Apr 7, 2026 · 1h 9m · cheeky-pint
The history and future of AI at Google, with Sundar Pichai
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Alphabet CEO Sundar Pichai joins John Collison and Elad Gil to reflect on his decade leading Google, detailing the company's full-stack AI strategy, custom TPU silicon, and the expanding role of autonomous agents. He also explores capital allocation across moonshots like Waymo and quantum computing while evaluating physical supply chain constraints and macroeconomic shifts.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. John holds 25.2% of the talking time here. How this is scored →
John as informed peer 5.5 Guest teaching 2.9 Guest disagreement 1.6 John pushing back 2.2
speaking balance: gold is John, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 17:11 Pichai dismisses the idea Google lacks AGI conviction
Pichai rejects the premise that Google is less 'AGI pilled' than rival labs, pointing to $180B in CapEx and early foundational work by pioneers across Google Brain and DeepMind.
Hardest push from John ▶ 34:41 Collison pushes back on total software vulnerabilityCollison directly interrupts and challenges Pichai's sweeping claim that AI models will break all existing software, citing battle-tested protocols like SSH.
Biggest teaching moment ▶ 0:34 Pichai explains the product origins of TransformersPichai corrects the common misconception that Transformers were an isolated pure-research byproduct, revealing they were engineered directly to solve Search translation and latency issues.
John holds their own ▶ 42:04 Collison articulates capital allocation hurdlesCollison demonstrates sharp financial literacy by laying out a textbook IRR framework and contrasting traditional industrial budgeting against Google's heterogeneous bet sizing.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | John as informed peer | Guest teaching | Guest disagreement | John pushing back | Why |
|---|---|---|---|---|---|---|
| Opening Welcome and Sundar Pichai's Decade as CEO | 4 | 5 | 3 | 2 | John Collison opens by framing Google's transformer invention as being productized elsewhere, which Pichai immediately reframes by explaining how transformers and TPUs were built directly for internal search and product needs like BERT and MUM. | |
| Product Latency, Budgets, and the Pareto Frontier | 6 | 3 | 1 | 1 | Collison draws on historical product knowledge regarding Google's emphasis on speed and latency in Search, Gmail, and Chrome, inviting Pichai to detail Google's millisecond latency budget accounting and Pareto frontier trade-offs. | |
| The Evolution of Search into Autonomous Agent Managers | 5 | 3 | 2 | 3 | The hosts press on whether traditional search boxes will be replaced entirely by autonomous agent managers within a decade, prompting Pichai to reject the zero-sum framing and explain asynchronous deep research queries. | |
| Market Sentiment Turnaround and Full-Stack AI Strategy | 5 | 2 | 1 | 2 | Collison analyzes the market sentiment shift from 2025 regarding Google's search vulnerability to recognition of its full-stack AI advantages. Pichai elaborates on custom TPU iterations since 2016 and multimodal Gemini design. | |
| Evaluating the AGI Mindset and Exponential Progress | 5 | 4 | 3 | 3 | Elad Gil brings up outside researcher critiques that Google is not sufficiently 'AGI pilled'. Pichai firmly pushes back by citing Google's $180 billion CapEx and pointing out foundational talent like Demis, Jeff Dean, and Ilya Sutskever. | |
| Executive Product Dogfooding and AI-Assisted Feedback | 5 | 1 | 1 | 1 | Collison shares Stripe's internal 'walk the store' ritual, prompting Pichai to explain his personal dogfooding of Gemini Live and using internal agentic tooling like anti-gravity for employee sentiment synthesis. | |
| Macroeconomic Impact and Real-World AI Diffusion Constraints | 4 | 4 | 2 | 2 | Gil asks about macroeconomic GDP realization from AI, referencing historic lags. Pichai discusses natural societal dampening mechanisms, compute curves, and regulatory pacing seen in Waymo deployments. | |
| Sponsor Segment: Stripe Network Intelligence | 6 | 2 | 1 | 2 | After an ad read on Stripe network intelligence, Collison drills into physical bottlenecks preventing even larger CapEx expenditures, leading Pichai to detail wafer capacity, memory constraints, and permitting delays. | |
| Compute Oligopolies, Open Models, and Cybersecurity Risks | 6 | 3 | 3 | 4 | Collison questions whether memory shortages limit industry expansion like the Strait of Hormuz oil bottleneck and challenges Pichai's claim that AI will break all existing software platforms. | |
| Frontier Bets: Space Data Centers, Quantum, and Robotics | 4 | 4 | 1 | 1 | Gil and Collison ask Pichai to reveal overlooked long-term frontier bets, leading to a deep dive into orbital space data centers, quantum simulation for molecular biology, and multimodal robotics partnerships. | |
| Capital Allocation Frameworks and the Journey of Waymo | 7 | 3 | 1 | 3 | Collison formulates a detailed corporate finance framework comparing heterogeneous project IRRs across YouTube, Waymo, and fundamental AI, asking how Google chooses which bets to kill versus sustain. | |
| Balance Sheet Leverage and ML Compute Budgeting | 7 | 2 | 1 | 3 | Collison challenges Google's historically under-levered balance sheet and questions the shift from headcount-based R&D budgeting to granular weekly ML compute allocations. | |
| Google Cloud MCP, Persistent Agents, and Docs Search | 6 | 2 | 1 | 2 | Collison provides granular product feedback praising the GCP Model Context Protocol before detailing specific search deficiencies in Google Docs compared to Gmail. | |
| Enterprise Workflow Shifts and the 2027 Agent Inflection | 7 | 3 | 1 | 2 | Collison outlines a structured multi-part framework on AI diffusion barriers in large enterprises, leading Pichai to forecast 2027 as the pivotal year for fully autonomous non-engineering workflows. |
Statements from this episode (32)
Pichai: BERT and MUM drove Google Search's largest quality leaps
“BERT and MUM, people underestimate how much, because we measure search quality so religiously, some of the biggest jumps in search quality in that period where search went ahead of everyone else was because of BERT and MUM. We built transformers and used it im…”
Pichai: Google held back LaMDA due to lack of RLHF and toxicity
“In fact, in the Google I.O. In, in 22, We launched something called AI Test Kitchen, and that was Lambda, but we had constrained it because internally we didn't have an end-to-end version which was RLHFed, right? So the version I saw was a lot more you know, t…”
Pichai: Garage startups will not ship a better iPhone
“I don't think people like wake up in a garage and ship a better iPhone. Like that's not going to happen, right? But that's not how consumer internet is.”
Pichai: Google enforces millisecond-level latency budgets across Search engineering sub-teams
“You know, I was speaking with the teams, right, like they now have for sub teams, like latency budgets, like in the milliseconds. You'll get 50% credit if, so if you ship something which, you know, shaves off three milliseconds, you earn 1.5 milliseconds for y…”
Pichai: Google improved Search latency by 30% over five years
“You know, last I checked the dashboards and the metrics, we've actually improved search latency by 30% in the last five years, but think about the functionality progression that's happened.”
Pichai: Gemini Flash models reach 90% of Gemini Pro capability
“The flash models are at like 90% the capability of the pro models but much faster much more effective to serve and the vertical integration helps and so on.”
Pichai: Google Search will evolve into an AI agent manager
“A lot of what are just information seeking queries will be agent taking search, You will be completing tasks. You have many threads running.”
Pichai: Google was behind in frontier LLM models
“So to me, we were behind in terms of frontier LLM models. But we had all the capabilities internally, and we had to execute to meet the moment.”
Pichai: Google designed Gemini models to be natively multimodal from day one
“I think we paid a bit more of a fixed cost upfront, but we designed the Gemini models to be very multimodal from day one.”
Pichai: Only two to three labs push the AI frontier
“I think there are two to three labs who are pushing each other pretty vigorously.”
Pichai: Google scaled capital expenditures from $30 billion to $180 billion
“Look, I think, you know, we probably have scaled our capex from thirty billion to a hundred and approximately a hundred and eighty billion.”
Pichai: Google shares the same foundational AI outlook as rival labs
“There are a lot of small attributes, which can probably make a difference, but I don't think at a foundational level, there is a difference in outlook on what the curve is.”
Collison: Stripe runs a weekly all-hands segment dogfooding its dashboard
“We do at our, like, little weekly All Hands Weaver recurring segment of Just Walk the Store, where we, like, click around in the dashboard together, and we're tripping over, like, why is that modal there, and that's a bit confusing or whatever, just so we're, …”
Pichai: Google CEO uses internal AI agent to summarize product launch feedback
“Internally, like, I would go fire to our earlier part. Like, I would query in anti-gravity, just our internal version of anti-gravity. Hey, we launched this thing. Like, what did people think about this? Tell me that, like, the worst five things people are tal…”
Pichai: Google Is AI Supply Constrained Amid Broad Demand
“To be very clear, you know, we are supply constrained. We are seeing the demand across all the surface areas.”
Gil: AI Coding Supply Could Expand the Software Market 10x
“And to some extent, I think that market has been so demand constrained for great software engineers that suddenly adding supply can 10 X that market, right? In other words, I think the market for software engineering and coding is dramatically bigger than anyb…”
Pichai: AI Will Meaningfully Accelerate US GDP Growth
“So, but I think the US economy is so much larger than it was 10 years ago. So to grow that, even at a half a percentage point higher than, you know, that, that's a massive contribution. So I expect it to play out that way.”
Pichai: Leading memory makers cannot rapidly expand short-term capacity
“There is no way that the leading memory companies are going to dramatically improve their capacity. So you have those constraints in the short term, but they get more relaxed as you go out.”
Pichai: Gemma 4 is based on Gemini 3 architecture
“Of Gemma four is based on Gemini three architecture.”
Pichai: AI models will break security across all regular software
“These models are definitely, like, really gonna break pretty much all software out there. Maybe already we don't know if we sit here and speak.”
Pichai: Google is exploring deploying data centers in space
“We're in the earliest stages of thinking about data centers in space right, but to your earlier discussion on constraint inspires creativity.”
Pichai: Quantum computing will have an advantage in simulating reality
“My instinct tells me there'll be, you know, simulating weather, simulating, you know, reality, all that, I think, quantum 11 advantage.”
Pichai: Gemini has state-of-the-art spatial reasoning models for robotics
“The Gemini robotics models are sort of on spatial reasoning, et cetera, so we definitely have state of the art models there.”
Pichai: Google is partnering with Boston Dynamics and Agile on robotics
“And we are partnering, ah, back in an ironic way with Boston Dynamics and Agile and a few other companies and in a determined way, making progress.”
Pichai: 40 million Americans will soon have Wing drone delivery access
“In some reasonable time period, like, forty million Americans will have access to a Bing delivery service, right? And I'm not talking years out or something like that.”
Pichai: Alphabet increased Waymo investment 2-3 years ago during industry pullback
“Waymo was a great example where I think we increased our investment two to three years ago when the rest of the world got pessimistic on it. When others, some of the people were backing off.”
Pichai: First-party hardware will be very important in regulated AI and robotics domains
“My lesson from Waymo and on the AI side with TPUs, et cetera, I think to really push the curve well, particularly in areas where you have safety, regulatory, everything, you want the firsthand experience of the product feedback cycle. So I think having first p…”
Pichai: CEO spends a dedicated hour weekly reviewing granular compute allocation
“I at least spend a dedicated hour a week thinking about that question at a pretty granular level.
So I will know by projects and by themes the compute units they are using, right?
And, you know, or at least I have that information
and I'm looking at it and …”
Collison: AI agents indexing docs are the ideal interface for complex APIs
“As we add more functionality to it, just the right way to navigate this big product surface area is an AI that's read all the API docs for you.”
Pichai: Consumer AI interfaces will run full coding models underneath
“Look, I think effectively the consumer interfaces are going to have Full coding models underneath, right? And the right harnesses and like the right scales and the ability to persist and run somewhere securely in the cloud, locally and in the cloud. So all tho…”
Pichai: Google rolled out agentic tool 'Jet Ski' to Search team
“For some strange reason, we have a different name internally than externally of the same product, but it's jet ski internally which is anti-gravity. And you're living on it. You're living in an agent manager world, your workflows, and you're kind of working in…”
Pichai: 2027 will be a profound inflection point for agentic workflows
“I definitely expect in some of these areas, 27 to be Important inflection point for certain things. Even the people doing it, that is the workflow through which they would produce it. And, you know, maybe for a while, you would check it in the conventional way…”