Writing
Notes from the native layer.
Three essays on the choice to ship an AI coding agent as a Zig binary instead of a Node or Python package, what that buys, and where it stops being the right call.
-
When systems languages start winning the agent layer
The first wave of AI agent SDKs is Node and Python. That made sense — the model APIs were Python-shaped and the audience was data scientists. The second wave is something else. Here is the case for why systems languages are an underrated answer for the agent layer.
systems-languageszigdesign -
C ABI as the universal AI SDK boundary
tokenworm's Python, TypeScript, and Go SDKs are all thin bindings over the same Zig-compiled shared library, called through a C ABI. That choice — to make the C ABI the source of truth — is the load-bearing decision behind the whole design.
ffic-abiarchitecture -
Why a 960KB binary changes the AI agent deployment story
Most AI agent SDKs ship as 200MB of language runtime plus their dependencies. tokenworm ships as a 960KB native binary that cold-starts in 8ms. That difference is not cosmetic — it changes which deployment shapes are tractable.
deploymentbinary-sizedesign