Zeke Sikelianos is an open-source AI engineer at Cloudflare whose work has shaped developer infrastructure at Heroku, npm, GitHub, Electron, and Replicate. He builds tools that make complex platforms usable by developers and, increasingly, the language models writing software alongside them.
At Wordnik, Sikelianos contributed to early Swagger tooling for interactive API documentation and JavaScript clients. At Heroku from 2012 to 2014, he redesigned the add-ons marketplace, improved Node.js deployment, and created semver.io to resolve Node versions.
He joined npm in 2014 as an early employee and helped rebuild npmjs.com. His projects explored package discovery and publishing automation, including an automatically updated inventory of npm package names.
At GitHub from 2016 to 2021, Sikelianos worked on Electron, helping transform its documentation into structured API descriptions that supported editor integrations and TypeScript definitions. He founded GitHub’s first documentation engineering team, helped launch and open-source docs.github.com, and contributed to the OpenAPI description for GitHub’s REST API.
Sikelianos joined Replicate in 2021 as its first employee and founding designer, working across its model playground, HTTP API, software libraries, documentation, and developer advocacy. He helped make generative models accessible without managing GPU infrastructure, including through a Stable Diffusion API, and created the image-editing experiments Inpainter and Outpainter.
What he builds for AI-assisted development
- Machine-readable documentation is product infrastructure. Sikelianos advocates Markdown documentation, llms.txt, concise cURL examples, and complete OpenAPI schemas that let coding assistants understand authentication, endpoints, and model behavior. At Replicate, that included model pages exportable as Markdown and consolidated documentation for Cog, its open-source model-packaging tool.
- Strong APIs make useful agent integrations possible. Sikelianos coauthored Replicate’s hosted Model Context Protocol server, connecting coding tools to model discovery and execution. He argues that reliable MCP integrations depend on comprehensive API schemas and compact responses that conserve a model’s limited context.
- Deployment friction can invalidate a successful prototype. His AI Engineer World’s Fair talk used Andrej Karpathy’s MenuGen application to identify failures that appear after local development: outdated documentation, changing APIs, cumbersome onboarding, rate limits, and abuse controls that block legitimate customers. He also supports exposing structured model pricing and capabilities so assistants can compare services on meaningful criteria.
At Cloudflare, Sikelianos applies these ideas to open-source coding agents. His Cloudflare development setup for OpenCode and sweet-sweet-opencode combine documentation search, browser tools, and platform integrations to help developers build applications across Cloudflare’s infrastructure.
