SDK Telemetry Drives Coding Agents
Augusto Marietti, CEO of Kong, on the end of tokenmaxxing
This is really a distribution and feedback loop play, not an SDK revenue play. When a model lab owns the layer that turns its API into the code developers actually import, it gets closer to the real workflow, which endpoints get called, which errors recur, which languages matter, and where agents break in practice. That context is useful for improving coding agents because coding tools win by handling messy real developer behavior, not just by having the best base model.
-
Anthropic said on May 18, 2026 that it acquired Stainless to improve how Claude connects to data and tools, and Stainless had already powered Anthropic's official SDKs. That points to SDKs as strategic infrastructure around the model, not a standalone software line.
-
Stainless generated assets go beyond client libraries into CLIs, docs, and MCP servers, and its docs show generated requests can include platform telemetry headers and request metadata. That means owning the SDK layer can expose a steady stream of implementation level signals about how developers and agents actually use an API.
-
This mirrors the broader coding land grab. Research on Claude Code and OpenAI's pursuit of Windsurf frames coding products as valuable partly because they capture high frequency code edit behavior and direct user workflows, which can feed product iteration and model alignment faster than selling raw API access alone.
The next step is more vertical integration by model labs around the full developer path, from model, to API, to SDK, to CLI, to agent. As coding shifts from chat to tool use and autonomous execution, the companies with the richest first party usage loops around real software work should keep compounding fastest.