Thursday 24 September 2026
Goodbye Google
Today I’m sending the following email:
I’m resigning from Google today.
This has not been an easy decision. I love my colleagues and my work environment, and being paid handsomely to solve fun puzzles has been amazing. But my team’s goal is ultimately to make AI much cheaper and lower-latency, and I don’t think that’s good for people right now: I firmly believe AI progress is currently far too rapid (and I have doubts about the destination too). It’s practically impossible for me to move to a different Google project that wouldn’t accelerate AI (partly due to my ties to New Zealand, where Google prefers not to do engineering), so my hands are tied.
There are millions of people contributing to AI acceleration and taking my foot off the accelerator will have a very small impact … but not no impact; some of my skills are rare. I explored trying to positively influence events from within GDM, but that effect does not seem to be strong, and I can have influence outside Google too. It’s tempting to just turn a blind eye to the impact of my work, but that would not be a Jesus-following thing to do. I have written more about these tradeoffs on my blog.
I don’t know exactly what I will do next. I will continue maintaining Pernosco and rr, and relatedly I plan to investigate how AIs debug code today and whether and how debugging tools could help. I have other project ideas I want to work on, some potentially lucrative, some not. Maybe I’ll find an existing project that’s compelling. I definitely want my future work to be unambiguously pro-human.
First, for those who don’t know me: I’ve been in the tech industry a long time and I have a lot of Silicon Valley connections, but I live in New Zealand so I live outside the industry bubble and also outside the American bubble. I’m a Christian, and actually an elder and occasional lay preacher in the English-speaking congregation of Auckland Chinese Presbyterian Church in Auckland’s inner city. That is, I am not a “tech bro”, nor do I fit into the self-described “rationalist community” … but I do think many of their arguments deserve to be taken seriously.
I have a lot of thoughts about AI, but I’m not going to elucidate them all in this post. In summary, I think the existential risks many people are warning about deserve to be taken seriously; a lot of the phenomena predicted by the “doomers” have come to pass (e.g., reward hacking, misalignment, deceptive models, model eval awareness, psychotic swarms). However, I am not convinced the chance of ASI doom is 100%. Rather, I think the risk is real but uncertain — but that itself is very alarming! We are morally obliged to make a massive effort to minimise such risk, and most likely the risk is high enough that aiming for ASI in the near future is inherently irresponsible. I’m also very concerned about other AI-related issues: cognitive surrender, AI-induced psychosis and loneliness, power concentration, economic disruption, cybersecurity, lack of accountability, and so on. I think the potential benefits of AI are quite unclear and currently, if I had to bet, I’d bet the negatives will outweigh the benefits … but I’m uncertain about that too.
Here are some things I’m confident about. I’m confident that the people in AI labs who are issuing warnings about AI are generally sincere. I’ve talked to many people in Google Deepmind about these issues and almost all of them have sincere and serious concerns, whether or not they voice them in public. I have seen no hard evidence that people are hyping AI risk as a means to boost company stock prices or regulate away their competitors. (I think national and international regulation is desperately needed!) I’ve seen a lot of arguments of the form “you can’t trust those people”, and maybe that’s true, but such distrust is not a good reason to disregard their warnings, as Russell Moore eloquently explained recently.
I’m confident that AI capability will continue to keep increasing steadily as long as we keep working on it. I wish that AI would hit some kind of plateau, or that we would identify important human cognitive abilities that AI will never replicate without a paradigm shift, but I don’t expect those wishes to come true. Model progress on benchmarks seems as fast or faster than ever, and with it, qualitatively new capabilities keep emerging. Even if model progress stopped today, we could spend years effectively unlocking new capabilities via new prompts and harnesses. Many prominent AI detractors (looking at you, Zitron and Doctorow) seem to think that AI is some kind of scam that won’t really work. I think it will.
I’m very confident that even if there is a path to a better future through AI, the current rate of change is far too high. AI is developing faster than humans can individually and collectively understand it and adapt to it. People trying to plan their futures, e.g. trying to plan for a world several years in the future as they enter university, can no longer do so the way previous generations could. I don’t think we’ve seen anything like this before, certainly not in the previous technological shifts I have lived through (PCs, the Internet, smartphones). Even during the Industrial Revolution, not only was change much slower but there were large swathes of human activity that were not and could not be directly impacted by the new machines. That is not very true anymore.
Why leave now and not earlier? It’s nothing to do with the recent spate of viral resignations or “AI slowdown” warnings; that’s a coincidence. I’ve had this date in mind for a while, because I have a long-planned ten-day backpacking trip with my friends starting Monday (Abel Tasman and Wangapeka tracks) and I wanted to go before that.
I did not work directly on AI capability, but on improved tools for hardware chip design. I really enjoyed the work, and for a while I told myself it was relatively harmless, but over time God forced me to confront the reality that the main impact of these tools will be to accelerate the design of a new breed of AI chips, which if successful will make AI much cheaper and faster — making AI more pervasive, and also more capable since we’ve learned to boost capabilities by burning more inference tokens. My duty to be a good employee meant I had to have an honest conversation with my skip manager and tell them I was at best reluctant to see their project succeed! Even after that I wanted to be really confident in my decision, because the great deal I had working for Google in New Zealand will probably never be available again. (Staying at Google and switching to a different engineering team not accelerating AI was impractical, because Google doesn’t have other engineering teams in New Zealand.) My aforementioned duty to my employer, and my respect for the people, was also a factor for not leaving too abruptly and trying to hand over my work in a reasonable state.
What next? The most important thing I’m confident about is that the Jesus of the Bible is real and therefore God has a plan that’s good for us. I don’t know what that plan is (and wish I did) but it lets me sleep at night in spite of the AI chaos. I expect his plan involves me continuing to make the best use of my talents. Even if the plan is for Jesus to return to rescue us from our folly, we’d better be busy when he returns! So, as long as the talent God gave me is valuable, I want to keep working. As I mentioned above, I plan to continue maintaining Pernosco and rr. Under the Pernosco umbrella, I plan to study how AI agents debug code and whether debugging tools that can make them more effective at that. I want to use AI agents to bring some of my hobby project ideas to life. I’m keen to reap the benefits of AI, but cautiously, in ways that benefit humans and keep my own mind sharp. As much as I can, I will continue practicing and advocating for that here in New Zealand.