Before the dust settles
An Introduction into AI agents for Africa
Africa’s possibly thousands-strong agent networks have been the story of distribution for everything from pharmaceuticals to call credits (airtime), micro-finance, digital finance, agricultural extension services and even loan disbursement and recovery services. In this continent, agents are jobs, cash comptrollers, product distributors and boots-on-the-ground “executives”.
I’ve been writing more and more about agents as the unique distribution fabric for Tech 1.0 and 2.0 in Africa. And I will be writing more about this, especially now that my experiment with trendsaf (commodity data for trade) is closer to going live. Please reach out if you’re in Kenya and can help us build an agro-dealer network for data processing.
With that said, these days agents mean something different. Especially as the blitzkrieg assault of artificial intelligence models on traditional software continues abated.
Njavwa Mutambo, founder of AI startup, Caantin, is a hardy entrepreneur who certainly has “agency”. For the uninitiated, “Agency” is the trendy term used especially by tech twitterati to describe people who have both the ambition and chutzpah to “entrepreneur” try to take control of something, and bend the(ir) world to their will.
I first met Njavwa in person some 21 months ago, at his then apartment in Nairobi.
Caantin was still a rebrand plus pivot from TopUp Mama, and I had written about it for TechCabal. He had a few friends over, and he invited me. Over wine, (and juice, since I don’t drink), some pastries and a documentary video of Picasso painting that was projected onto a wall, we had a good authentic conversation that night.
I remember leaving with an impression that the man fit the stereotypical startup founder, intensely trying to chase down and catch something. It was Caantin then, but from his story, it wasn’t always Caantin.
Since then, the brooding has developed into an obsession with transforming sales automation and now AI sales and customer support. Along the way, it has transformed Caantin from a restaurant and hospitality ERP into a product for sales automation focused on American business, and further transformed Caantin into the voice AI for customer support operations that it is today.
Ten days ago, Njavwa messaged me with a link to a short piece I had published in November last year for Briter and Mercy Corps AGRIFIN on agent networks in agriculture. “What do you think will tip agent networks to scale? More bundled services? Finance partnerships?” And then, “What’s the role of AI agents in this whole mix?”
I suspect Njavwa has an answer already. After all, he’s talking to dozens of potential customers and flying across the continent to close large enterprise deals. He gets to sell what Caantin AI can do, and see how his customers use it, or find new uses for which even he couldn’t think about.
But his question forced me to think – and reply eight hours later.
What will AI agents mean for Africa?
Since then, I’ve been paying more attention to this new-ish vector in the artificial intelligence blitzkrieg. I saw new-ish because some of what is being lumped under the agentic AI label are things that have existed before that are now being repackaged.
Honest-to-goodness agentic AI, are on my radar precisely because they take much of AI discourse out of the realm of boring benchmark model performance, and into enough of the realm of semi-autonomous capabilities, where it begins to mean something to users outside of the computer-device—even without them knowing.
Some things are obvious. But a lot of it is still hazy. And it will take a bit more listening to unravel. As with many things, the What, Why, How, Where and When framework is a useful approach to understand what is happening and what it means for us all today and tomorrow.
What do we know now?
#1. That AI models are commodities
We’ve known for some time that Sam Altman’s AGI spiel was to be taken with a shipload of salt. However, until DeepSeek, we did not understand the extent to which AI models would become extremely commoditised. We now know enough to know that incremental improvements will likely not be in leaps and bounds. So we’ve seen Satya Nadella go from extreme bullish on AI (and investing billions in one of the weirdest corporate structures to date) to recently dissing AGI and AI model benchmarkism as a metric for progress.
#2. AI for enterprise is niche and incredibly verticalised
Good examples of this include Thompson Reuters’ Westlaw Edge AI, recent drug-making breakthroughs, and so forth.
#3. Foundation models are still important a là DeepSeek and Grok.
But it is becoming more and more important for the AI competitors to underline the specialised use case capabilities of their base models. Hence, Google and Gemini is underlining Deep Research, Anthropic is underlining the coding capabilities of Claude. And DeepSeek is doubling down on being open and recently launched a new file system (pointing to a direction of controlling its infrastructure ecosystem at even deeper levels.
#4. The 20-dollar price point is becoming standard, especially figuratively
Most subscription prices for AI products start from $20. But that is only a pointer to the fact that in addition to everything, everyone is more conscious of cost, user interaction design and how to get people to use it, even as AI enters its blogspot era.
What will AI agents mean for Africa?
30 years and five days ago, Clifford Stoll, a brilliant American astronomer, cybersecurity legend and writer wrote (emphases mine):
“After two decades online, I'm perplexed. It's not that I haven't had a gas of a good time on the Internet. I've met great people and even caught a hacker or two. But today, I'm uneasy about this most trendy and oversold community. Visionaries see a future of telecommuting workers, interactive libraries and multimedia classrooms. They speak of electronic town meetings and virtual communities. Commerce and business will shift from offices and malls to networks and modems. And the freedom of digital networks will make government more democratic.
Baloney. Do our computer pundits lack all common sense? The truth is no online database will replace your daily newspaper, no CD-ROM can take the place of a competent teacher and no computer network will change the way government works.”
What this all means at a technological level, economic level, and social level will take some thinking. So barring a few exceptions and some planned posts, more essays in this newsletter will feature what I am learning as I attempt to generate and answer questions to help me build a framework for understanding what AI agents mean in and for Africa.
I know I’m not alone, and I’m happy to chat and trade thoughts. AI is like the internet. In the sense that it’s useless to throw around AI carelessly. But also in the way that it was shortsighted (in some ways) to dismiss it like Clifford Stoll did about the Internet in 1995.
AI will also unlikely be like the Internet partly because it is in the economic interest of a few people to sell it as the Next Big Thing and obscure the not-so-pretty details. There are enough critical takes on AI’s place globally. But the conversation in Africa is painfully narrow. I refuse to throw-up my hands and watch.
Reach out on: abraham [at] apercu [dot] pro (no spaces).