All-In

Ali Ghodsi on Management Moneyball

Ali Ghodsi· CEO of Databricks at Databricks
·~62 min·English·All-In Podcast
AgentsAI InfrastructureBusiness StrategyOpen SourceAI Company
TL;DR

Databricks CEO Ali Ghodsi argues the intelligence we need already exists — the work now is unglamorous and measurable: run the company by querying an agent for ground truth, treat cyber as a finite cleanup, and let the doom math wait while real value diffuses over a decade.

01How He Runs the Company

Management Moneyball

<strong>Ghodsi now runs Databricks by asking an always-on agent questions about his own company and getting ground truth back</strong>, which replaces the CEO's old job of playing detective and refereeing the blame game.

now I'm learning stuff about the organization. This is amazing. It would have taken me like a week to get the data team to do this analysis for me

— Ali Ghodsi, All-In
Key Insight
The real shift is epistemic, not just faster reporting. When the CEO can pull ground truth on demand, authority stops flowing from who tells the most convincing story in the room — the agent removes the people from the account of what happened, so a decision no longer rewards the best spin.

02What the Data Reveals

The L-Curve

<strong>When Ghodsi plotted code shipped across 2,000 engineers it formed an L — a few people ship far more because they run “automated factories,” and they don't necessarily cost more.</strong>

we looked into this and it turns out they're running automated factories. The rest are not. So, we push them

— Ali Ghodsi, All-In
Key Insight
The L is a management instrument, not a trivia chart. Because the top output didn't track with the top spend, the lever isn't “buy more tokens” — it's to make the factory pattern the default and drag the whole curve rightward, which is exactly the diffusion problem stated inside one company.

03The Risk He Takes Seriously

The Fruit Is Finite

<strong>Ghodsi grants that cyber is genuinely asymmetric — an attacker needs one win while the defender must be perfect — but argues the supply of vulnerabilities is finite and running down.</strong>

new code that's being produced is not going to have these vulnerabilities

— Ali Ghodsi, All-In
Key Insight
The optimism is conditional, and the condition is the tell. The garden only empties if new code is actually scanned on the way in and legacy is actually rewritten — and scanning is a product Ghodsi's own company sells, so the forecast doubles as a bet that the market buys the mop as fast as the spill spreads.

04Why Still Private

Avoiding the Roller Coaster

<strong>Ghodsi keeps a $7B-run-rate company private not to deny employees liquidity — annual tenders already provide that — but to escape the public-market roller coaster during the AI transition.</strong>

We are going to be public but right now I think during this transition period of AI there's just too much up and down

— Ali Ghodsi, All-In
Key Insight
The hidden cost of listing is a morale one. In a volatile stock, equity comp becomes a promise that silently breaks every time sentiment turns — “the employees literally are getting a pay cut” — so staying private is less a capital decision than a way to keep a long bet from being repriced by the week.

05Where the Doom Comes From

A Coder's Existential Crisis

<strong>Ghodsi reads the takeoff panic as projection: the people building AI watched it start writing its own code and mistook that for recursive self-improvement.</strong>

Just because 98% of the code is being written by AI, we don't have RSI. If it automates the last 2%, who cares?

— Ali Ghodsi, All-In
Key Insight
The argument is as much psychological as technical. For engineers who coded since childhood, watching the machine take the craft is an identity shock — and Ghodsi's claim is that the fear of a runaway loop is that shock looking for a theory, not a loop anyone has actually measured.

06How to Read the Claims

Pick a Number, Sound Credible

<strong>Ghodsi's objection is to manufactured precision: a sharp number like a 10% chance of extinction lends false credibility the same way promising a cancer cure in ten years lends false hope.</strong>

when you pick a number, it makes you seem credible to an average Joe

— Ali Ghodsi, All-In
Key Insight
Notice he refuses the move in both directions. Asked whether cancer falls in ten years he answers “maybe 10, maybe 30, maybe 50 — I don't know,” modeling the honesty he wants from the doomers: a decimal is a rhetorical device, and a range is the real state of knowledge.

07The Win That's Already Real

A New Token for Biology

<strong>The concrete payoff Ghodsi points to isn't a god-model but a narrow one — TEDDY, a transformer whose “next token” is a gene regulatory network rather than a word.</strong>

this genome is actually causing this disease this genome is just reacting to it so don't target it for a drug

— Ali Ghodsi, All-In
Key Insight
This is the quiet counter to both the hype and the doom. The value Ghodsi can actually name comes from pointing a transformer at a specialized, proprietary data space — telling a causal gene from a reactive one — which argues the near-term frontier is many narrow models wired to private data, not one mind that wakes up.