AI adoption isn’t stalling because the models don’t work. It’s stalling because enterprises can’t get their data in order. Critical business information remains locked in silos, trapped in proprietary platforms, or subject to shifting vendor roadmaps. Without a reliable flow of high-quality, first-party data, AI systems simply can’t deliver. Michel Tricot, CEO of Airbyte, has been watching this play out firsthand. His company has built one of the fastest-growing open source data movement platforms, with over 1,000 contributors building and maintaining connectors.

As he puts it, “Our mission is to help our community and customers address every type of data silo they might have internally — and to make this data actionable.”

The Push for Data Sovereignty
One of the defining issues shaping enterprise AI today is sovereignty. Tricot explained that Airbyte’s open source foundation ensures customers retain full control over their data pipelines: “We provide the pipes. We will never see what’s in the pipes. The only thing we provide is the framework.”

That’s a sharp departure from legacy players like Informatica and Fivetran, which either demand heavy expertise or can’t adapt quickly to schema changes. In contrast, Airbyte’s community-driven model enables rapid adaptation to source changes—whether a CRM redefines a field or a SaaS tool updates its API. For enterprises that can’t afford downtime in their data flows, the difference is enormous.

AI Readiness Depends on Flexible Data Pipelines
The real test comes when organizations try to apply AI at scale. “The first thing companies look at is building a chatbot,” Tricot said. “But the moment you need context—your own first-party data—that’s where they struggle.”

Integrating proprietary data into AI workflows isn’t just a technical hurdle; it’s also a security concern. Enterprises increasingly demand that vendors commit to not training models on their data. For Airbyte, the open source model provides the assurance that “the risk, if one exists, is only limited to the customer and not to the vendor.”

The roadblocks usually appear when legacy architectures, designed for human operators, collide with today’s GPU-driven workflows. Organizations with decades of entrenched systems face the challenge of reinventing themselves, step by step. Tricot compares it to the migration to cloud in the late 2000s: “Nobody wanted to put their resources on AWS in 2007. But bit by bit, they started migrating small services—and it grew from there. AI will follow the same path.”

Developers, Flexibility, and the Open Source Advantage
For developers, the appeal of open source is even more direct. They can adopt and experiment without waiting on procurement or security reviews, then scale success across their organizations. As Tricot explained, “That’s one of the main reasons open source is taking over the world. Developers want to innovate. They don’t want to be talking to ten different teams to use a tool.”

This bottoms-up adoption model also means enterprises avoid hard vendor lock-in. While migrations are never painless, the ability to carry over configurations, credentials, and community-driven fixes dramatically reduces risk. For highly regulated industries—finance, healthcare, government—the assurance that sensitive APIs and internal data never leave their perimeter is decisive.

Ultimately, Airbyte’s story highlights a fundamental truth: AI without accessible, high-quality data is a nonstarter. “You never know which type of data is going to make that product successful,” Tricot reminded. The companies that win will be those that balance security with speed, building flexible pipelines that unlock innovation without exposing risk.

For CIOs and CTOs weighing their options, Tricot’s advice is straightforward: protect your company’s gold—its data—but do it in a way that empowers teams rather than blocking them. In the AI era, that balance may be the most strategic choice they make.