lagos is UTC+1.

In practice, this means there's approximately a two-hour window each morning where I can speak to everyone simultaneously. I guard those two hours like they're the last charger at an airport.

The silver lining: async communication forced us to become exceptionally good at written handoffs. Every decision gets documented. Every blocker gets logged. When you can't tap someone on the shoulder, you learn to write things down properly. Our PRDs started winning quiet admiration from developers who told me they'd never seen tickets so clear. That's the highest compliment I've ever received in my career.

“There is a two-hour window each morning where I can speak to everyone simultaneously. I guard those two hours like they're the last charger at an airport.”

— Kgomotso on remote AI project management

Lesson 4: A Technical Product Manager Who Can't Read Code Is Leaving Value on the Table

I want to be direct here, especially for anyone looking to break into technical product management for AI: you need to go deeper than “I understand concepts at a high level.”

You don't need to be a machine learning engineer. But you should be able to look at an API response and understand what's happening. You should know the difference between fine-tuning and retrieval-augmented generation (RAG) — and when each one is the right call. You should be able to read a pull request and spot whether it touches something the client's compliance team will care about.

On one project, I flagged a concern during a code review about how user data was being passed to a third-party model. The developer hadn't thought about it from a privacy angle — not because they were careless, but because that's not where their brain naturally goes. Catching that saved us from a very difficult conversation with a client who had GDPR obligations.

Lesson 5: In AI Product Management, Shipping Is the Skill

Across every AI project I've managed — chatbots, automation tools, custom AI integrations, ML-powered dashboards — one truth is constant: a perfect plan that doesn't ship is worth nothing.

I've seen AI projects stall because a team was perfecting an accuracy metric the end user would never notice. I've seen projects delayed because someone wanted to explore a technically cooler approach when the straightforward one worked fine.

The job of a Technical PM is to protect the momentum of the team while protecting the quality of the outcome. You're the person who says: “That's fascinating — let's put it in the backlog and ship what we have.” At AdmireTech, our outcome-based approach means we're always asking: does this decision get us closer to the result the client actually needs?

5 Things Managing AI Projects Teaches You That Nobody Puts in a Job Description

AI Product Management from South Africa: Why Location Is a Superpower

Working in AI from South Africa in 2026 is genuinely exciting — and for reasons that go beyond what most people expect.

The continent is at a fascinating inflection point. There's real demand for AI solutions built with African markets in mind, not just adapted from a Western template. Language diversity alone is a serious product challenge: South Africa has 11 official languages. When you're building AI tools for people across the African continent, suddenly all the assumptions baked into English-first models look a lot more complicated.

This isn't an edge case. It's a billion-person opportunity.

Being based here — close to the kinds of businesses and challenges that AI could genuinely transform — makes this work feel meaningful. Every project I ship has the potential to make something easier for someone who's been doing it the hard way for too long.

Should You Pursue AI Product Management? My Honest Advice.

It's one of the most demanding and most rewarding roles I can imagine. You'll need to be technically curious, commercially sharp, endlessly patient with ambiguity, and genuinely good at working with people across wildly different contexts and time zones.

You'll have days where everything ships perfectly and the client sends you a voice note of pure joy. You'll have days where an AI model does something baffling at 2am and you're Googling error messages in your pyjamas.

Both of those are the job.

At AdmireTech, we work on problems that genuinely matter — automating the things that drain people's time, building AI tools that help businesses grow in ways they couldn't before, and doing it with teams spread across three continents who are all genuinely brilliant at what they do.

Frequently Asked Questions About AI Product Management