Decoding AI with Tarek Assaad

Tarek Assaad: I mean just because there's a threat doesn't mean it's a done deal. 

So there is tremendous opportunity. I just think the stakes are really high one way or another. 

You know if we sit back and watch AI become sentient and do everything then we didn’t do our part. I think there’s a lot of opportunity for value creation. 

———

Patrick FitzPatrick: Hey everyone. 

Way back in 2020, one of the first guests on this show was a venture capitalist named Tarek Assaad. 

Tarek runs Algebra Ventures here in Cairo. And this year, Algebra turns 10. 

Ten years of backing Egyptian founders through three devaluations, a pandemic, the end of free money, and now this AI moment that we're all living through — whatever this is.

And that's why I wanted to have Tarek back. It's why he's the first guest I've ever invited onto the show twice. 

It's not just Algebra's anniversary; it's the fact that Tarek is one of the smartest people I know in both technology and in business. 

I wanted to sit down with him and just figure out what the hell is going on with AI. Not the hype, not the doom — the actual thing.

And here's my promise: If you're someone who uses the free ChatGPT now and then, like a slightly chattier Google, and the rest of it kind of washes all over you, this one's for you. 

You'll leave smarter because I'm going to make Tarek explain things to me as if I don't know them, because half the time I don't, and I'm not too proud to say it. 

And if you're already deep into this — if you're building with Claude, betting on it, or if it's keeping you awake at night wondering what this means to your business and to your kids — well, stay with us because we're going to go deep. 

And that's the conversation.

Patrick FitzPatrick: Tarek, welcome back to Making It.

Tarek Assaad: Patrick, it's a pleasure. 

I think I came here six years ago and I said, "Hi, I'm Tarek. I'd like to be friends with Enterprise."

Patrick FitzPatrick: That was it, man. That was it. 

Listen, man. I want to start with the ground floor. You've got a brother who's a CFO, okay? Smart guy in his own right, but if he came to you today and said, "What the heck is this AI thing that everyone's losing their minds about?" What do you tell him?

Tarek Assaad: It's the big technology breakthrough of my lifetime.

In the mid-'90s when the internet came out, I thought, “Okay, this is it. I'm so lucky. I'm in my 20s and I'm witnessing this incredible technological innovation.” I went through subsequent technologies and they all felt like great technologies. This is something else. This is a breakthrough. This, I think, is maybe even bigger than the worldwide web.

Patrick FitzPatrick: Bigger than the web.

Tarek Assaad: The pace at which it's moving and the impact that it's having is just insane, even in terms of adoption. 

So it took 75 years for the phone to reach 100 million users. It took the internet 7.5 years. It took ChatGPT two months. So ChatGPT today is actually at a billion monthly active users, which is insane.

In Q1 of 2026, Anthropic grew at an annualized rate of 80x. They're doing about 45 billion dollars of revenue. Forget growth in users and just look at revenue — that's bigger than Salesforce, a little bit smaller than IBM and Oracle. IPOing at a trillion-dollar valuation. 

But probably at the end of the day, great these are amazing headlines that you can read about in the paper, but what does it do for me?

Patrick FitzPatrick: Yeah.

Tarek Assaad: It's two things: it's how quickly you can do things, particularly on the tech side, and who can do them. If I look at my portfolio companies and people who are using AI, there are the tech teams and there are the non-techies.

Patrick FitzPatrick: So the marketing people, the product—

Tarek Assaad: Yeah, the product managers, the finance across the board in any function — even HR. In every function.

AI has allowed non-techies to build functionality that only software developers could do a few years back, and that's transformational. 

That’s why it's not like a cool tool. With mobile, developers could do things that they couldn't do before, but still, it was in the realm of developers.

Patrick FitzPatrick: Yeah. It was the high priesthood in the holiest of holies developing stuff with technologies that the rest of us couldn't imagine putting together.

Tarek Assaad: Today, if you're a CFO or a marketing executive or in any function, you can sit in an afternoon and vibe-code something and get something out the door — and not like a broken prototype, but actually something that works. 

It will not be perfect, but it'll be really, really quick. And anyone who knows English can do it. Like they say, the biggest programming language right now is English.

Patrick FitzPatrick: I'll come back to vibe coding, but when did AI stop being a toy that you tinkered with and something that really changed how you work as a venture capitalist?

Tarek Assaad: It's been an evolution. 

From the early days of tinkering with ChatGPT to ask it to write better emails, etc., it was cute, but it wasn't really transformational. 

If I had to take one moment, I'd say probably the first time I used Claude Co-work.

When it started doing things on my behalf, I think that's when it became more powerful for me.

Patrick FitzPatrick: Claude is having a moment right now globally. 

Tarek Assaad: A trillion-dollar moment.

Patrick FitzPatrick: A trillion-dollar moment. first out the gate for an IPO, but explain to listeners who are not head-down in this: what is the difference between Claude Chat, which you can use in your browser, Co-work, and Claude Code? 

Tarek Assaad: So the chat version of Claude is just like ChatGPT or Gemini. It's basically a tool that answers questions for you. Claude Co-work takes a task or a project off your hands. 

So you set it up — you have to invest a little bit more in setting it up, you have to tell it more about who you are, what you do, etc., and show it what good or what great looks like for you.

Patrick FitzPatrick: And in this universe, you've got a little preferences section where you're telling it about your working style, what you do, where you work, and your level of expertise. But what is the fundamental superpower of an agent? What does agentic AI mean in the world of Claude Co-work? 

Tarek Assaad: An agent is a piece of software that can go out and perform tasks, and sometimes perform a hierarchy of tasks.

Patrick FitzPatrick: What does that mean?

Tarek Assaad: For example, let's say I'm writing a memo for my IC. It requires research. It requires going through the financials that the company sent me. 

I teach the agent how to do these things, I show the agent the outcomes that are good or great in my experience, and that's the benchmark that the agent sets. And then the next time I get a pipeline opportunity that I'm working on, the agent will take it, dissect it, and do all of it.

Patrick FitzPatrick: Because it already knows what you're looking for.

Tarek Assaad: It knows what I'm looking for because it's specialized in certain areas that I've told it about. And it all comes together in a project. 

It's very context-aware. You have to give it a lot more context so that it knows what you're looking for.

Patrick FitzPatrick: I like to think of it sort of almost like onboarding a new employee. You can hire somebody really smart, but if they don't understand your company, they don't understand your desires, they don't understand your function, how are they going to perform for you?

Tarek Assaad: And the new employee learns at an insane pace. So that's the thing, because you invariably start with a mediocre outcome.

Patrick FitzPatrick: Yeah.

Tarek Assaad: But then the pace at which… you know the more you feed it — and you can feed it everything — you could say… I have some skills that say: "Go run through all my emails and look for this and that," and it'll go do it in an incredibly efficient manner.

Patrick FitzPatrick: So are you emailing me through Claude, [is it] actually emailing me on your behalf?

Tarek Assaad: You know, I thought I'd be doing a lot more sophisticated tasks, especially because I used to be a software developer, so I enjoy tinkering with this stuff. But I use Claude Co-work a lot more than I use Claude Code. 

Patrick FitzPatrick: Seriously? 

Tarek Assaad: So Claude Code is the more sophisticated version; it's the version in which you have to do a bit of software development, maybe a little bit of architecture. 

But even with Claude Co-work, you're also designing the structure of the agents. It's not architecture in the classic sense, but it's you structuring things in a way that you think makes sense and then tinkering with that and optimizing over time.

But with Claude Code, you can also make software. 

So I use it more for portfolio stuff. Let's say I want to look across my portfolio and see variance in revenues, or how companies compare in how much they hire, or how much they spend on tech, or how much they spend on marketing, customer acquisition costs — anything. It can go and do these things, it can pull in the data, it can do the analysis, it can produce these dashboards. So it's closer to what a developer would have done than Claude Co-work. 

But I'm using Co-work a lot more just because it takes a lot of day-to-day, mundane tasks off my plate.

Patrick FitzPatrick: Does a smart exec need to understand how this thing works under the hood? And by that I don't mean how an LLM works. But how I should structure a conversation with Claude or with ChatGPT or with Gemini, and the types of best practices that are included in that.

Tarek Assaad: My advice for every executive is to learn as much about AI as you can. There's a lot going on, it can be confusing, it overwhelms everyone — myself included. 

But I think some basics are important, and then you build from there. 

I think just the way you query, the way you use your tokens, etc., and even the main functionality of what you can get out of AI is very important across the board, from the CEO down to a junior employee.

Patrick FitzPatrick: What's the best way to do that? Is it to have a conversation with the bot itself?

Tarek Assaad: It's incredibly useful in that way. And you can pick your modality: you can talk to it in speech, you can chat with it in text. 

I enjoy sometimes also watching some podcasts or YouTube videos that show things. But there's so much out there — if you want it, you're going to get it done in a few hours, easy.

Patrick FitzPatrick: So be a self-directed learner, as they used to say. 

You brought up this thing called vibe coding. Break it down for me. 

What happened to make this possible? There was this inflection point maybe late 2025 last year. What was that? What was the step change that made it possible?

Tarek Assaad: When I was a software developer, we would write the code, we would put in the comments, we would proofread it — like you had someone else look at your code. We obviously compiled it and debugged it, and you had to have a certain level of fluency in the specific language that you're using. 

Now you just tell the tool what you want in English, and it does it for you. It's not super, super simple — like you can't say, "Oh, write me a new enterprise application that replaces Oracle" — but you can get it to do a lot of things, especially simpler things, very quickly and without much or any skill in programming. 

So vibe coding — you vibe with the code, you can sort of manage it, but you don't require the technical skill to create it.

Patrick FitzPatrick: That comes with risks, though, right? 

Tarek Assaad: Depending too much on AI. It's not a magic wand, it's not going to do everything for you. 

It's an incredible tool that creates efficiencies that were unthinkable a few years ago, and it allows someone like a CEO to dabble with creating software. But it will not replace the entire tech team.

Patrick FitzPatrick: So I can't fire all my developers?

Tarek Assaad: You could, but you'll regret it.

Patrick FitzPatrick: Okay, okay. Walk me through that.

Tarek Assaad: It just raises the bar for everyone in an organization. 

If you have 10 developers, you might only need three, but the three that you need will be much more capable, and they'll be asked to do much more sort of managerial type of work. 

If you think across the organization, you will not need as many people sitting and writing code or doing basic debugging or things like that.  But you need a lot more people who are aware of the context, who understand your business, who understand the customer needs. 

It's like you had hand tools and now you have power tools.

I'll give you an example on just the level of efficiency. We had a portfolio company go through a really tough time and they had to let go of 75% of their workforce. So they were forced to adopt AI much faster and much more aggressively than they would have otherwise. 

They're probably more efficient in tech now. They ship things faster, they ship the full functionality, and they've probably lost 20% or so in other functions just because they very aggressively adopted AI.

Patrick FitzPatrick: There was a pretty rigorous study a couple months back that found experienced developers felt that they were like 24-25% more efficient using AI, when in fact they were sort of 19% slower. And then you look at individual, unique data points like the portfolio company you brought up. 

Where's the disconnect? How do we understand that? What did your portfolio company understand or do that other organizations are not that made them successful doing this?

Tarek Assaad: Go back to basics.

Patrick FitzPatrick: Okay, what does that mean?

Tarek Assaad: Define very clearly what it is that you want and what is the shortest path to get there, and use as much of the AI to achieve these objectives. It's as simple as that. 

For example, back in the day, a lot of people thought that if you developed a website, you were good on the internet — that's all you needed. Same idea. If you have AI, it doesn't mean that you've solved all your problems. 

If you deploy AI in a meaningful way, understand your objectives, and understand your business very well, and use AI to serve that, then you get good results. Otherwise, you can easily get lost in the sea of things that are happening, and that doesn't improve efficiency.

Patrick FitzPatrick: Aaron Levie, the CEO of Box, the data and storage company, is one of the biggest fans of AI, and he holds that AI is not going to result in a jobs apocalypse. To the contrary, it's going to result in two things: it's going to see software developers given the opportunity to finally tackle backlog — product managers the ability to tackle backlog. 

And that job growth comes because AI will make it possible for a company like Enterprise to hire software developers — that the job of a developer is going to become not just working for a large tech company, not working for a tech startup, but for all businesses. 

You agree or disagree?

Tarek Assaad: Interesting point of view. I disagree.

Patrick FitzPatrick: Really?

Tarek Assaad: Yeah. I mean, I think it will not be the software developers who eventually build the functionality; it will be the domain experts. Someone who understands supply chain really, really well, knows the problem better than anyone, and can just tell some automation tool, "Here's what I want from you." 

I think over time the role of software developers will probably be much more on the very technical stuff that your average hospital, or Enterprise, Algebra, or logistics company will probably not really need. 

But it's very hard to try to predict the future at this point, and everyone is trying to do it. There is so much going on, but I think that is the direction things are going in.

Patrick FitzPatrick: So the finance director who's been waiting for three months for internal IT to build her a dashboard gets to create her own dashboard in the future by talking to the tool.

Tarek Assaad: Yes, I think it's starting to happen. And I think over time, and at the pace of [how] things [are] moving it will not be very long. It’s just doing to democratize all of that stuff. 

Patrick FitzPatrick: Today, if I throw my stuff into free ChatGPT, it understands what I'm talking about, but it also stores that data and can train on it in the future if it wants to. 

Whose responsibility is it in this new world to think about security of corporate data and privacy?

Tarek Assaad: In my opinion, it should be the responsibility of the OpenAIs of the world. I don't think they're thinking about this right now — they're in a crazy race and trying to beat the next guy in every possible way. It scares me. 

The race is hot, and everyone is cutting a lot of corners along the way. So, given the absence of that, it's our responsibility to moderate it, and I don't think you can reach perfection. If you want to use these tools, you have to provide data. But we all have to think about what kind of data stays proprietary and what is shared.

Patrick FitzPatrick: I think one of the coolest stories we covered recently in the Saudi edition was about technology created by TII out of the UAE, one of the sovereign research institutes. They've come up with a homomorphic encryption technology that allows, for example, a bank to send a fully encrypted prompt and data to Claude. 

Claude understands the data, is able to think about it, and returns an answer, also fully encrypted. To Claude, it doesn't actually see the underlying data, and it comes back to you in plain text answers.

Tarek Assaad: Oh, that's cool, I haven't seen that.

Patrick FitzPatrick: Absolutely insane. 

Let's say I come to you on Thursday afternoon and say, "Tarek, I want to go to work on Sunday and start an AI transformation at my company, and I don't know anything about this stuff." 

What's my starting point? What do I do that weekend?

Tarek Assaad: Well, first you familiarize yourself with all the things we've talked about in terms of capabilities — what can it do for you, etc. Then I think you identify a few opportunities where you believe you get the highest bang for the buck, and you start implementing and you iterate. 

One of the interesting things about AI is the more you throw at it in terms of data, processes, experiences, and iterations, etc. the better it gets. 

So instead of trying to architect the perfect system on day one — much like a startup — start early, fail fast, iterate….

Patrick FitzPatrick: Iterate, iterate, iterate. 

Tarek Assaad: I think you'll be pleasantly surprised how well it performs.

Patrick FitzPatrick: After the break, the part the AI industry still hasn't fixed.

———

Patrick FitzPatrick: Tarek man, these things lie to me confidently. 

I was doing some background research for a series that I want us to work on as Enterprise, and it didn't just hallucinate transactions; it came up with the wrong link to real files on our system and then told me that the hallucinated transactions were in those files, right? And very confident about it. 

I'm like, you know, I can live with bullshit in prose, but a computer is supposed to be deterministic, man — like 1 + 1 equals 2, 3 + 3 equals 6. 

You had a similar version of this with a screenshot, you were telling me.

Tarek Assaad: Yeah. I play this game — it's a base-building game — and it has different types of little buildings. I was trying to upgrade something, and I asked Gemini. They said, "Oh, you should upgrade this building." I looked, couldn't find it. So I took a screenshot and I said, "Where is the building?" It changed the building — like, put something blue instead of red, something like that—and put a big arrow on it and said, "Here's the building."

Patrick FitzPatrick: "Here's the building, dummy." Oh my god. 

The AI nerds tell me that hallucinations like that are actually a feature of LLMs and not a bug, right? That it is fundamental to the technology.

Tarek Assaad: The way the technology works is it's a transformer. 

A transformer means you are constantly guessing the next string of text in a very sophisticated way, understanding the context and all of that stuff, but at the end of the day, it is guessing — it's like autocomplete in some sense. 

So that is a fundamental part of how the technology works; that will not go away.

Patrick FitzPatrick: How do I work against it?

Tarek Assaad: There are a lot of things. "Human-in-the-loop" is one way to go, or you can get one LLM to check on another LLM, which is also something that I personally use that catches some interesting glitches. 

At the end of the day, it's a tool — and it's not deterministic like software used to be. It's a tool that teaches itself. It's a tool that creates incredible efficiencies, but it needs to be managed.

Patrick FitzPatrick: It sort of seems to be a recurring thing with these models that they'll be great at something and then they will go through periods where they will become very bad at it. And no amount of changing the prompt or reworking the workflow necessarily brings it back to what it used to be able to do, until one day you wake up and they're good at it again. 

Is that redirection of compute to other models that these foundation model makers are training? Is it a feature of the landscape? How do I even begin to process that?

Tarek Assaad: It's very hard. I think one of the issues with these black-box systems is that you really don't know what's happening under the hood. You have very little control. And that's why many people shy away from depending on them completely. 

I think, at least as things stand today, I wouldn't trust AI to go do tasks without supervision.

Patrick FitzPatrick: Does it bug you that stuff leaves Algebra that no human you employ has read — none of your partners, none of the staff?

Tarek Assaad: I don't think any of the critical things are like that at Algebra.

Patrick FitzPatrick: Critical things but.. 

Tarek Assaad: Yeah, and I mean there are things, but humans make mistakes as well.

Patrick FitzPatrick: But when humans make mistakes, there can be a consequence, right? A human carries responsibility. What do you do when the model lies to you?

Tarek Assaad: It's still the responsibility of the human using it. 

It's your responsibility to make sure that what is sent in your name or done in your name is well managed. I think that's one of the skills that we all will learn over time: where do you draw the line?

Patrick FitzPatrick: What's something that you don't read before it goes out?

Tarek Assaad: Very generic emails, maybe summaries that are shared with the team for meetings, things like that. 

But anything that goes to partners, investors, or portfolio companies, I always check.

Patrick FitzPatrick: Do you worry about providing them with access to an automated dashboard, like a Co-work artifact?

Tarek Assaad: For now, that's all internal. So if there's a fire, we can still manage it. I think until we get more comfortable with it, we probably will keep it that way.

Patrick FitzPatrick: Who's more risk-averse than you are, and who is much more interested in embracing risk?

Tarek Assaad: Karim is the most risk-averse, and the one most worried about data at Algebra. 

We have to beg him to release some data to the AI but for good reason, because he was a CTO for a long time of a publicly traded company in the US, so he understands the risks. 

I think younger people on the team are more eager to get stuff done, but in my assessment, it's a healthy balance for now. But it's very early days, very early days.

Patrick FitzPatrick: It feels to me that nobody — not the frontier model makers, not the domestic business community here in Egypt, not the government — is really thinking in concrete terms about what these models mean for emerging markets, let alone for Egypt with all of our problems and all of our opportunities.

Patrick FitzPatrick: It feels to me that nobody — not the frontier model makers, not the domestic business community here in Egypt, not the government — is really thinking in concrete terms about what these models mean for emerging markets, let alone for Egypt with all of our problems and all of our opportunities. Small question for you: what does AI mean for us?

Tarek Assaad: That's the big question for me, and that's the big concern, to be honest.

Patrick FitzPatrick: Do you see anybody thinking about that?

Tarek Assaad: There are a lot of people thinking about it; I just don't see a lot of action. 

When I'm in the Bay Area, I see things moving at an insane pace…the types of problems that preoccupy them are so removed from the world we live in here that the disconnect really scares me. 

You know, Tim Berners-Lee, the guy who invented the worldwide web, said the worldwide web will be the great global equalizer — which means if you are a kid in San Francisco, or in India, or in Kenya, or in Egypt, the one with the best ideas will become the most successful. It's a complete meritocracy.

That's not at all what happened. 

You have trillion-dollar companies in Silicon Valley and many, many people under the poverty line in the rest of the world. I think it improved lives across the board, but it didn't become an equalizer. 

On the contrary, network effects kicked in and so on, and then you had this massive divide. And if AI is, I don't know, 10 or 100 million times more impactful than the web, what's going to happen there? 

I think there is a lot of opportunity to create value and be differentiated, but it needs a lot of work, and I'm not seeing much of that happen right now.

Patrick FitzPatrick: Whose job is it to think about that stuff?

Tarek Assaad: I think it's too big to be any one entity's sole responsibility. I think it's collective. It's part of why I'm excited about what I do — I feel I can contribute to that. 

But collectively, it should be government and the private sector.

Patrick FitzPatrick: Is anybody in government talking to you about it?

Tarek Assaad: No. People are interested, there's a lot of things around education and so on, but very, very little compared to the magnitude of the opportunity and the threat.

Patrick FitzPatrick: A big part of the argument about geopolitics and AI is who owns the foundation models, the frontier models. America, in the form of Anthropic, OpenAI, Google — we can argue SpaceX — is maybe 6 to 9 months ahead of the Chinese at any given time. The Chinese, on the other hand, are well ahead of the Americans in open-weight models. 

Before we can even unpack that, what's an open-weight model?

Tarek Assaad: These large frontier models operate in a way where all the parameters that make the software and make the AI work in the way that it does are proprietary to them.

Patrick FitzPatrick: So I can't see it.

Tarek Assaad: You can't see it, and you definitely can't change it. 

What the open-weight models do is they give you access to all the weights, which are basically all the drivers of how the model works. So you can change these things and customize them to make them work in a certain way — to make it, for example, much more specialized in a certain area or to put more emphasis on something that you're more interested in. And that allows you more control over how it works, but it also affects your data sovereignty because you can run these things yourself. You can download it, you can run it yourself. 

Patrick FitzPatrick: Put it in my own data center, run it on my laptop.

Tarek Assaad: And it's much less expensive because you don't need the model to know a ton about history if you're working in a medical field. 

The Chinese have done a very good job. Like you said, they are a bit behind — it's anyone's guess if it's 6 to 9 months or two to three months or what it is — but they are a bit behind.

Patrick FitzPatrick: But not by much.

Tarek Assaad: No, not by much, and I think the gap is narrowing. But also on the flip side, the traditional LLM US companies are also trying to get into the open-weight model space. 

So it's a race, and the scary thing about the race is when it is so intense and the prize is so big, people cut corners, and that worries me.

Patrick FitzPatrick: That worries you; that's keeping you awake at night.

Tarek Assaad: Yeah.

Patrick FitzPatrick: I think it's important, if we're going back to the 101 part of this podcast, that we differentiate between two different cost pools in AI. 

There's a cost pool called training, and then there's a cost pool called inference. 

Training is where OpenAI trains the next version of ChatGPT, teaches it how to do something, weeds bad habits out of it — maybe someday they'll convince ChatGPT to talk to me in something other than bullet points, which would be lovely. But that is a process of creation. 

And then there's the cost of inference, which is how much does it cost OpenAI when I ask a question of ChatGPT. 

How do I think about that as a CFO or a CEO when I look 12 months down the road?

Tarek Assaad: What I see people doing right now is [sending] the very critical tasks to the frontier models, like in our case, for example, looking at financial models or stuff like that. But I think or how I've seen people think about it is in terms of return — how much am I benefiting? 

The bills are starting to become painful, so what benefits am I getting out of these things? There are always ways to cut down on the cost significantly by using local open-weight models, etc., and they are catching up.

So the cost-benefit analysis, I think, is becoming more of a thing right now. 

In the beginning, people just wanted to be in on any AI.

Patrick FitzPatrick: Token maxing. 

Tarek Assaad: Yeah and you know the bills were coming in but they weren't horrible, and it was cool and new and so on. I think that thinking has changed. 

Everything else in the economy is rewarded on output; AI should not be any different. You should be rewarded on the result — the benefit that AI creates for you. 

So in making these decisions, I see people increasingly making them based on ROI and not based on novelty or a desire to look good in front of some stakeholder, investor, team, etc. 

And I think there is going to be more of that — more of this scrutiny. 

Patrick FitzPatrick: What about Arabic? The Gulf is putting some pretty real money into models like Falcon, Alam... 

Are they good enough for serious work? Are they the future, or do they become what happened with Arabic search and Arabic translation a generation ago, where they're sort of all subsumed by foundational Western tech?

Tarek Assaad: When I first started in venture, there were a few companies trying to do this — this was what, like 17 years ago?

Patrick FitzPatrick: We're getting old, man.

Tarek Assaad: Experience, experience. 

There were many local companies, Egyptian and regional, who were looking at native Arabic applications. 

And then someone like Google came along and did it parametrically for Arabic and for every other language just by understanding the correlation between the words, and blew all these things out of the water. 

Will that happen with LLMs or with AI or not? I'm not sure.

But when I ask the strongest CTOs in my portfolio what they use today for Arabic, they mostly use ChatGPT, Gemini, maybe Qwen. 

The Arabic models like Falcon are a contender, but it's not a category of its own that's distinct that people are clearly using, so it's iffy in my mind.

Patrick FitzPatrick: How real for you is the worry that this stuff is just making us dumber, that we're completely outsourcing our thinking?

Tarek Assaad: The dependency on it is worrying me. I see, especially with my kids' generation, this idea of, "Well, I need to have a novel thought, so let me ask AI."

Patrick FitzPatrick: Novel thought from AI.

Tarek Assaad: Yeah. "How can I do something cool? Oh, let me ask AI." 

For example, I was talking to a friend who's a CEO of a large company here and he said, "Well, I'm doing a CEO skill where I look at the questions that I ask when I go into a meeting. Typically the first five minutes of a meeting, I ask these types of questions, so I created this CEO skill, and that will go into the meeting before me to prepare."

Patrick FitzPatrick: Oh shoot me now. No, dude. No, man. 

Tarek Assaad: So where does that go?

Patrick FitzPatrick: I have a rule: if you send a bot to one of my meetings and you're not on the call, I kick your bot out. 

Tarek Assaad: But where does that go? 

If you think of this as the commoditization of intelligence, are you also commoditizing your essence — the essence of your thinking, the essence of your creativity, etc.— by handing over your know-how to AI, and eventually AI will work on it and be better at it than you are? And then where do we all end up?

I mean, that's the scary version of things, especially if you have an inherent belief that it's better than you are at everything, then you stop trying. 

You stop trying to be creative, you stop trying to problem-solve, you stop trying to figure out the next wave of whatever it is that you're working on, and you create more of a dependency. I think that's the risk — that it no longer stays a tool because it is sentient or will become sentient…

Patrick FitzPatrick: You think it becomes sentient?

Tarek Assaad: It will. I mean, eventually. 

Patrick FitzPatrick: That’s a big thing man.

Tarek Assaad: Yeah I mean that's what people who are much, much smarter than I am are all saying. 

Eventually we will not contribute to the economy, we will not contribute to innovation; we might be able to contribute capital or resources, but we're also big polluters. So where does that leave us is another question.

Patrick FitzPatrick: Us Egypt, or us humans?

Tarek Assaad: Us Global South.

Patrick FitzPatrick: I kind of wonder if there's not something for us in what Chris Sacca... For listeners who don't know Chris Sacca, he's probably best known for having raised a relatively small fund that outperformed by investing in names including Airbnb, Twitter, Uber, etc. 

He was quite a thing for a while at the height of VC mania, and today if he invests in anything technology, it is tech that is adjacent to real-world experiences. He would rather have a sports team, an arena, or a performance space for artists — and sure, maybe some tech around that — than any AI investment. The only differentiator for us becoming our humanness.

Tarek Assaad: If you look at chess, chess has become much more popular in the past 10 years even though a computer can easily beat a human. Again, what happens in the future is anyone's guess. 

There are still certain aspects, like ethical boundaries, that will always be the role of a human. 

Patrick FitzPatrick: We both have daughters in college. They've both just finished their first year. 

What are you telling your daughter in terms of getting ready for a world where AI plays a more prominent role in one form or another?

Tarek Assaad: What I told her is there's a threat and an opportunity. Increasingly, fresh grads — even ones who graduate from top schools — are finding it hard to stay relevant in the workplace. 

But on the flip side, there is tremendous opportunity for entrepreneurship, for AI-augmented tasks across the board. 

Across the board this is a phenomenal time to start a company.

Patrick FitzPatrick: A phenomenal time to start what type of company?

Tarek Assaad: A company that solves a problem that you're passionate about.

Patrick FitzPatrick: So that hasn't changed.

Tarek Assaad: No, I don't think that has changed. This is the golden time for it. 

You can do so many things with so little and do it while others haven't caught up yet. But in order for you to play in that world, you have to have the know-how and the familiarity with these tools that other people don't have. 

So for someone who's at college, I think the opportunity is to stay close to what's happening in AI, be very conversant, be completely fluent in it, understand how to use it, and go do something that you're passionate about. I think that type of endeavor is very exciting.

You know if you want to go get a job at a large multinational and go through the ranks, I think that’s less exciting.  

Patrick FitzPatrick: The idea that now is probably a golden age to create something stands in opposition to your fear for the future, right? 

And I know from our past conversations that you worry about this stuff more than I do. I am from an industry that your industry tells me is dead, will be automated — I'm out of here. 

And yet you have literally said that if AI learns to teach itself, we're screwed. 

How do you reconcile in your own mind, man, the difference between "now is the best time for my kid to create something" to "if this thing becomes recursively self-improving, we're all screwed"?

Tarek Assaad: I mean just because there's a threat doesn't mean it's a done deal, right?

Patrick FitzPatrick: Okay.

Tarek Assaad: So there is tremendous opportunity. I just think the stakes are really high one way or another. 

If we sit back and watch AI become sentient and do everything then we didn’t do our part. I think there’s a lot of opportunity for value creation. It’s like you turn any stone and you find a small treasure or a big treasure.  

On the one hand I am worried about the gap between have and have nots — particularly for developing markets, for the Global South, for Egypt and other similar markets — but I also see tremendous opportunities. 

I can imagine it and I see it practically happening. 

Patrick FitzPatrick: When we come back, I want to make it personal to you, Tarek, and ask: what does AI mean for your business today?

———

Patrick FitzPatrick: Let’s bring this home to you man. You guys are ten years old now. If you were building Algebra again today what would you build?

Tarek Assaad: I would probably start with more efficiencies in mind. But I think fundamentally the type of work that we do entails a lot of connecting with other people. AI is not going to do that for you — at least for now. It's a people business at the end of the day.

Patrick FitzPatrick: Let the bot handle the due diligence, but...

Tarek Assaad: It'll run the financial model, it'll do the market research and all of that stuff — it does that very well. 

But at the end of the day, you connect with people, you build relationships, you sort of sit in the trenches with the founders, and that's not something that AI can do.

Patrick FitzPatrick: How is AI going to change what you invest in?

Tarek Assaad: We were looking at opportunities in the past year or maybe a year and a half that we passed on, that we definitely would have jumped on four or five years ago.

So there's that. There is understanding what does AI do, how close are these businesses to the models, and how easily can they be disrupted? That's one thing. 

For the portfolio companies; we have portfolio companies that are AI-native, and we have portfolio companies that are implementing AI to improve their efficiencies or to cut costs, etc. 

So I think there's a lot of thinking of how AI can fit into the different stories of portfolio companies that we're invested in.

Patrick FitzPatrick: Are you more attracted to a company that has AI as a revenue driver or AI as an operational deliverer of cost efficiencies?

Tarek Assaad: It's contextual, Patrick. I'll give you a couple of examples. 

We invested in a company called DeepEcho, which does prenatal ultrasound scans. So they have an AI engine that sits on top of the scanner, and it started out as automating the scans, but then over time, now it also detects preeclampsia and gestational diabetes. It detects things that are very fundamental to the health of an expecting mother. And they've taken FDA approval in the US — actually, the FDA designated them as a breakthrough technology.

So I think the level of ambition that I'm seeing from local and regional entrepreneurs has also changed a lot. These guys are going after the global market.

Or I can tackle the US market because I have a technology that nobody else has. 

We invested in another company that has an AI technology for credit scoring, and they have clients in Latin America.

Patrick FitzPatrick: From the Arab world.

Tarek Assaad: Yeah, it's an Egyptian company. 

And they're expanding; they're looking at options in Asia, etc. So again, if you go back to the analogy of, "I had a hand tool and now I have a power tool," with a power tool, I'm emboldened. I think I can go conquer the world. I'm standing on firmer ground. I have more behind me, more momentum, more resources with little capital. 

So these things don't require Silicon Valley levels of funding for them to work. They require the right setup, and you can move very, very quickly. 

Patrick FitzPatrick: We had one of your partners, Laila, on stage in October 2025 at our Egypt Forum and she said something that still kind of echoes in my mind: "We are timing out any investment that could be disrupted by AI until we have a clearer view." 

Sounds like you have a clearer view now.

Tarek Assaad: We do in some areas. And we have passed on businesses that performed phenomenally well in the past three or four years — regional expansion, amazing margins, very strong moat relative to what is there today — but if we look a little bit into the future, it looks much more murky. So we've decided not to invest in these types of businesses.

Patrick FitzPatrick: Without naming a name necessarily, can you give me an example?

Tarek Assaad: For example, training — large enterprise training companies that can sign large government contracts, large enterprise contracts, provide the training, do the design, etc. 

The customer intimacy part they're very, very strong in, but everything else is threatened.

Patrick FitzPatrick: How defensible are those companies from the frontier model makers themselves? 

The risk that Claude [or] Gemini climbs the chain — what are the odds that these companies get sherlocked, that the fundamental producer of the technology decides, "Hey, you're a feature in my app now"?

Tarek Assaad: Anytime you're close to a large platform, you're running the risk of that happening. 

A company that does AI for ultrasound scans will not be eaten by an LLM. I think things like network effects, being the first to market, and building a position are the things that allow you to be defensible against an LLM — but also being thoughtful about where you position yourself in the value chain. 

If you're close to the customer, if you understand the market, if you have the right domain expertise, and you build around that with data, connecting with partners, and building an ecosystem — just like what happened in every other technology wave — I don't think AI is different in that perspective.

Patrick FitzPatrick: That's again kind of optimistic ya Tarek.

Tarek Assaad: I am an optimistic person. I'm a little worried about AI because of the reasons we talked about, but yeah, no, I am definitely optimistic. 

I think overall there's tremendous opportunity, even for us in Egypt. I think there is opportunity, and I wouldn't be doing what I'm doing if I wasn't.

Patrick FitzPatrick: Your model leans pretty hard on the big DFIs, development finance institutions. The Europeans and the IFC have consistently backed you through two different funds. 

You're going to be raising again in the potentially not terribly distant future. How do you see conversations with potential LPs will change as a result of AI?

Tarek Assaad: There are a few things at play here. 

The magnitude of the opportunity for us has increased — like I said, with founders being more ambitious with AI enabling founders to do more. 

But also the types of problems that can be solved, which are interesting for DFIs both from an impact perspective and a financial return perspective — access to healthcare, access to financial services the types of problems that we have been trying, working to solve for for a long time now will cost less to solve or require less resources.

So I think that's exciting for everyone. I think the, you know, LPs in general want to see exits.

We’ve had a few, we're working to get more, but I think that's something also that we need more momentum. 

And on that, I think acceleration that comes with AI, will mean faster results earlier. 

Patrick FitzPatrick: Get them to the size where they're more exitable.

Tarek Assaad: Yeah, so that's something we're more excited about.

Patrick FitzPatrick: You just made two of your younger partners into managing partners — Omar and Laila. 

10 years in, does this mean that you're handing over the keys to the next generation, or are we still going to see you at the table 10 years from now?

Tarek Assaad: Honestly, Omar and Laila have been doing that role for a while, so this is just a recognition of an existing role. 

We are a team, we work well together, and I think we bring different perspectives. 

And I think as long as I'm enjoying it, I want to keep doing it. And I'm very excited about founders from our part of the world creating solutions that work in other markets as well. 

I think that's been something I've been very keen to see. I think the next wave will be a higher level of ambition, which is very exciting for me.

Patrick FitzPatrick: 10 years from now when we sit down again to do this interview, what does Algebra look like?

Tarek Assaad: Broader geographical footprint. In the past four years or so, we've gone into North and Sub-Saharan Africa and have found great partners to work with, great co-investors, and phenomenal founders to back. 

When we turned 10, it was a good time to reflect on what went well and what didn't. 

And I think one of the things that I'm very proud of is how much we've learned in Egypt, and I really feel that when we go to other markets. 

I feel it because we can show the experience that some of our founders have had, and we can work with partners to apply these lessons in other markets. 

We did our first deal in Nigeria, we're looking at South Africa quite closely, we want to do more in Morocco — we want to expand beyond these geographies eventually. I think there's tremendous….

Patrick FitzPatrick: Expand beyond Africa and the Middle East? 

Tarek Assaad: Very possible, very possible yeah. 

Having spent time in the Bay Area earlier in my career, and then the vast majority of my career here, I see the disconnect. I see that there are a lot of local problems, Global South problems, that the very advanced economies aren't really solving, because they're solving bigger… not because of anything bad it’s but…

Patrick FitzPatrick: It’s different.

Tarek Assaad: Yeah it's different and they have their big opportunities that they’re running after. But it means that many of the other opportunities get left behind, and these will only get solved by people who really understand these problems and potential solutions for them — and these are the founders on the ground in developing economies. 

So I think that's the more exciting thing for me.

Patrick FitzPatrick: Tarek, thank you very much for joining us today, man.

Tarek Assaad: Patrick, a pleasure.

Patrick FitzPatrick: Habibi.