A Data+AI conversation with Jonathan Frankle, cofounder of MosaicML and Chief AI Scientist of Databricks

Data+AI in the Enterprise: From Research to Real-World Impact

When Jonathan Frankle co-founded MosaicML, the goal was clear: make AI training more efficient and accessible so that companies of all sizes—not just the hyperscalers—could train and fine-tune their own models. That mission resonated so strongly that Databricks acquired MosaicML for $1.3 billion, where it’s now playing out at an even larger scale.

But turning AI research into real-world impact is an entirely different challenge than making breakthroughs in the lab. AI adoption has moved beyond the hype cycle, and the real work—deploying, scaling, and refining AI in production—is well underway.

In a recent podcast conversation with Jonathan, we unpacked what it really takes to build and scale AI in the enterprise.

The Real Lessons in AI Adoption

For founders, builders, and investors, the biggest AI breakthroughs won’t just come from technical advances—they’ll come from solving real-world problems. And based on Jonathan’s experience scaling MosaicML and leading AI research at Databricks, here’s what he sees as the critical lessons:

1. Optimize for Learning, Not Just Validation

Enterprise customers provide credibility and long-term contracts, but they move slowly. Startups, on the other hand, move fast, iterate quickly, and give rapid feedback—which is critical for early-stage AI companies. The best approach? Target enterprises, but don’t depend on them.

Jonathan’s advice: Strive for enterprise customers, but don’t block on them. Early on, MosaicML partnered with fast-moving teams like Replit, who were willing to take a bet before Mosaic had a polished product. That kind of early traction is invaluable.

2. Stand for Something

The most successful AI startups don’t wait in stealth mode—they openly share their research, experiments, and philosophy.

MosaicML gained its first wave of customers before launching a product by blogging about their AI efficiency breakthroughs. They documented their process, explained their approach, and attracted customers who believed in their vision.

If you want great customers, start attracting the ones who need your product—before they even start looking.

3. Find Your First Customer & Nail the Storytelling

Getting one strong customer in a segment makes it dramatically easier to win others in that industry. Enterprises don’t like to take risks—but if they see a competitor successfully using your tech, they’ll start paying attention.

Every vertical has a different adoption curve. The first bank to work with you is the hardest win. The first healthcare company takes the longest. But once you land one, more will follow.

4. Some enterprises are (finally) deploying AI at scale today

Some enterprises are deploying AI at scale today. Others are still disillusioned after the hype spike of 2023. Founders need to identify where their customers sit on the adoption curve and meet them there.

Jonathan laid it out simply:

✅ Early adopters hit their AI hype peak in 2023, struggled in 2024, and are now deploying in production.

✅ The next wave of companies is in the “trough of disillusionment”—figuring out what works, what doesn’t, and what they actually need.

✅ Some companies still think AI will solve everything—they’re about to be disappointed, but in two years, they’ll be productive.

Founders should time their messaging and offerings accordingly.

5. Jump on a Plane & Meet Your Customers

The best insights don’t come from a slide deck or a demo. They come from sitting in your customer’s office, watching how they work, and understanding what actually matters to them.

Jonathan’s approach? If a customer has an interesting AI challenge, get on a plane and meet them in person. This hands-on approach helps you find the right problems, build better solutions, and strengthen relationships.

6. The Future Belongs to Problem-Solvers, Not Just AI Experts

Jonathan put it bluntly: The future of AI isn’t just in the hands of PhDs. It’s in the hands of people who find real problems and imagine solutions.

The biggest AI opportunities aren’t just in model-building—they’re in applying AI to industries like healthcare, finance, logistics, and education.

• OpenAI’s scale-driven approach is one path.

• Databricks’ data-driven AI approach is another.

• But in the end, AI is only valuable if it solves a tangible problem