Airbyte CEO: How ChatGPT changed the data integration game
AI needs data, lots of it.
Because we know all the truisms that circulate in this arena (i.e. garbage in, garbage out and so on), we need to pay particular attention to the provenance of our data for AI, the cleanliness and readiness of the data we feed into AI language models and execution engines… and, crucially, the way we approach data integration in this space, especially where AI “feeds” on data sources composed of more than one initial source.
Michel Tricot, CEO & Co-Founder at Airbyte thinks this is a big issue; his firm is known for its open source data integration, data movement and data activation platform that moves and consolidate data from various sources into data warehouses, data lakes and databases.
Tricot says that the rise of ChatGPT has driven a fundamental shift in the data integration game.
Why?
Because the core raison d’être of data movement has moved from a function focused on traditional analytics and dashboards to one where it is now deeply concerned with the need to feed AI agents and large language models (LLMs) with real-time context.
AI-native data integration
It’s the difference between data integration for “traditional” analytics and AI-native data integration for analytics at every conceivable level.
“We started Airbyte in 2020 to solve data movement. Back then, analytics was the main use case. In 2023, everything changed,” blogged Tricot. “ChatGPT hit the market and suddenly, every company needed to be AI native yesterday. The old playbook died overnight. Data movement wasn’t about dashboards anymore. It was about feeding agents and making models smarter with real-time context.”
It drove Airbyte to completely rethink how the company builds out its own roadmap and platform. The company had to sit back and think about a restructure that would deliver services directly tuned to AI teams need across unstructured data sources, metadata preservation and sovereignty.
Beyond rebranding, this is rewiring
Airbyte CEO Trictot: Becoming AI-native isn’t a rebrand… It’s a case of rewiring.
Tricot says that his engineers didn’t think about refactoring or enhancing the company’s core product; they also changed the way they work.
Today, they confirm that AI reviews their code, debugs customer issues and helps the team write documentation.
“Becoming AI-native isn’t a rebrand… it’s deeper than that, it’s a case of rewiring,” explained Trictot. “The last decode was all about structured data and looking for form and function where we could achieve it. The future is about data in context.”
Tricot recounts a brief history of structured data progress and reminds us that – in post-millennial times, if not before – we were focused on creating structured data (often for business intelligence use cases) that had perfectly clean schemas, transformed models, beautiful dashboards and so on.
The age of entity intelligence
“Companies spent millions of dollars de-siloing data and transforming information sources into formats to feed these use cases. But AI agents need something completely different,” said Tricot. “Context is entity intelligence. What’s an entity? It’s an object of interest, such as a customer, a product, or a company. When an agent needs to understand a customer, it has to read transactions, support tickets, sales call transcripts and contract terms etc.”
He says that just like a human student who might be penning (okay, typing) a history report by reading books, articles and Wikipedia, context in the age of AI is the integrated layer that connects all those disconnected bits of information about specific entities.
This is where he sees the future of data infrastructure.
