How AI is changing software testing

🔓 Unlock full access to deeply reported, important stories on startups, venture capital, and tech in MENA you won’t find anywhere else.

Start a FREE 7-day trial

Hi friends! 👋

Shipping software has never been faster, yet keeping it from breaking has never been harder.

Take the recent Tea‑app breach. A single misconfigured Firebase bucket went live late on a Friday. By Monday, an independent researcher had quietly exfiltrated roughly 72,000 user photos and more than a million private DMs.

Tea’s engineers were shipping product updates at record speed, but their QA net, focused on UI and API tests, never touched the storage layer.

The same tension runs through the “vibe‑coding” landscape.

Platforms like Lovable, Bolt, and Replit, dev tools like Cursor and Windsurf, and agentic coders like Devin let teams ship features in hours, but QA still drags on for days and often becomes the growth‑stage bottleneck.

Global software testing spend is already in the tens of billions and is projected to climb toward $90B by 2030. Every release adds complexity that human testers or brittle Selenium scripts struggle to tame.

Thunder Code believes that logjam is fixable.

Tunis-born, Paris-based entrepreneurs Karim Jouini and Jihed Othmani sold Expensya to Sweden’s Medius in 2023 for a reported sum of more than $120 million, marking one of Africa’s largest tech exits.

They had promised each other they would not start another company after selling Expensya, but they once again found themselves face‑to‑face with a problem they couldn’t ignore: QA. Six weeks later, a prototype convinced them they were onto something bigger.

Thunder Code’s answer is automated personas. An Accessibility Auditor here, a UX Specialist there, each brings domain‑specific expertise to QA. Describe a test in plain English, pick a persona, hit Run, and the agent spins up a browser, walks the scenario end to end, and returns tailored feedback designed to mirror how a human expert would review the product.

The duo went back to the very investors they had once told “never again” and those investors still piled in. Silicon Badia led a $9 million seed round with Janngo Capital, Titan Seed Fund, and angels including Roxanne Varza (Station F) and Karim Beguir (InstaDeep).

Here’s what we covered:

This conversation is one of our all‑time favourites. Karim is among the most thoughtful and deeply insightful founders operating in the region, and his perspective on building, scaling, and staying ahead is one you will not want to miss.

We hope you find it every bit as valuable and eye‑opening as we did 👇

If you're short on time, here’s what investors, founders, and operators, should take away from Karim Jouini’s Thunder Code playbook.

Premium members get the full version of this article, plus a TLDR summary with key takeaways and actionable insights right here.

Okay, let’s dig into it 👇

Tell me a bit about how the company came to be – the thesis behind it, how you spotted this particular pain point. I only realised during my research that you’re quite a prolific angel investor too. Was this something you picked up on through your portfolio companies?

So, first of all, with my cofounder, Jihed, we had promised each other we wouldn’t start another company after selling Expensya, it was really hard. But the reason we decided to do this is that, through our software experience, whether at Microsoft, Expensya, or with our portfolio companies, the thing that always slowed companies down as they scaled was quality assurance.

You start in a single market with a single product. The reason people choose you over incumbents is because you offer better quality. But once you start to scale, quality assurance becomes your biggest challenge, how do you maintain that same level of quality?

It’s a problem everyone knows exists, but people grow numb to it. There are really only two ways to assure software quality. You can have humans test it, actual customers or employees acting like customers, which delivers high quality but doesn’t scale. Or you can write a huge amount of test code. At Expensya, we had a million lines of test code, that is just the tests, not the actual product code. Complexity increases as you grow, so you automate as many scenarios as possible.

We had grown used to this problem, but then generative AI came along and made us rethink it. What if that non-scalable human testing could suddenly become scalable? That’s how it started. We hated quality assurance and testing so much that we decided to make it our mission to transform the field.

And, honestly, I was also on paternity leave at the time. I had six weeks off, and somewhere after my boredom threshold was reached, over a few drinks, we decided this is what we are going to do.

Once you decided to pursue it, what did that look like in practice? I read you shipped the Thunder Code MVP in six weeks – how? Did you use low/no-code tools, given their rise (if I read about Lovable hitting $100M ARR again, I might lose my mind)?

First, we had to agree on who we are. When you say “testing,” there are so many ways to approach the problem. As a startup, you’re tempted to do a lot of things and try to make everyone happy, but in reality, you need to define who you are, and then focus on who needs you.

We spent maybe a week or two debating that and doing a little research. We spoke to CTOs and CPOs around us, probably around a hundred people. We pushed them to give us answers quickly. Eventually, we agreed that we would be the “manual” of this space, our tech would be the equivalent of a human tester, but scaled.

Once we defined that, we asked ourselves: what kind of prototype would allow us to reach our goals? We needed to convince ourselves the problem was solvable. We weren’t sure if the technology was mature enough. We needed to hire people, and to hire people we needed to show them something. The same goes for investors – we needed something tangible.

So we tested different AI and low-code solutions, picked one, and then spent a few late nights over three or four weeks building that prototype.

The prototype was very specific: it was about testing our previous company, Expensya, which we knew inside out. We built test scenarios for it and looked at whether this approach was much more efficient than what we were used to. We were convinced the answer was yes. That’s when we decided to launch the company.

You’ve said before that you’re not afraid to dilute early in order to hire top talent. What has shifted in how you think about equity and leverage compared to your days at Expensya? And especially now, with AI talent being such a hot topic, how are you approaching it differently?

Honestly, I think about it the same way. I’ve always believed in sharing equity. I had that intuition early, and then I experienced it firsthand at Microsoft. When I joined, I was just a random software developer in a company of 100,000 people, and I got a decent amount of equity. Even if they hadn’t given it to me, I would have taken the job, but having equity changed how I felt about the company and how much I wanted Microsoft to succeed. I was 21, and it pulled me into really feeling part of the company.

That showed me, from the recipient’s perspective, that it works. Then as an entrepreneur, you realise what you’re trying to do is incredibly hard, and you can’t do it alone. You need people with skin in the game.

So my thinking hasn’t changed, but the difference with Expensya is that at Thunder Code, we’re doing it much faster. At Expensya, we didn’t have the cash to hire top talent right away. Equity is great, but you still have to pay people, even if they accept less than before.

With Thunder Code, we were able to bootstrap. We launched the company, won our first customers, and raised money while burning less than $200k. We’re very frugal, but we were also not afraid to push hard from day one. That meant hiring top talent right away.