1. You’re overseeing a tech organization of 900+ people supporting products across multiple international markets. How do you maintain engineering velocity and coherence at that scale, and where does AI fit into how you run the org itself?
First of all, coherence at this scale comes from shared platforms and clear decision rights, not central control. Autonomy is essential, as we scale judgment, not headcount. We invest heavily in ensuring engineering quality standards are clear and well-calibrated across all roles, because setting a high bar consistently is what keeps 900+ people moving in the same direction.
The classic failure mode is teams solving the same problem five different ways. Our answer to that is paved roads and strong defaults: the right thing should also be the easiest thing. We’ve simplified our systems and architecture considerably, and we empower every engineering and product leader to operate as the CEO and CTO of their own organization, with a strong emphasis on communication across teams to build efficiently.
When it comes to embedding AI, our focus is on reducing the time from question to answer, whether it is incident investigation, code understanding or data access. That means smaller teams can own larger surfaces. Today, 98% of our engineers use AI, and more than half do so in highly advanced ways. The results speak for themselves: roughly three times more code written this year compared to last, with no increase in AI-supported incidents. That’s been possible because we’ve invested just as heavily in our testing and development harnesses as we have in productivity, making sure the added volume is safe to launch.
2. Super is a complex, multi-market platform. What does embedding AI into a product portfolio at that scale actually look like in practice, and what separates the initiatives that stick from the ones that don’t?
The most successful initiatives always start from a real customer or operational problem. “Add AI” is never a useful brief. What AI actually does for us is accelerate personalization and enrich the experiences we invite customers into.
Before any of that works, though, you need the unglamorous foundations in place: data quality, evaluation harnesses (testing), guardrails, the ability to measure impact, and the ability to experiment in rigorous, repeatable ways. Equally important is building an institutional memory of past efforts and failures, one that improves the whole organization over time rather than just preparing people for the next sprint.
Operating across multiple markets adds another layer of complexity. A platform capability has to be able to adapt to local regulation and local content. Our principle is to build 80% of every feature centrally on our platform, reserving the remaining 20% for genuine localization needs: payment methods, government integrations or regulatory requirements specific to a given market. Basically, we are building once and then configuring for each market to make future product and market launches faster and better.
3. You’ve led significant platform work, from distributed systems at Facebook to Applied Machine Learning at Wayve to what sounds like a major AI strategy shift at Super. What’s the hardest part of a large-scale platform migration that rarely gets talked about?
The hardest part is running two worlds at once. Keeping the old system healthy while the new one earns trust is an operational and leadership challenge that gets far less attention than the technical work itself. Building clean transition tools from the old system to the new one is what represents the real coordination complexity.
There is also a timing problem that most people underestimate. You are only done when the last team moves, and that last 1% is just as difficult as the initial 99%. The early, middle, and late stages of a migration each require a different strategy and a different style of leadership. A forcing function for the final stretch is essential.
The organizational and incentive cost is the thing that surprises people most. Someone has to own the old system while everyone else is excited about the new one, and getting those people engaged in the transition is a genuine leadership challenge. In our case, our product and operations teams had to learn entirely new workflows after moving to our central platform. The amount of cross-functional collaboration required to make that shift smooth and productive was a real test of the organization.
4. There’s a lot of noise around AI strategy right now. What’s the question you think CTOs should be asking themselves that most aren’t?
When the cost of intelligence approaches zero, entirely new things become possible. The question is whether you have the clarity to act on that, and the discipline to let go of what no longer serves you. The cultural change required for AI transformation in bringing the organization along the journey with you is the key focus from the executive role.
A second question that deserves more attention: are we building evaluation and measurement capability as fast as we are shipping AI features? Shipping without that foundation is how organizations lose their ability to learn.
The third, and perhaps the most underappreciated: the majority of AI transformation is a leadership and management challenge, not a technology one. The CTO can set the direction, but the real leverage is in getting the full management team aligned.
Technology moves fast. Organizational alignment is the harder, slower, more consequential work.
