Staying Small: The Strategic Choice Behind Claude Code and CData's Library Approach

When Anthropic built Claude Code, they made a choice that's easy to miss because it looks like not making a choice: they didn't build an IDE.

They could have. The obvious move for "AI for developers" is a big, opinionated environment: a new editor, a new surface, a place you're supposed to go. Instead, they did the simple thing. They put AI around the shell that developers already live in. No new home to move into. No workflow to adopt. Just intelligence wrapped around the most composable, durable tool in computing: the terminal.

It's the Unix instinct, really: do one thing, compose with everything, don't try to own the whole world. And it works precisely because it's small. A tool that meets you where you already are can go anywhere you go.

We've been making the same bet for over two decades. We just call it something less glamorous: the CData Driver library.

The unglamorous decision

The foundation of CData is the driver. A small, embeddable thing that speaks SQL to hundreds of data sources. It doesn't decide where you run the data operations. It wants to live inside your application.

That sounds modest. It's actually the most strategic decision we ever made, for the same reason Claude Code's was a small primitive that doesn't compete with its host can be embedded everywhere. Because we never tried to own the surface, every surface could adopt us.

And they did:

• It lived on the desktop, inside the analytics and reporting tools developers ran locally.

• It lived in server applications, powering data access behind the scenes.

• It lived inside ISV products, embedded by the software vendors whose names you know, shipping as the quiet connectivity layer underneath their own offerings.

• It lived in the cloud, as services moved off the desktop.

• And now it lives with AI, handed to a coding agent through a CLI.

Five eras of computing. One library. We didn't have to become a new company for each wave. The form factor around the driver changed; the driver kept doing it’s one thing well.

The wave wraps the primitive instead of replacing it

When the AI era arrived, we didn't rebuild the driver. We didn't need to. A well-made primitive doesn't get replaced by the next wave. It gets wrapped by a thin new surface that fits the era. Claude Code didn't reinvent the shell; it wrapped it. The CData CLI didn't reinvent the driver; it wrapped it: a small, terminal-friendly, AI-friendly surface over two decades of proven connectivity.

The query engine, the push-down, the schema, the SQL: all of that was already there, already battle-tested in tens of thousands of deployments. The CLI is just the adapter that lets an AI agent reach it through the terminal it already works in. Thin surface, deep foundation.

That's the advantage of having stayed small. You don't chase the trend. You let the trend come to you, and you greet it with an adapter instead of a rewrite.

And the CLI isn't the only surface this era asked for. MCP, the protocol for handing tools to an AI agent at runtime, is another. So we built one of those too, on the very same drivers.

Why small things survive

It's tempting to think the durable thing to build is the big thing: the platform, the destination, the moat made of surface area. The opposite is true. Surface area is what gets replaced. Big, opinionated environments are bets on a particular shape of the world, and the world keeps changing shape.

A small, composable primitive makes no such bet. It doesn't care whether the work happens on a laptop, a server, in someone else's product, in the cloud, or inside an AI agent's reasoning loop. It just needs to be embeddable, and being embeddable is the one property that never goes out of style.

Claude Code understood this: keep it small, keep it composable, meet developers in the terminal, and you can go anywhere the work goes. We've been quietly proving the same thing for two decades.

No matter where the technology trend goes next, and it will keep going somewhere, the library is small enough to follow it there.

That's the whole idea with CData Drivers. Keep them small. Go anywhere.

Build agent-driven workflows with CData

Start your 30-day free trial of CData Drivers and connect hundreds of data sources through one consistent SQL interface. Pair them with the CData CLI to hand that same connectivity to Claude Code and build your own agent-driven workflows.