Skip to content

All-In Interview Recap: Databricks CEO Ali Ghodsi on AGI, Enterprise AI and Staying Private

Key takeaways and a timestamp guide from the Nov 17, 2026 All-In interview with Databricks CEO Ali Ghodsi: why he thinks AGI is here, why big companies lag, a live AI demo, the IPO question and the doom debate.

By the GNOUP team Β· Β· 8 min read

The All-In Podcast released a full-length interview episode on November 17, 2026, "Databricks CEO: AGI Is Already Here, Why Companies Are Behind, and the Superintelligence Debate". It runs about an hour. Host Jason Calacanis talked one-on-one with Ali Ghodsi, co-founder and CEO of the data and AI company Databricks, as part of the show's interview series rather than its usual weekly news roundtable. Here is a plain-English summary of what they covered, plus a timestamp guide so you can jump to the parts you care about.

This is an independent summary written by GNOUP. It is not affiliated with, sponsored by or endorsed by the All-In Podcast, Databricks, or any host or guest, and GNOUP does not endorse any company or product mentioned. The views described below belong to the speakers. Nothing here is financial or investment advice.

Key takeaways

  • Born in a Berkeley research lab. Ali said Databricks grew out of AMPLab at UC Berkeley, where about 20 researchers worked on combining data processing with AI. He said the industry wasn't ready to adopt their work, so the team started the company in 2013 partly "out of despair", at a time when he said there was almost no commercial interest in AI. He said the group's first real use case was helping Netflix recommend movies.
  • Three contrarian bets. Ali said the company bet everything on the cloud, on AI use cases and on open source, and that all three were controversial at the time. In his view, being early on those bets is what gave Databricks its head start.
  • "AGI is already here," by his definition. Ali argued that, by the definition of artificial general intelligence his lab used around 2009 and 2010, AGI was reached last year, and that the goalposts have since moved. He said that even if all AI progress stopped today, current models would be enough to transform the world. He separated this from "superintelligence" in the sense used by philosopher Nick Bostrom's 2014 book, which he described as something far beyond today's systems.
  • Big companies are using AI in basic ways. From his meetings with customers, Ali said most large organizations still use AI mainly as a "glorified Google search" and for coding help, with coding speedups he put at roughly 20 to 30 percent rather than dramatic gains. He said very few are running large fleets of AI agents, and people still need to stay in the loop.
  • Four obstacles for enterprises. Ali listed what he sees holding large organizations back: giving AI enough context (he said one big-bank CEO told him the bank would never record its meetings), cybersecurity and insider risk once agents can reach company data, rising AI usage costs that aren't matched by rising revenue, and the strategic worry about lock-in and protecting a company's own data and know-how.
  • No rush to go public. Asked when Databricks will IPO, Ali said the company will go public eventually, but that during this period of AI-driven swings he would rather avoid the stock-price roller coaster that can distract and demoralize employees. Jason opened the show citing a reported run rate of about $7 billion, and Ali described the business as growing roughly 70 to 80 percent a year. These figures are as stated on the show; Databricks is private and GNOUP has not verified them.

Newsworthy insights

These are claims and opinions voiced on the show, not independently verified findings by GNOUP, unless noted.

Running a company with AI: the live demo

  • Ali screen-shared a Databricks product called Genie that he said builds an "ontology" of the company: a constantly updated map of people, reporting lines, meetings and their transcripts, email, chat messages and documents.
  • He walked through an example from an internal employee survey with more than 10,000 responses. He said his finance chief suspected work-life balance was driving low scores on whether leadership acts on feedback, and the tool showed a correlation but argued it was not the cause, pointing to a manager whose team reported hard work but high confidence that their concerns were acted on. Slicing by department, he said, surfaced different complaints for product, engineering and sales operations.
  • He also showed a chart of code shipped by Databricks' roughly 2,000 engineers that formed an "L" shape. He said the top contributors were running automated coding "factories" and that they were not always the biggest spenders on AI usage, which he took as a sign that costs can be managed by routing simpler questions to cheaper or open-source models.
  • Ali said one company goal is to reduce "human blockers": waits for legal, finance or another team's sign-off. Jason described building his own bot that tracks major decisions across his team's chats and documents, and the two compared it to "management Moneyball" for spotting who actually moves work forward.

Staying private and the competition

  • Ali listed what he sees as the upsides of being public: easier access to capital and a quarterly stage to tell the company's story. He said Databricks instead runs a yearly tender offer for employee share sales, and that employees have never used up all the money set aside for it, which he read as a sign most staff are optimistic.
  • Jason noted that well-funded AI labs, including Anthropic, are launching new products at a rapid pace and moving into other companies' markets. Ali said Databricks will keep expanding into adjacent areas and mentioned recent moves into security and marketing.

The superintelligence and "doom" debate

  • Ali said he believes much of the fear of a runaway "recursive self-improvement" loop comes from AI researchers watching AI take over coding. He argued that AI writing 98 percent of code does not by itself mean AI is improving itself.
  • He laid out four conditions he thinks would all have to hold for such a loop: each new model trains much faster than the last, uses fewer resources, gets better across the board rather than in narrow areas, and keeps doing all three in a repeating cycle. In his view none of this is happening now, since frontier models are taking longer and costing more to train and still need a lot of hand-tuned, domain-by-domain work.
  • He called cybersecurity the biggest real risk today, while arguing it is manageable: he said new code can be scanned for flaws as it is written and that existing software has a finite number of hidden bugs that can gradually be cleaned up. He acknowledged that attackers have an advantage because they only need to succeed once.
  • Asked about the effective-altruism community and bold predictions from AI leaders, Ali said people on all sides believe they are the good guys and that he understands why Bostrom's ideas worry people, but said he sees no evidence of runaway self-improvement right now. On whether AI could cure major diseases within a decade, he said he didn't know, noting that new treatments still have to go through human trials.
  • He pointed to health research as a real upside, describing Databricks' work with Regeneron on the genetics of chronic liver disease and a newer model built with Merck aimed at speeding up drug discovery.
  • Ali predicted that spreading AI's benefits across the economy will take something like 10 to 15 years, slower than many AI labs expect.

Watch the episode

Databricks CEO: AGI Is Already Here, Why Companies Are Behind, and the Superintelligence Debate on YouTube (All-In Podcast's official channel, published November 17, 2026)

Episode guide

Chapter times are taken from the episode's own YouTube description. Select a time to jump to that point on YouTube.

  • 0:00: Ali Ghodsi joins the show: Databricks' early bets on cloud, AI and open source
  • 8:50: Is AGI already here? Why enterprise AI is falling behind
  • 22:04: Live demo: how Databricks uses agentic AI for efficiency
  • 38:25: Databricks IPO: the case for staying private
  • 44:02: Recursive self-improvement, cyber risk and the doom debate

About this summary

This recap was prepared from the episode's publicly available transcript text and chapter list. Transcripts can contain errors, especially with names and product names, so figures and wording may differ slightly from the audio; watch the episode for the exact words. Company figures, plans and research results described here are the speakers' own account. Nothing here is financial or investment advice, and mentions of companies are not recommendations to buy or sell any security.