The Neuron: AI Explained

Danny Wu, Head of AI Products at Canva, leads the team building editable, layered design models. He explains how Canva plugs into ChatGPT, Claude, and Copilot to produce editable designs, why layered outputs matter more than flat images, and how 24 billion AI uses revealed unexpected creative workflows. He also shares advice for creatives and how AI integrations are changing discovery.

Ryan Carson, founder of Treehouse who taught over 1M people to code and now builds agent-first tooling at Untangle. He rethinks how people should learn to code when AI writes syntax, demos agent-driven workflows and his viral 3-file system, and argues the new hard skills are deployment, databases, security and shipping reliable products.

Brandon Sorbom, co-founder and Chief Science Officer at Commonwealth Fusion Systems, builds fusion power with high-temperature superconducting magnets. He explains tokamak design, why HTS magnets changed the game, and how AI speeds plasma control and digital-twin simulations. Manufacturing scale-up, real-time control with reinforcement learning, and timelines for Q > 1 are also explored.

Logan Kilpatrick, a DeepMind product and engineering lead who helps bring frontier AI to developers. He demos Gemini 3 Flash in action. They discuss how smaller, faster models unlock new multimodal and coding workflows. Conversation covers cost-driven choices, rapid prototyping in AI Studio, and how cheap, speedy intelligence changes developer tooling and product building.

Stefano Ermon, Stanford CS professor and founder of Inception Labs, explains Mercury — a diffusion approach that drafts full text then refines it. He discusses why diffusion can edit many tokens in parallel, how that reduces latency and GPU memory bottlenecks, which real-time applications benefit most, and the tradeoffs around quality, context length, and industry adoption.

Matt Price, founder and CEO of Crescendo and former Zendesk exec, builds AI-native customer experience that blends AI agents with human experts. Conversation highlights LLMs as a CX inflection point. Multimodal AI handles voice, vision, and APIs. Discusses forward-deployed human experts, outcome-based pricing, and the future roles and structure of customer service.

Taylor Mullen, Principal Engineer at Google and creator of Gemini CLI, who previously worked on GitHub Copilot and VS Code. He recounts building an agentic terminal from a hackathon prototype. Topics include how the CLI self-builds to ship 100–150 features weekly. Discussions cover managing swarms of parallel AI agents, verification and testing workflows, Conductor planning, and the Ralph Wiggum parallelism technique.

Alexander Embiricos, product lead for OpenAI Codex who builds agentic coding tools, demos GPT-5.1 Codex Max live. He shows where Codex runs, how it executes safe terminal actions, and how teams customize agents and workflows. Demos include installing IDE extensions, scaffolding projects, making live code edits, planning long tasks, and integrations with GitHub, Slack, and issue trackers.

Eve Bodnia, founder and CEO of Logical Intelligence and former theoretical physicist, builds Kona: an energy-based, token-free approach to AI. She explores energy landscapes for non-linguistic reasoning. Short takes cover spatial planning, robotics control without language, reducing hallucinations via constraints, and how EBMs might integrate with large language models.

Predictions about which model types and companies will rise or fall in 2026. Debate over diffusion models challenging transformers and alternative architectures. Forecasts for small on-device models, world models for immersive experiences, and natural-language-built agents. Discussion of tooling, interfaces, pricing pressures, chip geopolitics, and which incumbents could regain leadership.