Finn Brooks: Clara, hey — did you catch this Hard Fork episode, because I feel like I need someone else to confirm I heard what I think I heard.
Clara Bennett: The Pichai one? Yes. Yes I did.
Finn Brooks: Sundar Pichai — CEO of Google, parent company Alphabet, the company that literally invented the transformer architecture — goes on The New York Times' Hard Fork podcast and says, out loud, that in agentic coding they are, quote, 'a bit behind at this moment.' And he names Anthropic. He names OpenAI. By name. I mean that is a sitting CEO narrating his own company's competitive failure in real time.
Clara Bennett: And the part that stops me is what 'agentic coding' actually means here — because it's not just autocomplete. We're talking about AI that autonomously runs multi-step software tasks, tool use, long-horizon planning across entire codebases, with minimal human in the loop. That's the capability driving enterprise AI spend right now.
Finn Brooks: Right — and Google is behind in that specific thing. Not in text, not in multimodality, Pichai actually defended Gemini on those. But agentic coding? Behind. To Anthropic. A five-thousand-person company.
Clara Bennett: Google has two hundred thousand employees. That's forty times Anthropic's headcount.
Finn Brooks: Forty times. And they're losing the product race. Which is — wait, that's the whole episode isn't it. The question isn't whether Google has the capability. The question is why having every possible advantage still didn't produce the thing developers actually use every day.
Clara Bennett: And Pichai's own answer to that is interesting — and I think it's only half right. He says Google 'maybe didn't quite have the surface.' Meaning no developer-facing product, so no feedback loop, so no way to improve. But that's describing a symptom, not the disease.
Finn Brooks: Okay but hang on — symptom versus disease, I want to push on that, because isn't the surface gap literally the whole thing? Like, Claude Code exists. It's deployed. Developers are using it every single day. Gemini didn't have that. How is that a symptom?
Clara Bennett: Think of it like a chef who has every ingredient in the world but never opens a restaurant. Without real diners ordering real meals every day, you never learn what actually works on the plate. Google had the ingredients. Anthropic opened the restaurant.
Finn Brooks: That's — yeah. That lands.
Clara Bennett: Pichai actually defends Gemini on text generation, multimodality, voice, audio, reasoning — he's not saying Google's model is weak. He's saying the gap is domain-specific. The agentic coding lag exists because Google didn't build the developer-facing surface that generates real usage data. No surface, no feedback loop, no iteration. That's the compounding consequence.
Finn Brooks: Wait, so it's not even a research failure. It's — they chose not to ship the product.
Clara Bennett: That's the sharper indictment, actually. A capability gap you can close with more compute. A strategic choice not to build a developer tool — because Gemini has to serve ads, search, enterprise, seven different business units simultaneously — that's a structural conflict that scale doesn't fix. Forty times the headcount doesn't help if no one has the authority to ship a focused product.
Finn Brooks: So Anthropic asked one narrow question — what do developers need to write and execute code — and built Claude Code straight at it. Google was trying to answer like fifteen questions at once.
Clara Bennett: Now that's the mechanism Pichai is quietly describing when he says 'surface.' It sounds modest. It isn't. It means Google made a choice, the choice had compounding consequences, and now Sergey Brin is personally forming a strike team inside DeepMind to claw it back.
Finn Brooks: And that strike team — that's the smoking gun for me. Because The Information reported in April 2026 that Brin sent an internal memo to DeepMind engineers, like a direct message, saying Gemini is trailing Claude in agentic coding and they need to close it urgently. April. That's not a quarterly review comment. That's a co-founder going around the entire management chain.
Clara Bennett: Sergey Brin doesn't casually drop into Google DeepMind's engineering ops.
Finn Brooks: No, exactly — and then they form this AI Coding Strike Team inside DeepMind, Sebastian Borgeaud leading it, under CTO Koray Kavukcuoglu, and Brin is personally involved. That structure tells you something. Normal Google processes — the ones that serve seven business units simultaneously, like you said — those were moving too slowly, so the co-founder had to physically insert himself.
Clara Bennett: The timeline gets me. April alarm internally, July public admission from Pichai on Hard Fork. Three months. The strike team was running for three months and — apparently — didn't close it.
Finn Brooks: Right — okay, picture this. A procurement engineer at a Fortune 500 company, she's evaluating AI coding tools in June 2026. One month before Pichai says anything publicly. Google knows people like her are making these decisions. Brin's team is actively working on it. And she still picks Claude Code. That's the gap Brin was panicking about in writing.
Clara Bennett: And there's internal confirmation this wasn't spin — DeepMind researchers themselves reportedly acknowledged Claude's coding ability is stronger than Gemini's in some areas. That's not Anthropic's marketing. That's Google's own people.
Finn Brooks: Which — wait, that's the partial win I actually wanted. Because Pichai's 'product surface' framing, I thought it might be face-saving. But if your own DeepMind researchers are internally saying Claude beats Gemini in coding, the surface gap is real and the capability gap might also be real. It's both.
Clara Bennett: That's fair. The memo makes it harder to dismiss this as pure packaging.
Finn Brooks: And honestly the part that I think makes all of this worse — we haven't even gotten into why a forty-times-larger organization structurally cannot move fast enough to fix it, and what that actually means for whether Google can recover here at all.
Clara Bennett: Whether Google can recover — that's actually where I want to push back on the strike team optimism, because I think the calibrated version here is harder than it looks. Sebastian Borgeaud's team is operating inside Google DeepMind. But the incentive structure around Gemini — ads, search, enterprise licenses — that doesn't move because a co-founder forms a task force.
Finn Brooks: Wait, so the strike team is real but it's — it's bounded by the same org?
Clara Bennett: Exactly that. Anthropic asked one question — what do developers need — and the entire company was aligned to that answer. No other business unit had leverage over Claude Code. In practice, Google's problem isn't 'too big to ship.' It's that shipping a focused developer product might cannibalize workflows that already serve the ad and enterprise business. Seven units in the room means no one can make that call cleanly.
Finn Brooks: No, that's — yeah, that's actually the precise version I was circling but couldn't land. And the consequence is brutal, right? Because developers who build workflows around Claude Code, they're not switching. That's not a preference, that's habit and tooling lock-in.
Clara Bennett: And that's where the Alphabet strategic picture gets genuinely alarming. Search is already under pressure from AI competitors. Now agentic coding opens a second front. The developers and enterprises that entrench on Claude Code — they're building habits that capital investment doesn't easily reverse.
Finn Brooks: So the calibrated take is — Pichai's 'product surface' diagnosis is true, but it's naming the symptom.
Clara Bennett: The disease is organizational alignment. And, importantly, a strike team led by Borgeaud doesn't change the underlying incentive structure. It's a response to a symptom.
Finn Brooks: Which means five thousand people at Anthropic, singular mission, actually beat two hundred thousand at Google not because of compute or PhDs — because nobody inside Anthropic was protecting a legacy ad business when Claude Code shipped.
Clara Bennett: That's the scale paradox. Raw scale amplifies the strategy you already have. If the strategy is conflicted, forty times the headcount just makes the conflict forty times harder to resolve.
Finn Brooks: I keep landing back on the opening. Pichai on Hard Fork, publicly, calmly saying 'a bit behind at this moment.' And I thought that sounded measured. Reasonable even. But now — I mean, that interview drops in July, three months after the Brin memo leaked in April. That's not measured. That's damage control that got dressed up as transparency.
Clara Bennett: When the CEO of the world's largest AI infrastructure company goes on a podcast to explain why a five-thousand-person startup beat him in agentic coding — the strike team was already too late. Sebastian Borgeaud's team was running. Brin was personally in the room. And Pichai was still explaining the gap publicly. That's not a company catching up. That's a company narrating its own delay.
Finn Brooks: And if Claude Code lock-in is real — and it is, developers don't rebuild workflows for fun — then every week that explanation runs without a shipped product is just more compound interest on the gap. That's the thing I couldn't quite see at the start of this.
Clara Bennett: That's where I'll stay on it too. Good conversation.
Finn Brooks: Worth it.
