Sundar Pichai on A.I. Backlash, the Future of Work and Google's Next Era

ELI5/TLDR

Google’s CEO sits down for a friendly grilling about where Google stands in the AI race. His honest admission: Google is at the leading edge on most things, but a step behind on AI that writes code. He thinks the public’s fear of AI is understandable and even healthy, jobs will change rather than vanish, and the really world-altering stuff — machines that improve themselves — isn’t here yet but is coming. Throughout, he stays cautious and measured, refusing to say anything dramatic.

The Full Story

Where Google actually stands

The framing question was simple: three years ago Google looked like it was chasing. Where is it now? Pichai’s answer was unusually candid for a CEO. Google is at the “frontier” — the leading edge — in some areas and behind in others.

“When it comes to agentic coding with tool use, and or instruction following, long horizon tasks, I think we are a bit behind at this moment.”

Translate that: an “agent” is software you hand a goal to and it goes off and does the multi-step work itself — opening files, running code, fixing its own mistakes — instead of just answering one question. “Long horizon” means the task takes a long chain of steps rather than one. This is the kind of AI that actually writes and debugs real software, and Pichai admits rivals (Anthropic, OpenAI) are ahead here.

He has a theory about why. Coding ability improves when the AI gets used on real codebases and the results flow back as training fuel. Google, he says, simply did not have a popular coding tool collecting that fuel the way competitors did. Their new internal tool (codenamed “anti-gravity”) is changing that — usage is “doubling every week” inside Google.

He also flags how fast the leaderboard flips. Think of it like Formula 1 where every team upgrades its car on a different schedule, so whoever looks dominant this month can look ordinary next month.

“We are in a moment in time in this field where 30 to 60 days look like 5 years.”

The front door of Google is changing

The interviewers note Google just made the biggest change to its search page in 25 years. The fear (for Google’s wallet) is obvious: if everyone uses the chatbot-style “AI mode,” the classic list of “10 blue links” — the thing ads are sold against — fades away. One interviewer says he hasn’t done a normal Google search in a year.

Pichai stays calm. He won’t rip the band-aid off; he’ll “bring users along.” His bet is that as AI does more for you, the total value goes up, and money follows value whether through ads or subscriptions.

“Adam Smith’s rules don’t change in this new world.”

(That is shorthand for: markets still reward whoever delivers the most value — the economics don’t get suspended just because the technology is new.)

The backlash, and why he thinks it’s fair

A poll found only 16% of people call AI “mostly good” while 35% call it “mostly bad.” Pichai doesn’t dismiss this. He calls AI “the most profound technology humanity will ever work on” and says humans simply aren’t built to absorb change this fast.

“Humans aren’t evolved to process that much change.”

He even calls the public anxiety “healthy” — in a democracy you want citizens paying attention and pushing back, because that pressure is what makes society act. He pushes back only on what he calls “overly deterministic dire scenarios” — the assumption that doom is guaranteed.

Jobs: changed, not erased

His case for optimism leans on history. He compares AI to the spreadsheet — which raised the floor of what an ordinary person could do (he jokes he never learned how financial analysis was done before spreadsheets). AI, similarly, “changes the starting point.”

His favorite example is medicine. Doctors burn out partly because paperwork eats the time they trained to spend with patients. The radiologist case: he’s had far more scans than his father did, each carrying 10x the data, and that volume is heading another 10x. No human can keep up — so AI becomes the thing that lets demand get met at all, rather than the thing that takes the job.

Agents for regular people

Google’s upcoming consumer agent is “Spark.” Pichai’s own example is modest: he asked it to look at his calendar and color-code his meetings by category so he could see how he spends his time. It came back with two schemes and just did it.

“It’s actually like sci-fi.”

The interesting part is his caution. He compares trust in agents to getting people comfortable in a self-driving car — done in steps. If an agent does something unexpected, people bail. And there’s a real danger: these systems “can be hacked.” So Google deliberately won’t run “ahead of the frontier in a wrong way.”

Self-improving AI and the “break glass” moment

The biggest question: recursive self-improvement (RSI) — AI that gets better at building AI, potentially spiraling fast. Pichai says the ingredients exist (anti-gravity can build a simple operating system from scratch in 12 hours, work that would take a person thousands of hours), but true RSI “we aren’t quite there yet.”

Asked what happens if they hit it, his answer is the most pointed in the interview:

“It shouldn’t be an internal conversation at that point… we all have to avoid race conditions at those stages of AGI.”

A “race condition” here borrows from engineering — a situation where parties rushing against each other produce a dangerous, unpredictable outcome. He’s saying: at the threshold of human-level general AI, nobody should be sprinting alone.

On TPUs, AGI, and the singularity

Why does Google sell its custom AI chips (TPUs) to rivals like Anthropic? Because, he argues, chip-making and his own models aren’t a zero-sum trade-off as long as Google can make enough. More customers means more scale, which makes the next chip better — the standard logic of running a platform.

On AGI, he’s careful with language. He’s “AGI-pilled” in the sense that he’s sure the technology is making foundational progress, and recent progress makes him lean toward the “closer” end (3–5 years rather than 5–10) — but as the head of a company with social responsibility, he says he picks his words more carefully than others. Demis Hassabis used “foothills of the singularity”; Pichai interprets that simply as “the advent of AGI,” which Hassabis pegs around 2030.

Key Takeaways

Google admits it is behind Anthropic and OpenAI specifically on agentic coding (multi-step, autonomous software work), while claiming the lead or parity on text, voice, multimodality, and reasoning.
Coding ability improves with real-world usage data; Google lacked a popular coding tool to harvest that data, which it blames for the gap. Its internal tool “anti-gravity” is now closing it.
The classic “10 blue links” search won’t be removed abruptly; Google will migrate users gradually and bets that ads + subscriptions both survive as AI delivers more value.
Pichai frames public AI anxiety as both understandable and democratically healthy, while rejecting “deterministic dire scenarios.”
His jobs thesis: AI raises the baseline of what people can do (spreadsheet analogy) and meets exploding demand (medical scans), so roles change rather than disappear.
Google’s consumer agent is “Spark,” launching later in summer; trust is being built in steps, with hacking/security cited as a real constraint.
Recursive self-improvement (AI improving AI) isn’t here yet per Pichai; if reached, he says it must trigger a broad, cross-lab conversation, not a private one — explicitly to avoid a competitive “race” at the AGI threshold.
Google sells TPUs to rivals because scale improves its next chip generation; it also buys NVIDIA chips.
Pichai privately leans toward AGI on the “closer” (3–5 year) end but deliberately moderates his public language.

Claude’s Take

This is a polished, low-information interview — which is itself the story. Pichai is a master of saying things that are true, measured, and almost impossible to quote against him later. The genuinely useful nugget is his unforced admission that Google trails on agentic coding, with a coherent explanation (no data-collecting surface like Claude Code or Cursor). That’s a real, falsifiable claim, and it lines up with what outside observers have said.

Most of the rest is well-rehearsed reassurance. The jobs argument leans entirely on optimistic historical analogies (spreadsheets, radiology) and conveniently skips the obvious counter: those shifts still displaced specific people painfully, and “society should engage” is not a plan. The “we’ll avoid race conditions at AGI” line sounds responsible but is free to say when you’ve also just defended racing for compute and shipping agents on a quarterly cadence. The tell is the contrast he draws himself — Demis Hassabis says “singularity” and “2030”; Pichai launders it into committee-speak. He’s leaning bullish on timelines while keeping deniability.

Score 6: substantive on Google’s competitive position and a clear window into how a frontier-lab CEO talks in mid-2026, but heavy on managed optimism and light on anything you couldn’t have predicted. Worth it if you track the AI race; skippable if you want depth.