Our customers use AI software agents: having a client instead of API docs speeds them along. A coding agent pointed at a client library has the method names, the required fields, and the error cases in front of it, written in the language it is already working in, rather than having to turn prose documentation into working requests and find out later which assumptions were wrong. Every repo ships worked examples next to the client itself for the same reason.
We chose these five languages because they're the ones our customers run in their production systems.
Now, engineers who build on the Gradient Labs API can read exactly how each library constructs and handles requests, pin to a version that suits their release cycle, and raise issues or pull requests directly rather than waiting on us to relay a fix.
Where the clients fit
If you run headless AI agents behind your own interface, the client library is the integration layer between your product and our agent. Everything from live conversations to outbound conversations, back-office tasks, and voice passes through it, and it's now yours to read and change.
What this means for the future of software development
A few years ago, when we released our Go client, we'd have thought about each subsequent client as an individual asset that needed to be created and maintained. Perhaps then we'd need to discuss it from an opportunity cost perspective. None of that needs to happen anymore.
Getting an AI agent to build a new client when we have reference implementations in other languages, and the implementation is fully verifiable with end to end testing, makes AI particularly well suited for this kind of build. A single AI coding agent can now implement one backend API change across all of our clients, the same shift we saw when our engineers adopted AI coding agents. If a prospect comes along with a new language, an AI agent can spin up a new client for them in short order.
That is the part worth sitting with, because it generalises well beyond client libraries. Work that used to be weighed up as an asset someone had to own and maintain for years is now work you can produce when it is needed, as long as you can verify that the result is correct.
If you're working out how a headless agent would sit inside your stack, book time with us
Share post
Copy post link
Link copied