Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/ALL IN/Databricks CEO: AGI Is Already H…
POD
// EPISODE
ALL IN

Databricks CEO: AGI Is Already Here, Why Companies Are Behind, and the Superintelligence Debate

DATE November 17, 2026SOURCE ALL INPARTICIPANTS ALI GHODSI, CHAMATH PALIHAPITIYA, JASON CALACANIS
// KEY TAKEAWAYS6 ITEMS
  1. 01AGI Is Already Here
  2. 02The Four Enterprise AI Blockers: Context, Security, Cost, and Lock-In
  3. 03The Context Gap: "If It's Not in the Meeting, It's Useless"
  4. 04Org Charts Are Latency; AI Can Cut Across Them
  5. 05Company Ontology / "Management Moneyball"
  6. 06Remove Human Blockers: The Real Metric for AI-Native Operations
In this episode

1. Key Themes

AGI Is Already Here — The Bottleneck Is Diffusion, Not Intelligence

Ghodsi argues that by the definition used in the Berkeley AMPLab circa 2009–2010, AGI was reached last year, and the goalposts were simply moved. The practical implication is that even if model progress froze today, there is enough intelligence to transform the economy; the work is deployment.

"I actually think we already have artificial general intelligence. I've been saying it since last year." — Ali Ghodsi 00:09:36

"If you happen to stop all progress right now, I think we have all the intelligence we need to completely transform the planet." — Ali Ghodsi 00:10:06

The Four Enterprise AI Blockers: Context, Security, Cost, and Lock-In

Ghodsi frames why enterprises are stuck at "glorified Google search" and a bit of coding. He lays out four problems that must all be solved sequentially: capturing context, securing agent access, making token costs efficient, and avoiding lock-in/IP leakage.

"They're just using it as a glorified Google search. They're using chatbots all day." — Ali Ghodsi 00:12:18

"If the token cost is going up exponentially, is your revenue at your company going up exponentially? And I don't know any company who started using AI. And as of their use of AI, the revenue started accelerating exponentially, matching their cost curve." — Ali Ghodsi 00:15:50

"How do we avoid getting locked in? And how do we own our own IP? Because, you know, otherwise, you know, there's no business to be made in the future." — Ali Ghodsi 00:16:19

The Context Gap: "If It's Not in the Meeting, It's Useless"

The deepest blocker is that AI lacks the organizational context humans have. Ghodsi cites a bank CEO who refused to record meetings, which he sees as disqualifying for AI usefulness.

"If it's not in the meeting, it's useless. It doesn't matter if you have Terry Tao or your greatest mathematician in the world. If they're in the cold and they don't know what decisions or what you're doing... there's no way it can help you." — Ali Ghodsi 00:14:11

Org Charts Are Latency; AI Can Cut Across Them

Ghodsi argues organizations are built as trees, but value creation cuts horizontally (customer → product → engineering → marketing → sales). An AI that understands the entire business can collapse those handoffs.

"So you want to cut out that latency so that you can be really, really fast. But that requires that you have an AI that actually understands your whole business, your IP. It needs to understand your org chart. It needs to know who's who." — Ali Ghodsi 00:20:39

Company Ontology / "Management Moneyball"

The live demo of Genie shows an ontology graph of people, meetings, Slack, docs, and decisions powering CEO-level analytics: employee survey causal analysis, org span-of-control analysis, engineering productivity distributions, and culture drift by leader.

"The ontology is a graph that it constantly is constructing for your company. That graph understands all of the people, who reports to who, what are they working on, what are the emails that are coming in, what meetings have happened, and what's the transcripts of all those meetings." — Ali Ghodsi 00:23:00

"Get the culture of this subgroup in a 12,000-person organization. You know, figure out how the Japan organization at Databricks, I don't even speak Japanese... What's the culture like?" — Ali Ghodsi 00:37:00

Remove Human Blockers: The Real Metric for AI-Native Operations

Databricks measures "human blockers," the points where fast AI work stalls waiting on legal, finance, marketing, or sibling-org agreement.

"One of our main goals is reduce number of human blockers. So we measure that throughout the organization." — Ali Ghodsi 00:33:10

Superintelligence vs. RSI: Four Conditions That Aren't Being Met

Ghodsi rebuts the recursive self-improvement narrative by listing four necessary conditions (faster training runs, fewer resources, broad accuracy gains, and a sustained loop) and observing that none currently hold.

"That flywheel doesn't exist right now. The models are taking longer to train, right? And there's more data cleaning and there's more manual whack-a-mole-ing happening." — Ali Ghodsi 00:48:24

Cybersecurity Is the Real, Tractable AI Risk

He treats cyber as the genuine near-term danger due to attacker/defender asymmetry, but believes the vulnerability surface is finite and will be cleaned up as new code is scanned and legacy code rewritten.

"Once we've secured that code base, there is no more. So it's not like there's an infinite thing and this is going to go on forever. No, it then stops." — Ali Ghodsi 00:53:21

Staying Private: Avoiding the AI-Era Volatility Tax

Ghodsi explains why Databricks delays its IPO: stock volatility during the AI transition demoralizes employees and CEOs, while annual tenders deliver liquidity and acquisition currency.

"I think during this transition period of AI, there's just too much up and down. And I think it's distracting to companies. It's distracting to CEOs. It's demoralizing to CEOs." — Ali Ghodsi 00:43:16

Contrarian Founding Bets: Cloud, AI, and Open Source

Databricks' success came from three controversial bets made in 2013 that were all wrong in conventional wisdom at the time.

"We bet everything, the whole farm, on just being a cloud company. We bet everything on AI, the use cases of AI. And we bet on open and open source. And at the time, actually, all three of these bets were controversial." — Ali Ghodsi 00:06:42

2. Contrarian Perspectives

AGI Already Happened — The Debate Is Over Moved Goalposts

Most people treat AGI as a future milestone. Ghodsi says it was achieved last year per the definition held by the Berkeley AI community, and he verified this with former lab colleagues.

"I actually went and checked with some of the people that were in the lab. Like, hey, is it just me? Am I hallucinating? And they said, no, no, no. That definition for sure. But we've moved the goalpost." — Ali Ghodsi 00:10:06

98% AI-Written Code Does Not Mean Recursive Self-Improvement

The Silicon Valley existential-crisis narrative says AI writing its own code implies the takeoff has begun. Ghodsi rejects the inference: automating the final 2% is irrelevant if the training loop itself isn't accelerating.

"Just because 98% of the code is being written by AI, we don't have RSI. If it automates the last 2%, who cares? Does it matter if it's 98% or 100%?" — Ali Ghodsi 00:46:17

Enterprise AI Transformation Will Take Decades; GDP Moves Slowly

Contra the labs' "six to 12 months" timelines, Ghodsi predicts the context gap will take 10–15 years to close and GDP effects will take decades.

"The labs think this will happen like in the next six months, 12 months, in, you know, two years. But I actually think GDP will move very slowly and it's going to take decades for us to actually get through it." — Ali Ghodsi 01:00:20

Digital Natives Are Ahead; Incumbent Giants Are Structurally Stuck

Even the best 50-year-old Fortune 500 companies, loyal employees, and beloved products are structurally prevented from adopting AI by regulation, legacy software, and process.

"You have so much software. You know, you have so much regulation. And you have so many processes around security and so on. That it's very hard for you to adopt this." — Ali Ghodsi 00:21:24

Token Spend Is Not Correlated with Output

His engineering data shows the top committers aren't necessarily spending more on tokens; smart routing and open-source models can cut costs significantly while raising output, undercutting the "just spend more on AI" thesis.

"It's not even the case that cost necessarily is higher for those people... There is a way to spend tokens in a smart way. You can use open source, you can route simpler questions to simpler models." — Ali Ghodsi 00:30:13

3. Companies Identified

Databricks

Data and AI platform, private, ~$7B run rate growing ~80% (per Chamath). Mentioned as the subject; its bets on cloud, AI, and open source were contrarian and paid off. Ghodsi cites 2,000 engineers, 10,000+ survey respondents, and a recurring tender process.

"In 2015, they were doing $1.5 million in revenue. And it's August, or the rumor I heard, August, 2026, Ali, $7 billion run rate growing 80% year over year." — Chamath Palihapitiya 00:01:42

Genie (Databricks product)

Ontology-driven agent/assistant that builds a graph of the company's people, meetings, Slack, docs, and decisions.

"The key thing that Genie has... is that it has an ontology. The ontology is a graph that it constantly is constructing for your company." — Ali Ghodsi 00:23:00

Lakewatch (Databricks product)

Cybersecurity product built on lakehouse data with automated agents that detect intrusions and monitor customers' own agents.

"We actually have a whole product in the cyberspace called Lakewatch built on all this data that has automated agents that actually detect these kind of intrusions." — Ali Ghodsi 00:50:14

Block / Square (Jack Dorsey)

Cited as likely the furthest along at the "company world model" concept, matching what Ghodsi does at Databricks.

"He's calling it the company world model. So, I think that they're probably pretty far ahead." — Ali Ghodsi 00:21:24

Anthropic

AI lab praised for enabling intelligence while also flagged as expanding into adjacent products (Excel-like, Photoshop-like, dashboards) that threaten partners and customers.

"A company called Anthropic seems to be releasing every two weeks, a half dozen products that just now they're competing against Microsoft." — Chamath Palihapitiya 00:17:06

OpenAI

Named alongside Anthropic as a platform with data-leakage/lock-in risk for enterprises.

"If you use Anthropic, OpenAI, or any of these platforms, there is this risk that it will leak, that it will get incorporated into their next version." — Chamath Palihapitiya 00:17:06

Regeneron

Biotech that used Databricks to identify the genome response for chronic liver disease and derive drugs.

"We were involved with Regeneron where they found the actual genome response for chronic liver disease using Databricks." — Ali Ghodsi 00:57:09

Merck (TEDdy model)

Databricks partnered with Merck to train a transformer-enabled drug discovery model over gene regulatory networks.

"We trained a new one now with Merck and it's called TEDdy. Transformer-enabled drug discovery... instead of the next token being English, it's the gene regulatory network." — Ali Ghodsi 00:58:02

Crisis Text Line

Nonprofit using Databricks and LLM agents to detect self-harm risk in kids' texts.

"They actually detect if kids want to do self-harm and it helps prevent that, right?" — Ali Ghodsi 00:58:32

Virtue Foundation

Mentioned as a Databricks user in social-good use cases.

"Virtue Foundation uses Databricks." — Ali Ghodsi 00:58:58

Zipline

Drone delivery company that used Databricks to manage flight and delivery data.

"Zipline, this was a use case that was reported before, used Databricks to deliver food. They had these drones." — Ali Ghodsi 00:58:58

Mercor

Data vendor sourcing skilled-labor training data for labs; cited as evidence that model improvement still depends on manual, domain-specific effort.

"You have vendors like Mercore that has to actually go get the data from very skilled labor." — Ali Ghodsi 00:48:24

Snowflake

Named as Databricks' headline competitor.

"Obviously, you got a lot of competition, the snowflakes of the world." — Jason Calacanis 00:01:42

Netflix

Original use case for the Berkeley work (recommendations) and the Netflix Prize.

"The first use case, actually, it was the first use case ever was at UC Berkeley when we wanted to help Netflix recommend better movies." — Ali Ghodsi 00:03:33

Airwallex

Sponsor: AI-native platform for global accounts, cards, and payments.

"If you were building a global financial system from first principles today, you wouldn't build it on 50-year-old legacy rails. You'd build Airwallux." — Chamath Palihapitiya 00:00:44

OpenClaw

Cited as an example of an agent unleashed on all enterprise data, creating a hacking/insider-threat risk.

"You could unleash something like OpenClaw and just have it access all your data everywhere. Well, then you get into this problem that you're going to get hacked." — Ali Ghodsi 00:14:40

Replit and Cloudflare

Mentioned as companies making peripheral product moves into adjacent markets.

"Cloudflare is looking at features next to them. Replit is looking at features next to them." — Chamath Palihapitiya 00:41:43

Uber, Stripe, SpaceX

Referenced as examples of companies that stayed private for long periods.

"Stripe has been private closer to 20 years, I think, than 10... SpaceX took a little while. Uber took a little while." — Chamath Palihapitiya 00:38:54

4. People Identified

Ali Ghodsi

CEO of Databricks; former UC Berkeley AMPLab researcher. Mentioned as the guest who built a data/AI company on three contrarian bets and runs it with an ontology-driven AI.

"I had a standing bet with one of the most important analyst firms out there where they said, it's a strategic blunder. You're going to go under because you're betting on the cloud." — Ali Ghodsi 00:06:42

Nick Bostrom

Philosopher and author of Superintelligence (2014), who defined superintelligence and RSI; Ghodsi notes his views have moderated.

"Bostrom has kind of changed his view now. If you actually follow him now, he's actually not anymore." — Ali Ghodsi 00:45:44

Jack Dorsey

CEO of Block/Square, pushing the "company world model."

"Probably block and square with Jack Dorsey. Since he's pushing, he's calling it the company world model." — Ali Ghodsi 00:21:24

Dario Amodei

Anthropic CEO, who claimed cancer and Alzheimer's could be gone in under 10 years; Chamath cautions this gives false hope.

"Dario, you know, during the roadshow is saying cancer and Alzheimer's, I believe will be gone in under 10 years." — Chamath Palihapitiya 00:54:24

Ben Horowitz

a16z co-founder, who previously cited Ghodsi's framing of why superintelligence rollout can be controlled.

"Ben Horowitz... was like, the best explanation I've ever heard for why we are going to be able to control the rollout of superintelligence... came from you." — Chamath Palihapitiya 00:08:54

Terry Tao

Mathematician, used as an example of top intelligence that is useless without context.

"It doesn't matter if you have Terry Tao or your greatest mathematician in the world. If they're in the cold..." — Ali Ghodsi 00:14:11

Robert Morris

Author of the Morris worm that took down ~10% of the internet, later an MIT professor; example of past tech-risk that society absorbed.

"There was a kid called Robert Morris, who took down 10% of the internet with the worm... then he became a famous MIT professor." — Ali Ghodsi 01:01:07

Robin Dunbar (via "Dunbar's number")

Source of the 150-person coordination threshold Ghodsi uses to explain why orgs break at scale.

"It's somewhat easy up until Dunbar's number, which is 150. You can know everyone." — Ali Ghodsi 00:20:09

Masayoshi Son

Referenced as a splashy capital provider for long-private companies like Uber.

"Luckily, Masayoshi-san was splashy cashy." — Chamath Palihapitiya 00:38:54

Chamath Palihapitiya (host)

Described his own "major decisions" bot that scans Slack, Notion, and email to surface who actually makes decisions.

"I came up with a bot just called major decisions... give me a report on the most important decisions, who participated in them, and what was the resolution, what's not resolved." — Chamath Palihapitiya 00:34:27

Jason Calacanis (host)

Coined the "management Moneyball" framing.

"We're actually getting into like management money ball, aren't we?" — Jason Calacanis 00:36:22

5. Operating Insights

Pressure-Test Hypotheses with Live Data Instead of Debating in the Room

When the CFO speculated that work-life blend drove low "action" scores, Ghodsi queried the AI mid-meeting, which found correlation but not causation, and then sliced by department to find the real drivers (decision-making in product, tools in engineering, career growth in sales ops). Meetings stop being opinion contests.

"My CFO says... work-life blend is causing low action. And I'm like, that doesn't make any sense. But I just entered here immediately as we were discussing... it says, no, actually, it's not causal." — Ali Ghodsi 00:25:37

Measure Culture Per Leader and Region by Reading Their Actual Meetings

Instead of surveys, have AI derive the actual culture each leader implements and compare against stated principles (truth-seeking, first principles). The point isn't to enforce uniformity but to surface divergence.

"Maybe they're right and the rest of the organization, including me, are wrong. But then we need to, at least, it tells me that there's this divergence." — Ali Ghodsi 00:37:00

Diagnose Productivity Distributions, Not Averages

Plotting PRs per engineer revealed an "L-shaped" curve where a small minority running automated factories commit far more. Rather than add headcount or cut costs, the response was to push the rest toward the factory pattern.

"It has this L shape... the top people are actually committing way more... it turns out they're running automated factories. The rest are not. So we pushed them." — Ali Ghodsi 00:30:13

Pipe Customer Transcripts Directly into the Engineering Backlog

Automatically aggregate feature requests across all customer-meeting transcripts and route them to engineering, which then runs automated build factories. This removes the "game of whispers."

"Having it look at features that are requested from all of our customers automatically from all the transcripts... they're immediately going to engineering. And then we have factories in engineering that can build features automatically." — Ali Ghodsi 00:32:43

Use Tenders to Create Predictable Liquidity and a Signal on Employee Conviction

Annual tenders let employees take liquidity on a predictable schedule, and unused tender capacity signals bullishness; liquidity also underwrites billion-dollar acquisitions paid in equity.

"The money that we set aside for tenders, they never use up all of it. So most employees are bullish." — Ali Ghodsi 00:40:02

6. Overlooked Insights

The Scarce Resource Isn't Intelligence, It's Permission to Record

The bank CEO's offhand "we're never recording the meetings. Like not in a million years" is arguably the most important data point in the episode. It implies that liability and regulatory culture, not model capability, set the ceiling on enterprise AI value in the most valuable (largest, oldest) organizations. It also hints at a structural advantage for companies and jurisdictions willing to record everything, and an opportunity for governed-context infrastructure (the exact market Databricks sells into).

"One of the big bank CEOs told me the other day, no way. Like we're never recording the meetings. Like not in a million years." — Ali Ghodsi 00:14:10

The Founding Thesis Was a Bet Against the Researchers' Own Blind Spot: Governance

Buried in the origin story: the Berkeley research was technically strong but "what was missing... was that we didn't think about the security, the governance." The first years of Databricks were about making AI processing compatible with enterprise compliance, which is the same moat now re-emerging in the agent era (the second and fourth blockers). Governance, not raw capability, repeatedly turns out to be the monetizable layer, and it compounds: every agent connecting to Databricks "just send[s] us more queries."

"What was missing from our research at UC Berkeley was that we didn't think about the security, the governance." — Ali Ghodsi 00:04:30

"When they start using that, they just, you know, send us more queries." — Ali Ghodsi 00:19:26