AI models, applications, and developer tools
OpenAI
OpenAI develops AI models and applications for consumers, developers, and organizations. ChatGPT supports conversational assistance, while ChatGPT Work carries out tasks across connected apps and files, producing spreadsheets, presentations, documents, and web apps. Users can review progress, steer work, and approve important actions. Codex helps developers build software and review pull requests; OpenAI’s APIs let developers build on its models. Together, these products span direct assistance, software development, and workflows involving multiple tools.
Founded as a nonprofit in 2015, OpenAI began with Sam Altman and Elon Musk as co-chairs, Greg Brockman as CTO, and Ilya Sutskever as research director. Altman is its current CEO. Its InstructGPT research applied reinforcement learning from human feedback to improve instruction following: human demonstrations and ranked model responses supplied training data, and a learned reward model guided further training. This approach addressed the gap between predicting text and responding to user intentions.
Following its completed 2025 recapitalization, the nonprofit OpenAI Foundation continues to control OpenAI Group PBC, its commercial public benefit corporation. In March 2026, OpenAI reported more than 900 million weekly active ChatGPT users and over 50 million subscribers. That month, it closed a funding round with $122 billion in committed capital at an $852 billion post-money valuation. By July 2026, the company reported more than 5 million weekly Codex users, including more than 1 million using it outside software development.
22 talks
Newest firstFrom fork() to Fleet: Designing an Agent Sandbox Cloud — Abhishek Bhardwaj, OpenAI
AI Engineer World's Fair 2026 · 44:34
Full Workshop: Setting Yourself Up for Success — Jason Liu, OpenAI Codex
AI Engineer World's Fair 2026 · 1:15:02
The Golden Age of AI Engineering
Alexander Embiricos · Romain Huet · Peter Steinberger
AI Engineer World's Fair 2026 · 25:13
Harness Engineering: How to Build Software When Humans Steer, Agents Execute
AI Engineer Europe 2026 · 46:21
Model-Maxxing: RFT, DPO, SFT (Fine-tuning with OpenAI) — Ilan Bigio, OpenAI
AI Engineer World's Fair 2025 · 1:46:15
Your realtime AI is ngmi — Sean DuBois (OpenAI), Kwindla Hultman Kramer (Daily)
Sean DuBois · Kwindla Hultman Kramer · Yaxin
AI Engineer World's Fair 2025 · 16:30
From Text to Vision to Voice: Exploring Multimodality with OpenAI
AI Engineer World's Fair 2024 · 23:39
25 speakers at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here
- Harness Engineering: How to Build Software When Humans Steer, Agents Execute
Start with the essay based on Ryan Lopopolo's talk to learn how repository boundaries, structural checks, and review processes make team standards usable by coding agents.
Ryan Lopopolo · Vibhu SapraAI Engineer Europe 2026
- Building Effective Voice Agents
Use this talk to understand how model delegation, constrained tools, evaluations, and observability fit into a customer-service voice system.
Anoop Kotha · Toki SherbakovAI Engineer World's Fair 2025
- Agent Reinforcement Fine Tuning
Read this session for the mechanics of training tool-using agents, including public tool endpoints, rollout tracking, and isolated execution environments.
Will Hang · Cathy ZhouAI Engineer Code 2025
- The New Code
OpenAI alignment researcher Sean Grove argues that precise, durable specifications—not prompts or generated code—are becoming the fundamental artifact of software development.
Sean GroveAI Engineer World's Fair 2025
Messages from the stage
Make agent behavior explicit and enforceable
Sean Grove argues for durable specifications that preserve intent; Ryan Lopopolo describes delivering requirements through context and feedback loops. Fouad Matin adds execution boundaries through isolation, restricted network access, and human approval.
Voice architecture involves product tradeoffs
Anoop Kotha and Toki Sherbakov compare chained speech pipelines with speech-to-speech systems across accuracy, determinism, and latency. Dominik Kundel’s workshop makes those interactions concrete through interruptions, human approval, handoffs, and transcript-based guardrails.
Train for useful behavior within a budget
Will Hang and Cathy Zhou connect agent training to production-representative evaluations, inference-budget penalties, and continuous rewards. Ilan Bigio’s fine-tuning workshop supplies broader context on supervised training, preference optimization, distillation, and when fine-tuning is warranted.
Affiliations reflect each recorded session, not necessarily current employment.





















