CJ Avilla works on platform and developer experience at Anthropic. He produces developer screencasts, cohosts Build and Learn with Colin Loretz and has worked across teaching, product engineering and API developer tools.[1][2][3]
Start with Why One AI Agent Isn’t Enough (And How to Run 1,000) - CJ Avilla & Harrison Stall, Anthropic →
Career
Avilla’s personal career timeline lists App Academy instruction in 2012–2015, senior software engineering at MyVR in 2015–2019, developer advocacy at Stripe in 2019–2023 and full-stack engineering at Craftwork in 2023–2024. He then worked in developer relations at Stainless on SDKs and APIs. His current public X and Anthropic-focused GitHub account identify platform and developer-experience work at Anthropic. His personal About page still describes Stainless, so the transition has no inferred date.[4][1][2]
Contributions
Avilla’s Anthropic account describes work on developer documentation, deployable quickstarts, cookbooks and SDK examples. These serve different stages of adoption: a quickstart gives a working application, while a cookbook isolates one technique so a developer can understand and adapt it. His personal screencasts and Build and Learn conversations extend the same teaching approach to Rails, JavaScript and applied AI.[2][3]
Ideas
In his conference workshop with Harrison Stall, Avilla explains dividing work into phases, assigning bounded tasks to agents and evaluating the combined result. The presentation distinguishes isolated contexts and independent review from merely running many agents at once. It also discusses controls around long-running work. The workshop’s stated service limits describe that presentation’s environment and should not be treated as permanent platform guarantees.[5]
Tools, writing & talks
Writing, research & talks
Start here
Why One AI Agent Isn’t Enough (And How to Run 1,000) - CJ Avilla & Harrison Stall, Anthropicyoutube.comA chaptered conference workshop on agent flows, delegation, review and operational controls.
Start with the conference workshop’s chaptered agent-flow examples. API and coding-agent familiarity help; running the examples requires the services and credentials named in the workshop, and its stated agent limits are time-specific.
Cospeaker with Harrison Stall[5]Original source