Hey 👋

Sheriff here with a quick question.

Do you feel like your workload keeps growing even though you’re using AI?

There’s a reason for that, and it could have far-reaching implications for people, and companies too.

This week, we’ll dive into that reason. Let’s get into it.

Each one goes the same way. 

But once you have something you like or can work with, you close the tab. And in that click of a button, all the context stays frozen there forever.

It’s the same across all LLMs; they’re not the best at remembering you. Instead, they work like a smart person with ADHD.

And it makes sense if you understand how AI works. Your AI bot reads what is in front of it, predicts the next word, produces an answer, and stops. The next session opens with the same blank slate as the last one.

For one person, it can be annoying, but for companies, it can compound into major missed opportunities.

At its core, companies are a mix of two things: people and skills. 

The amount of work produced by any company can be expressed as a product of these two variables. Adding more people can translate to more output. The same dynamic applies to skills: more skills mean more output.

AI breaks this formula: it makes it possible to have fewer people while increasing the amount of skill in a company.

So, one person can make a deck, work on a report, conduct in-depth research, and still take notes during a meeting, all in the same hour.

If one person is this productive, it stands to reason that the entire company’s productivity goes up, right? 

But that’s often not the case.

In 2025, Glean AI did a survey on AI use in companies. About 6,000 (90% of total) full-time digital workers said AI automation saved them 11 hours every week. 

When the companies were asked, only 13% of leaders said their companies had benefited from it.

Even at the individual level, workers report spending 6.4 hours every week doing something called “botsitting”. This is when you feed AI context it should already have, cross-check its output, and clean up its work when it’s poor. 

The reason this happens is that AI is a forgetting machine. Because our initial formula (people x skills) is simply not enough;  a company is more than just people and skills.

There’s a third variable: institutional knowledge.

This refers to the nuance and context in which all work in a company gets done. The texts on Slack where decisions are made, live meetings where ideas are shared spontaneously, the bigger picture in which the project exists, and where everything gets housed.

It seems like a tiny detail that shouldn’t matter, but many years ago, it almost killed one of the fastest-growing companies in the history of the world.

Last year, Ben Horowitz, co-founder of A16Z, one of the world’s largest VC funds, told a story on a podcast interview about Facebook that sounds like folklore. 

In 2008, Mark Zuckerberg was thinking of firing his entire management team.

Facebook’s growth had flattened, he’d turned down acquisition offers, and the management team seemed to have lost faith in his ability to bring back growth.

Ben asked if Mark knew why growth stalled, and Mark explained that they’d hired hundreds of new engineers who were writing to the Facebook database, instead of a MySQL layer that interacts with the database and makes user requests faster.

This meant that logins took about 30 seconds longer, and in that time, people dropped off. So, signups flattened.

Ben then explained that when a company is young, there’s no knowledge in it. It’s just people and their talent doing the hard work of creating the company. 

But as it grows, knowledge gets created, and people who join need to absorb all the right information to do work that’s additive, not subtractive.

Mark went on to create a 40-hour crash course for every engineer who joined Facebook, reversed the database issue, and growth came back.

This was a knowledge problem that stalled the growth of Facebook, well before LLMs ever existed.

In Africa, that problem arrives with a worse twist. AI is supposed to fill the continent's skill gap, which is why African CEOs plan to put 26% of their budgets into it this year.

But AI in Africa could simply be swapping skill gaps for context gaps instead. 

It’s most likely why those same CEOs investing in AI also listed integrating AI into core operations as their biggest operating challenge by far, even more than regulation or cybersecurity.

Two guys in Nairobi saw that context gap coming.

In 2024, Lorcan O Cathain and Stefan Kruger were thinking about what to do with AI.

Lorcan was a serial founder who’d started and led a few companies across Africa; and Stefan was vice president of engineering at Paystack, which had been acquired by Stripe.

Both believed in the same thing: that the companies that will win in the AI era are the ones that will manage their AI workers as intentionally as their people.

So, they started building custom AI agents to help different companies automate their operations.

It started with one client. The client was running customer support over WhatsApp, with a team working through thousands of messages a day by hand. 

The duo built an AI agent that sorted through the messages, so the support team only had to step in when a conversation genuinely needed a person.

Word spread quickly, and before long, Lorcan and Stefan were building agents for more companies under a new name: Lua AI.

Lua’s AI agents could complete entire workflows from outreach to onboarding customers and even handling support tickets. 

Within its first three months, Lua reached $1 million in annualised revenue. In April 2026, it raised $5.8 million from Norrsken22, Y Combinator, and a number of other investors.

So far, the company has deployed over 5,000 agents and processed 50 million messages.

Then Lua started to notice the cracks in how these companies use AI. The agents could complete tasks, but nobody could see how they got there, or how to make the next task better.

It was like having an employee who does great work once but can never tell you how. And in this kind of universe, if things break, it’s not obvious where to look, because knowledge was never shared.

So, Lua built Workspaces. 

Think of it as a shared institutional memory. As your team works across different tools, Workspaces keeps up with conversations and decisions that would be difficult for any one person to follow.

Here is what having Workspaces inside your business looks like in practice.

After a week away from work, Lorcan asked his Workspaces one question: “How did the team get on, what moved on our deals, and is there anything that needs me first?” 

A few minutes later, he knew what had moved, what had stalled, and what was waiting on him. Nobody wrote a report to make that possible, because Workspaces sits inside Slack, email, and meetings and builds the connections as the work happens.

In essence, Lua is…

Let’s go back and fix our original formula for quality output. We now know that..

Quality Output = People x Skills x Institutional knowledge

Here's what makes this moment exciting. 

The first two variables are getting cheaper by the month. Skill used to take years to build; now it's a subscription. Headcount used to be the constraint; now one person does the work of four.

Which means institutional knowledge is about to become the only variable left that separates one company from another. 

And companies that are able to compound their knowledge the fastest will be best placed to win.

That compounding engine is what Lua is building with Workspace.

At Tech Safari, we’re working with Lua as our AI partner, and you can have special access to join the waitlist for Workspace before it’s released next week. 

And if you want to go deeper into using AI in your company, Caleb Maru is doing a session with Lorcan on how the best companies get the most out of AI.

Cheers,

Sheriff.

That’s it for this week. See you on Sunday for a breakdown on This Week in African Tech.

Cheers,

The Tech Safari Team

PS. refer five readers and you’ll get access to our private community. 👇🏾

Wow, still here?

You must really like the newsletter. Come hang out. 👇🏾