All-In Podcast

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

17 Nov 2026 · 62 min · Episode page ↗

Artificial General Intelligence (AGI) is already here, but its transformative potential in enterprises is hampered by practical challenges in data access, security, cost, and intellectual property ownership, not by a lack of raw AI capability.

Summary

This conversation with Databricks CEO Ali Ghodsi explores the current state of AI, arguing that AGI is already a reality, but its real-world application in large organizations faces significant hurdles. Ghodsi contrasts the tech labs' pursuit of 'superintelligence' with the practical challenges businesses face in deploying existing AI effectively, offering a grounded perspective on AI's impact on productivity, organizational structure, and the economy, while also addressing common 'doom' narratives.

Interesting ideas

AGI is Already Here, But Not Yet Distributed. Ghodsi asserts that Artificial General Intelligence (AGI), as defined by early AI researchers, has already been achieved. He argues that the current debate around AGI and 'superintelligence' often moves the goalposts. The actual challenge isn't creating more intelligence, but rather enabling organizations to effectively harness and apply the AI capabilities that already exist, which are far from fully utilized.
Four Obstacles to Enterprise AI Adoption. Despite AI's power, enterprises struggle with adoption due to four key issues: 1) The 'context problem' of giving AI full access to an organization's internal data (e.g., meeting transcripts). 2) Cyber security risks associated with widespread AI access. 3) Exponentially rising token costs, making AI use financially unsustainable without efficiency. 4) The need for data sovereignty and owning AI-generated IP to avoid vendor lock-in. Solving these, not just raw AI capability, is crucial for impact.
AI as an Organizational 'World Model' and Blocker Finder. Databricks uses its 'Genie' AI, which builds an ontology (a 'world model') of the company, mapping people, data, communications, and decisions. This enables real-time insights into HR, engineering productivity (e.g., identifying automated coding 'factories'), and customer feedback. This shifts management from detective work and blame games to data-driven decision-making, highlighting organizational 'blockers' and accelerating processes across hierarchical structures.
Recursive Self-Improvement (RSI) is Not Currently Happening. Ghodsi refutes the idea of imminent recursive self-improvement (RSI) leading to 'superintelligence' and existential risk. He outlines four conditions for true RSI (faster training time, fewer resources, generalized accuracy improvement, continuous loop) and argues none are currently being met, citing the continued need for exponential hardware and human intervention in current model training. He suggests fears are often conflated with AI's coding ability and sci-fi narratives.
Cybersecurity as the Primary Real AI Risk. While dismissing 'paperclip maximizer' scenarios, Ghodsi identifies cybersecurity as the most significant and immediate risk from AI. He acknowledges the asymmetric advantage of attackers (needing only one successful exploit) but argues that AI will also enable better defense, by scanning new code for vulnerabilities and eventually rewriting legacy code. He believes this risk, though real, is manageable and will lead to a more secure codebase over time.

Questions worth thinking about

If AGI is already here, how should our societal and regulatory discussions shift from 'if' to 'how' we integrate it safely and equitably?
Given the enterprise AI adoption hurdles, what new business models or services will emerge to bridge the gap between advanced AI capabilities and organizational readiness?
If AI's current impact on GDP is slow despite its advanced capabilities, what structural economic or societal changes are necessary to unlock its full potential?

Written by AI from the episode's full transcript. It can get things wrong: if something here looks off, tell us at hello@podpings.com.

More from All-In Podcast

All 23 All-In Podcast summaries →