AI Enterprise - Databricks & Glean | BG2 Guest Interview
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
This BG2 podcast episode features Altimeter partner Apoorv Agrawal interviewing Ali Ghodsi (Databricks CEO) and Arvind Jain (Glean CEO) in a candid discussion about the current state of enterprise AI. The conversation dives deep into why 95% of AI projects fail, how LLMs are becoming commoditized, and where durable competitive advantage actually lies in the AI landscape. (02:15) The speakers explore three distinct camps in AI development: the super intelligence quest, academic researchers, and practical enterprise implementers focused on extracting real economic value from existing AGI capabilities.
- Core theme: Bridging the gap between AI's theoretical capabilities and practical enterprise implementation, with emphasis on data moats, workflow integration, and moving from the current 5% success rate to broader adoption
Speakers
Ali Ghodsi
CEO and co-founder of Databricks, a leading data and AI platform company serving thousands of organizations globally. Ghodsi previously worked in UC Berkeley's AMP Lab (Algorithms, Machines, and People), where he was exposed to early AI research and development that shaped his current perspective on artificial general intelligence.
Arvind Jain
CEO and co-founder of Glean, an AI-powered enterprise search and knowledge platform that recently crossed $200 million in revenue run rate. Jain has extensive experience in enterprise software and is focused on making AI practical and accessible for everyday business users through intelligent automation and personalized AI companions.
Key Takeaways
LLMs Are Commodities - Data Is Your Competitive Moat
Ali Ghodsi emphasized that large language models have become interchangeable commodities, similar to getting gas from different stations. (06:57) The real competitive advantage lies in your company's proprietary data and unique business processes that competitors can't replicate. Companies succeeding in AI are those leveraging their exclusive datasets and workflows rather than relying on generic model capabilities. For example, Royal Bank of Canada built agents that analyze earnings reports using their proprietary financial data and processes, reducing equity research time from two hours to fifteen minutes.
95% Failure Rate Reflects Necessary Experimentation
Rather than viewing the high AI project failure rate as problematic, both leaders frame it as evidence of healthy experimentation. (02:33) Ghodsi notes that if all projects succeeded, companies wouldn't be pushing boundaries hard enough. The 5% that work are those that focus on company-specific data and processes rather than generic AI applications. Success comes from rigorous engineering, proper evaluation systems, and building solutions that truly understand your business context.
We Already Have AGI - Focus on Making It Useful
Ali Ghodsi argues that we already possess artificial general intelligence by the standards defined 30-40 years ago, but the goalposts keep moving. (21:39) Instead of waiting for super intelligence, companies should focus on expanding the current 5% success rate to 10%, 20%, and beyond by solving actual enterprise problems. Current LLMs can reason and perform tasks smarter than many humans, making the focus on practical implementation more valuable than pursuing theoretical super intelligence.
Proactive AI Will Drive Mass Adoption
Arvind Jain highlighted that the future belongs to AI products that understand users deeply and proactively bring value to them, rather than requiring users to come to the product. (41:31) This paradigm shift from reactive to proactive AI will move adoption from 5% power users to 100% of employees. Glean's vision involves creating personal AI companions that know your work context, goals, and needs, automatically handling tasks before you even ask.
Speech Interface Will Replace Keyboards
Ali Ghodsi is particularly bullish on speech as the future interaction method, believing keyboards will largely disappear. (40:03) Despite feeling like we've solved speech recognition, the continued reliance on keyboards indicates we haven't truly nailed natural voice interaction yet. This shift represents a fundamental change in how humans will interact with AI systems, moving from typing commands to having natural conversations.
Statistics & Facts
- 95% of AI projects fail according to MIT research, but the speakers view this as healthy experimentation rather than a problem indicator. (02:07)
- Royal Bank of Canada reduced equity research report generation from 2 hours to 15 minutes using AI agents that analyze earnings reports and market data automatically. (04:07)
- The software industry generates approximately $400 billion in revenue, while AI infrastructure spending approaches $500 billion, creating a physics problem requiring about $1 trillion in AI revenue to justify current CapEx investments. (16:01)
