Stephen Batifol
9 Apr 2026Live at AI Engineer
Stephen Batifol (Black Forest Labs): World Models & Video Gen
Stephen Batifol outlines BFL's research-first approach: recent small models for near real-time image editing, a paper from a month ago on training from images, video, actions and audio, and world models for robot training as the long-term goal beyond video generation. He describes developer advocacy as inspiring developers and creatives, correcting misconceptions about how models are trained, and converting attention into API and open-weights usage. He also defends the deliberate Freiburg-rooted, research-first branding — build cool things and people come — and argues Europe has the talent to build frontier models while keeping an SF office.
Marlene Mhangami (Microsoft & GitHub): Copilot & Open Source
Marlene Mhangami describes her role spanning Microsoft and GitHub, the cultural contrast between the enterprise-focused parent and the more playful GitHub, and her developer-advocacy work on Python, open source and Azure APIs. She reports growing Copilot usage — including beginners using it to contribute to open source — and notes Copilot supports models from both Anthropic and OpenAI. The conversation turns to her take on the Mythos announcement right as this part ends.
Marlene Mhangami: Harnesses Matter More Than Models
Marlene Mhangami closes out her segment, saying she loves the Claude models and uses Opus through Copilot, which supports all models. She argues LLMs are generalists trained on the mean, so harnesses like Copilot matter more than raw model choice for specialized domains, and doubts current models are at AGI level. Asked what excites her, she highlights multi-model workflows and GitHub's new Critique release, where one model critiques another's output, before pointing listeners to GitHub Copilot and Copilot CLI.
Swyx: AI Engineer London, 10 Downing Street & EU/acc
Final guest Swyx joins live to discuss AI Engineer London. He explains the push to publish talk clips immediately after sessions and describes the morning visit to 10 Downing Street, praising the UK government's AI fellowship and noting the UK Chief AI Officer will give a closing keynote. He highlights the ElevenLabs booth with its Michael Caine voice, calls Luke Harries one of the world's top AI growth people, and cites Gradium's speech-to-speech work as evidence Europe can build its own AI champions. He also explains the EU/acc logo in the conference branding, saying builders must push capabilities alongside regulators.