Gmail Creator Paul Buchheit On AGI, Open Source Models, Freedom

Buchheit says Google was built around large-scale data and machine learning, from PageRank to the search infrastructure. He joined in 1999 and recalls building the first “Did you mean?” feature, using web pages and query logs to improve spell correction, including for proper nouns.

Speaking as an outsider who left Google in 2006, Buchheit attributes its slower AI response partly to a shift toward protecting the search business and greater risk aversion. He says AI could disrupt ad-supported search by giving users direct answers, while concerns about offensive outputs and regulators discouraged launches.

Buchheit recalls YC discussions about building AI in the public interest rather than leaving it inside Google. He describes OpenAI’s early roots in YC Research and says its promise to share research helped attract talent. He supports open-source models as a way to distribute AI capabilities, while noting that the cost of training frontier systems could still concentrate power.

Buchheit believes AI is on a path toward AGI because investment is now producing better systems and attracting further investment. He describes current workflows and multi-agent systems as interim ways to support more deliberate reasoning. He predicts many digital knowledge jobs could be automated within a decade, and argues that AI should give people more agency and creative tools.

Buchheit warns that concentrated AI power could enable censorship and authoritarian control, and argues for open development and open-source models. He says the goal should be to give people useful tools and let them make decisions, rather than assume institutions can plan the future. He also points to startups as a way to spread the ability to build with AI.