OpenAI CEO Sam Altman and Databricks CEO Ali Ghodsi sat down for a deep conversation about the future of enterprise AI and agents. The two companies have combined vast enterprise data with powerful AI models, signaling massive innovation ahead in 2026-2027. They emphasized that agents will evolve beyond simply writing code to autonomously handling complex business tasks, and stressed the critical importance of data foundations to make this possible.
1. Background of the OpenAI-Databricks Partnership
The conversation was moderated by Databricks CTO Hanlin Tang. The two companies recently announced a partnership making OpenAI's models natively available within the Databricks platform. Ali Ghodsi explained this was entirely driven by overwhelming customer demand.
Enterprise customers want to apply OpenAI's powerful models to their own corporate data, but they also want to maintain data privacy, auditing, and GDPR compliance. Satisfying both requirements simultaneously is quite challenging, and this partnership addresses that problem.
An interesting note: OpenAI is also a Databricks customer. Altman and Ghodsi both acknowledged they serve as demanding yet excellent customers for each other.
"You are a very demanding customer, in the best possible way. You've pushed us to become a much better company." (Ali Ghodsi)
2. The Enterprise AI Revolution and the Importance of 'Context'
Sam Altman candidly acknowledged that while AI models have been popular with consumers, they were not yet capable enough to handle complex enterprise tasks. However, as model capabilities mature, he predicted massive changes in the enterprise environment by 2026 and 2027.
He cited the coding revolution of 2025 as an example, suggesting that AI will soon handle a significant portion of intellectual labor across all enterprise functions. Ali Ghodsi emphasized that the most critical factor here is 'context'.
"AI has already learned all the publicly available data on the internet, but it cannot know a company's proprietary internal data. Providing that data as context to agents is the real solution." (Ali Ghodsi)
3. The Future of AI Research: Agents That Work for Longer Periods
The moderator asked where AI research is headed next. Sam Altman explained that beyond simply making models smarter, the key metric will be 'how long a model can sustain a task successfully'.
Back in the GPT-3.5 era, AI could only successfully complete 5-second tasks. GPT-4 extended that to 5 minutes, and at the GPT-5 level, models can now handle 5-hour tasks.
"How long is the task that a model can complete with a 50% success rate? In coding, for example, we started at 5-second tasks when GPT-3.5 launched, and now we're up to 5-hour tasks. That's truly remarkable progress." (Sam Altman)
But enterprise work can take days or months. Going forward, a key research direction will be deeply integrating AI into all enterprise data and processes so it can handle longer-horizon work.
Sam also added a striking observation about the current state of AI models:
"It always amazes me that today's model performance will be the worst I'll ever see for the rest of my life." (Sam Altman)
4. Governance and the Future of Open Source Models (GPT-OSS)
As AI agents start taking real actions within enterprise systems, responsible AI and governance have become more important than ever. Ali Ghodsi explained that enterprise AI requires robust guardrails and audit logs -- ensuring models don't recommend competitor products or make statements misaligned with brand values. Without these safeguards, AI adoption becomes impossible.
The conversation also touched on open-weight models, which are popular among security-conscious enterprises. Sam Altman stated that OpenAI plans to continue releasing open source models.
"I'd love to eventually release an open source model with GPT-5-level performance that can run on a personal device. I don't know how yet, but I intend to try." (Sam Altman)
Altman expressed his conviction that people will want high-performance models running locally, without internet connection, for privacy and freedom.
5. The Future 5-10 Years Out and Current Use Cases
Sam and Ali also discussed the long-term changes AI will bring. Ali pointed out that while we currently focus on coding, coding accounts for only 20% of an engineer's work. The remaining 80% -- meetings, documentation, planning -- will also see agents working alongside humans as colleagues.
Sam Altman predicted that the economic structure will undergo the biggest transformation 5-10 years from now. But he wasn't worried about human roles disappearing.
"Humans have this incredible 'main character energy.' They won't really care what machines are doing. GPT-5 is probably already smarter than most people, and nobody seems particularly bothered." (Sam Altman)
The most interesting current enterprise use cases involve tasks that humans would never have attempted. For example, AstraZeneca used AI to review 400,000 documents. Sam called this the "wouldn't do" category -- things companies never attempted before that are now possible thanks to AI.
6. Advice for Executives: What to Prepare Right Now
Finally, when asked what enterprise leaders should do immediately, both CEOs unanimously pointed to building data foundations.
Ali Ghodsi said that while it may sound boring, consolidating siloed data scattered across the organization is the single most important step.
"It sounds boring, but it comes down to data foundations. That data hidden somewhere in your enterprise is exactly the context AI needs. Only when that context is connected can AI understand concepts like 'revenue' or 'churn rate' the way your company defines them." (Ali Ghodsi)
They closed by agreeing on the "bitter lesson" -- that when building agents, it's better to let smarter models figure things out on their own rather than overengineering complex designs.
Closing Thoughts
This conversation demonstrated that AI is evolving beyond simple chatbots into agents that perform real enterprise work. The key factors are data integration and security. How well companies organize their data and provide it as context to AI will determine their competitive edge going forward. To prepare for the age of agents arriving after 2026, the time to start building data foundations is now.