The wiki

Researcher and designer of malleable software, exploring how AI can help people build, understand and reshape their own tools.

Artificial Analysis co-founder and CPO working on independent AI benchmarking, agent evaluation, and the trade-offs behind model selection.

Creator of tinygrad and founder of tiny corp, pursuing simpler AI software and more accessible compute.

Microsoft’s software development platform expanded from sharing Git repositories to code suggestions and delegated agent work, with branches, tests, and pull requests keeping proposed changes inspectable.

From reconstructing a workweek to redesigning Copilot, GitHub’s agent ambitions depend on shared context, credible review, and infrastructure built for changing workloads.

Cowork combines a coding agent, a controlled execution environment, browser access, and reusable instructions. Its central challenge is making work easy to delegate—and safe enough to leave running.

Glean turns scattered workplace information into searchable context, conversational answers, and agent workflows, with permissions and personalized retrieval at the center of its architecture.

An AI interpretability research lab developing tools to inspect model internals, monitor behavior, and guide training.

Production monitoring, live steering, and scientific discovery expose both the promise of mechanistic interpretability and the distance between finding a feature and controlling a capability.

Google connects AI research, custom computing infrastructure and widely used products, from Search to Gemini and Google Cloud.

Google’s AI research team helped turn neural-network research into translation systems, open-source infrastructure, and experiments in robotics and creativity before joining DeepMind in 2023.

Google’s AI research and model-building organization, combining multimodal models, scientific tools and research into reasoning, robotics and security.

Google’s AI product organization builds experimental tools for research, coding, filmmaking and interactive workflows, working closely with DeepMind.

Long-context reasoning, controlled coding agents, practical prompting, and workload-dependent costs in the April 2025 conversation.

An enterprise AI platform for custom models, with finance-specific models, million-token Llama 3 research and ambitions for autonomous workflows.

Natural language processing researcher and All Hands AI cofounder whose work spans neural-network tools, realistic agent benchmarks, and open software-development agents.

Founder of Wand, building AI drawing tools around visual control, artistic skill and personalized models.

Gamma’s co-founder and CEO on content-first creation, AI onboarding, and building a lean team.

Designer and engineer working on generative AI interfaces, from hand-controlled image exploration to agentic canvases, video tools and live music instruments.

An AI security company combining adversarial testing, runtime policy enforcement and community red teaming, with roots in Carnegie Mellon research.

OpenAI cofounder and engineering leader whose work spans infrastructure, reinforcement learning and the practical design of coding agents.

Google Brain co-founder and word2vec co-author whose healthcare AI work emphasizes clinical validation alongside model capability.

ARC Prize Foundation president whose work spans AI evaluation, long-context retrieval testing, and practical AI applications.

Engineer and Subroutine cofounder working on AI agents, trusted skill distribution, and development tools that make agent behavior easier to test and constrain.

An AI inference company whose LPU technology and GroqCloud evolved into a global cloud business alongside a non-exclusive technology license to Nvidia.

NVIDIA research scientist working on agentic robotics, reusable agent skills and learning game controls from internet video.

An AI reliability company whose products span output validation and simulated agent conversations, acquired by Harvey in September 2026.

Output specifications, validation and repair give developers more control over model behavior. SQL makes the mechanism concrete; an email assistant exposes the harder problem of specifying what the user wants.

Mistral’s co-founder and Chief Science Officer has contributed to efficient language models, unsupervised translation and neural symbolic mathematics.

Machine-learning engineer and independent AI consultant focused on evaluating, debugging, and improving AI products.