Romain Huet

Romain Huet is Head of Developer Experience at OpenAI. His work includes demonstrating how developers can use reasoning models to architect applications and connect real-time voice to external tools. He announced his move to OpenAI on September 21, 2023; his personal GitHub profile uses the handle romainhuet.[4][5][1]

At Stripe Sessions 2024, Huet distinguished two strategies for builders: filling gaps in models’ present capabilities, or planning for models to improve. He endorsed the latter, predicting that developers would spend more time orchestrating agents through natural language and applying product taste and design expertise, while models handled more routine work such as writing tests.[6]

His technical explanation paired context with task decomposition. Retrieval or fine-tuning can supply knowledge of a codebase; an action plan can then break a complex problem into subtasks assigned to models with different quality, cost, and latency trade-offs. He used Cognition’s Devin as an example. Huet also described spending time with developers to understand those trade-offs and help them bring applications into production.[6]

In his DevDay 2024 interview, Huet said developer requests had influenced releases including prompt caching, prompt generation in the playground, and vision fine-tuning. He invited feedback ranging from feature requests to intricate API endpoint details, describing the roadmap as “heavily influenced by them and their work.”[1]

At DevDay 2024, Huet demonstrated an International Space Station tracker for iPhone and a drone application. For the iOS demonstration, he used o1-preview in ChatGPT alongside Xcode to brainstorm, consolidate files, and architect the application, then manually transferred generated code into Xcode. The app required fetching API data, consulting documentation, parsing JSON, deciding where to store data, and wiring the components together. He described o1 as “your architect, effectively,” emphasizing planning as well as code generation.[1]

Huet reported that the iPhone app built on the first attempt after transferring the code. Interviewer swyx questioned how much setup and prior iOS knowledge the demonstration required. For his second demonstration, Huet chose o1-mini to connect frontend and backend code, UDP packets, and WebSockets. His model-selection advice at the time was to try both: mini for math and coding tasks, and preview when broader knowledge was needed.[1]

Huet chose travel as the use case for the Wanderlust demonstration because it combined pictures, locations, and potential translation needs. The app appeared at DevDay 2023 with a text-based assistant; in 2024, he brought it back to demonstrate a real-time voice conversation that could also call tools and functions. One interaction placed a phone call through the Twilio API.[1]

Wanderlust was a team effort. Its original app creators were Simon and Carolis, according to OpenAI developer-experience engineer Ilan. Ilan added the voice and phone-calling components, with the API team helping debug the integration.[1]

Huet qualified his launch guidance by describing the Realtime API as a beta in 2024. For robust client applications, he suggested WebRTC-based approaches rather than working directly with sockets at scale, pointing to LiveKit and Agora integrations. He distinguished client-side and server-side voice requirements and said developer feedback would be critical to further iteration.[1]

At the AI Engineer World’s Fair 2026, Huet presented with Alexander Embiricos on Codex. In Richard MacManus’s conference account, the pair argued that using multiple agents through loops could improve productivity. MacManus’s subsequent analysis attributed to Huet the view that tools such as Codex make it easier for engineers to collaborate with agents.[2][3]

Before OpenAI, Huet worked at Stripe and Twitter. His earlier career also included cloud software: Tariq Krim recalls cofounding Jolicloud with Huet after they had worked together at Netvibes. These roles preceded his work helping developers build with OpenAI’s models and APIs.[4][8][6]

On March 20, 2018, Huet authored Stripe’s announcement of new developer tools in the Dashboard. He explained how developers could filter errors by endpoint or type and inspect the request logs behind them, investigate webhook events and response times, and see which API versions an integration used alongside potential breaking changes before upgrading. The announcement illustrates the concrete integration and debugging concerns in his earlier developer-platform work.[7]