According to Business Insider, William Tunstall-Pedoe, the founder whose 2012 startup acquisition by Amazon became the foundation for Alexa, decided to leave the company in 2016. He used Amazon’s own signature six-page memo process to weigh his decision, concluding he had delivered all he could and that the product’s success was assured with thousands now working on it. After founding True Knowledge in 2006 and pivoting to the voice assistant Evi, his 30-person company was acquired by Amazon, turning its Cambridge office into a major Amazon hub. Since leaving, he launched a new deeptech startup called Unlikely AI in 2019, focused on neurosymbolic AI to make systems more trustworthy.
The Big Company Trap
Here’s the thing about building something world-changing inside a giant like Amazon: you get insane leverage, but you trade away a certain kind of freedom. Tunstall-Pedoe nails it when he points out the exposure Alexa got by instantly landing on Amazon’s front page. That’s a superpower most founders would kill for. But he also highlights the flip side, and it’s a brutal one for novel ideas. In a big company, it only takes one manager’s “no” to kill a project. The resources get reallocated, and that’s that.
In a startup? You just need one “yes” from one investor to keep going. That fundamental difference in decision-making physics is why so many foundational technologies are born in garages, not corporate R&D labs. It’s not that big companies can’t innovate. They’re just optimized for scaling and refining, not for the chaotic, contrarian exploration that defines true invention. Once Alexa was a hit, the job became about managing that scale, not asking the next wild “what if.” And for a certain type of builder, that’s just not the gig.
The Next Big Bet
So what does a guy who helped kickstart the voice assistant era do next? He goes after what he sees as the core flaw in today’s AI. His new company, Unlikely AI, is betting on neurosymbolic AI. Basically, it’s about marrying the intuitive, pattern-recognizing power of modern machine learning (which is amazing but often a “black box” that gets things wrong) with the old-school, rule-based logic of algorithms (which are predictable and reliable).
Think about it. We’re surrounded by incredibly powerful but hilariously unreliable AI. It can write a sonnet but can’t reliably remember your lunch order. Tunstall-Pedoe’s bet is that the future isn’t just about making models bigger, but about making them trustworthy. That’s a huge, hard problem. And honestly, it’s exactly the kind of long-shot, foundational moonshot that a well-funded startup is probably better positioned to tackle than a corporate division with quarterly goals. It’s a contrarian idea in an era obsessed with large language models. Will it work? Who knows. But it’s the kind of bet that changes everything if it does.
The Builder’s Dilemma
This story isn’t really about Alexa or Amazon. It’s about the eternal tension for creators: scale versus agency. Do you want to see your idea touch millions (or billions) immediately, with all the resources in the world? Or do you need the freedom to follow the weird hunch, to pivot on a dime, and to own the entire vision, for better or worse?
Tunstall-Pedoe clearly loves the startup chaos, even while admitting he gets nostalgic for the structure of a big org. That’s real. The stress is different. The stakes feel different. But for some people, the thrill of building from zero to one is just more compelling than managing from one to one billion. His memo in 2016 was him realizing which type of person he was. And look, the tech world needs both. We need the Amazons to deploy and scale, and we need the unlikely startups in Cambridge garages to dream up the next thing that Amazon will one day scale.
