Why Google moved fast on AI in Search but held back on chat

This article summarizes an episode of Stripe’s video series featuring Sundar Pichai, CEO of Google.

Sundar Pichai, CEO of Google/ Photo credit: Sundar Pichai

When the AI boom started, critics labeled Google a slow company that failed to capitalize on its own inventions. Sundar Pichai, CEO of Google, rejects the idea that his team ignored the technology.

He argues that the company immediately integrated new tools into its search engine while intentionally holding conversational software to a higher safety standard.

This approach shifts the focus from releasing unfinished chat programs to building physical data centers and ensuring global products function reliably.

The gap between inventing a tool and using it

Ensuring this reliability meant keeping early developments behind the scenes rather than launching a flashy, unstable product. Critics argue that researchers published the first papers on language models and then abandoned the technology.

Pichai says this misrepresents the company’s actions, noting that engineers immediately applied these discoveries to handle billions of daily searches.

He explains, “Transformers [language AI model] were all done to solve a specific product need. We built Transformers and used it immediately in Search.”

Weighing the risks of public release
Releasing a public chat program carries entirely different risks than updating hidden search code.

New startup companies can afford to release unfinished products and survive the public backlash. Established brands face severe financial damage if a program invents facts or generates offensive material.

Pichai notes that while the team built early versions of these chat tools, the software lacked the necessary safety filters. He recalls that the early test version “was a lot more toxic. We couldn’t have possibly put it out at that time.”

Balancing fast updates with large operations

Holding back unfinished software allows the company to focus on its most critical operational metric: system speed.

Managing a massive user base requires engineering teams to operate under strict speed limits. Every design choice must preserve system speed because any delay frustrates users and costs the company money.

Pichai details these strict rules, saying, “If you ship something which shaves off three milliseconds, you earn 1.5 milliseconds for your latency budget, and 1.5 milliseconds gets passed on to the user.”

The physical limits of global expansion

Changing how people search the internet

The hurdles to building independent software