Investor interest in deep tech and AI-first infrastructure has quietly intensified in early 2025 alone, global funding for early-stage deep tech startups surpassed $4.2 billion, according to PitchBook.
At the same time, a growing number of commercial leaders are stepping away from the C-suite not to build the next consumer marketplace but to take on developer-first tools, AI infrastructure, and R&D-heavy platforms.
This shift echoes earlier waves from the PayPal Mafia to former Palantir executives now behind companies like Anduril and ElevenLabs. To understand what’s driving this transition, we spoke with Ted Chalouhi, a founder-in-residence at Antler, one of the world’s leading early-stage accelerators.
Before launching his current startup, Chalouhi held executive roles at high-growth companies like Foodbomb (acquired by Ordermentum), Deliveroo, TripAdvisor and Uber. Now, he’s building Duku AI, a platform using AI and reinforcement learning to automate QA testing by simulating real user behaviour across staging and production environments.
Q: You’ve chosen a highly technical and specialized niche for your startup , developer-focused, AI-driven QA automation. What inspired you to focus on this particular area?
Ted Chalouhi: Testing is at the core of every digital product, yet it remains one of the most vulnerable and costly processes. Statistics show that up to 40% of engineering time is spent not on building new features but on writing and maintaining test scripts that quickly become outdated. This is a huge reserve of time and budget that doesn’t actually drive business growth.
I came into this space from a revenue leadership background, and I know firsthand how bugs in production silently kill sales. At Foodbomb, a single checkout error blocked revenue for one of our largest clients for four days. Poor code quality lowers conversion rates, breaks onboarding, and damages user trust. QA has always been treated as a cost centre, but in reality, it impacts revenue.
The idea for Duku AI came directly from this misalignment: companies burn valuable engineering cycles on low-leverage QA tasks, which slows down product velocity. This turns QA from a bottleneck into a growth multiplier and that’s exactly what Duku AI is designed to do.
Q: Looking back at your time with Uber, Deliveroo and Foodbomb, which parts of that experience have proven most valuable in building a new deep-tech startup?
Ted Chalouhi: Working at Uber and Deliveroo showed what it takes to build processes that can handle rapid scale and keep teams moving fast with minimal friction. At Foodbomb, there was a clear, practical lesson, too: over a few years, the sales team grew from just two people to forty-seven, which contributed to the company’s successful exit. But as scale increases, so do the pressure points that tend to break first.
QA was always one of those critical bottlenecks. For example, every major release at Foodbomb triggered manual regression testing across ten or more screens this cost the team hours of work each sprint and directly slowed down how quickly new features could ship.
This experience highlights a core principle that’s crucial for any deep-tech startup: it’s essential to pinpoint where manual effort or legacy workflows limit scale. The solution shouldn’t just address a single task it needs to prove ROI fast and turn a bottleneck into a growth driver. In QA, this is especially clear: the right testing architecture has a direct impact on release speed and revenue reliability.
Q: Many founders with a strong commercial background usually build B2C products. Why did you decide to invest your experience in developing a deep tech solution for engineers instead?
Ted Chalouhi: Today, this kind of niche is exactly where you can deliver real impact for the market. There’s a clear trend now: people who deeply understand how to monetise, build distribution, and meet user needs are increasingly stepping into spaces that used to be purely technical. If you can speak both product and engineering, you can turn complex technology into tangible business results.
For me, building a startup isn’t about creating a flashy app to chase downloads and buzz. It’s about solving deep, overlooked problems that have a direct impact on how teams work and how reliable their products are. I want to build core infrastructure that makes shipping faster and prevents costly failures because that’s where real, durable value comes from.
Q: What are some of the core challenges in QA workflows today that you feel are still underserved?
Ted Chalouhi: Most teams either underinvest in QA or rely too heavily on automation that doesn’t scale well in practice. The result is unstable tests, constant manual upkeep, slow feedback loops, and poor visibility. In fast-moving projects, testing often drops down the priority list and product quality takes the hit.
In recent months, over 150 conversations have been held with CTOs and engineering leaders. Almost everywhere, the same pattern shows up: teams either skip testing altogether or spend 20–40% of sprint time writing and fixing tests that break for reasons unrelated to the core product logic. One CTO summed it up perfectly: “Our test suite is more work than the feature itself.”
The biggest gap is that most tools still assume a human needs to write and maintain scripts. The future of QA is fully AI-native: autonomous agents that learn how an app works, adapt to changes, and continuously validate critical user flows all without relying on human-written scripts.
Q: How do you evaluate the current competitive landscape in automated QA, and what blind spots do you think the market still has?
Ted Chalouhi: Right now, most solutions are moving incrementally , they make test writing faster or debugging slightly easier. However, very few challenge the core idea behind it all: almost everyone still assumes a human has to write and maintain tests manually.
That’s the biggest blind spot. I believe it’s time to remove that model altogether. QA is following the same path DevOps did , it moved from manual deploys to CI/CD. Here, brittle scripts are being replaced by continuous simulation.
Everything comes back to three things: team velocity, depth of coverage, and developer experience. In reality, it’s rare to find tools that truly deliver all three at once. For us, that’s non-negotiable.
Q: What role does AI play in solving these QA challenges , and are you building your own models or working on top of existing ones?
Ted Chalouhi: AI opens up new possibilities, but the real breakthrough comes from reinforcement learning. Instead of using LLMs to generate test scripts, the approach goes a step further: autonomous agents are built to learn how an application works, explore real user flows, and simulate behavior across different environments. This method is closer to robotics than to traditional test automation.
The approach is hybrid: some models are fine-tuned, while custom reinforcement learning pipelines are developed and trained on real user data. In QA, trust is everything , an unreliable AI can cause more harm than having no tests at all. For this reason, ensuring that agents remain stable and provide clear explanations of their actions is a top priority.
Q: What are some of the biggest challenges founders face when building AI startups in deep technical categories , especially when it comes to fundraising or early team building?
Ted Chalouhi: First, the market is very noisy. These days, almost anyone can put together an impressive demo, but there are very few products that are genuinely defensible and hard to replicate.
Investors now ask much sharper questions: What’s your unique edge? Are you training your own models or just using off-the-shelf ones? Can you actually sell this, not just show it on stage?
Another key challenge is building the right team. You need engineers who are comfortable working in uncertainty and can move across products, infrastructure, and AI. That skill set is rare, and competition for that talent is intense.
And finally, focus. In AI, it’s easy to get carried away trying to solve everything at once. But deep tech only works when a team goes deep on one clear problem and solves it better than anyone else.
Q: After holding leadership roles at companies like Uber, Tripadvisor, Deliveroo, and Foodbomb, what motivated you to start something from scratch?
Ted Chalouhi: Working at companies like Uber, Tripadvisor, Deliveroo, and Foodbomb gave me a unique perspective on how to scale complex products quickly, build strong teams, and handle real operational challenges at every stage of growth. That experience shaped how I think about product, revenue, and execution.
But after years of helping large organisations grow, I wanted to build something more fundamental , not just a company with good numbers, but a solution that tackles a real, daily pain point for software teams.
Duku AI is my answer to that. It’s a bet on a better way to build software, where engineers don’t lose hours fixing brittle tests, and quality is built in by design. QA stays invisible when things run smoothly but protects you when it matters most.
That’s the kind of impact that makes the years of work worth it.
Looking ahead, how do you envision the future of QA and developer tooling?
Ted Chalouhi: I believe we’re about to see a major shift. Quality won’t be something checked after the fact anymore, it will be built into every stage of development, from the first commit to production. Testing tools will evolve from static scripts to intelligent co-pilots that watch, learn, and adapt in real time.
QA will become an invisible safety net: continuous, autonomous, and capable of ensuring stability without slowing teams down or eating up engineering time. When teams can release new features quickly and confidently without compromising quality, engineering stops being a bottleneck and becomes a real growth engine. That’s the direction we’re all moving toward.