AI Engineer 2026 — Keynote Notes

Illustrated, timestamped summaries of the AI Engineer 2026 mainstage keynotes — the World's Fair in San Francisco and the Singapore edition.

Each day's mainstage livestream runs 8–10 hours. These pages distill them into per-talk summaries — who spoke, what they argued, the punchy quotes, key slides, and clickable timestamps that jump straight into the source video. Speaker names and talk titles are taken from each conference's own running order and on-screen title cards, and every stream is split by talk (not arbitrary time-chunks), so every entry maps to a real session.

🌉 World's Fair — San Francisco

Moscone West · June 30 – July 2, 2026

Day 1 — Software Factories 33 talks

June 30 · ~8h37m · OpenAI, Google DeepMind, Microsoft, Cursor, Warp, Factory, OpenClaw, HuggingFace, Z.ai, MiniMax, HumanLayer…

Software engineering as “factory engineering”: agents that triage, spec, implement, verify and ship in a loop — and the counter-melody of “slop” when they run unreviewed.

Day 3 — Autoresearch 16 talks

July 1 · ~8h52m · Anthropic (Fable), Google DeepMind, Amazon AGI, You.com/Recursive, Prime Intellect, Weco, GEPA, Cursor, Arena…

Agents that do research and recursively self-improve. The verifiable loop, reward-hacking, and why humans move to taste, judgment, and accountability.

Day 4 — Harness Engineering + Startup Battlefield 15 talks

July 2 · ~9h11m · AWS, Anthropic, DSPy, PostHog, Restate, Vercel, Oracle, Morgan Stanley, Google DeepMind, Garry Tan, Theo Browne…

Everything wrapped around the model that makes an agent safe in production: sandboxes, credential proxies, “prompts are not security,” adversarial red-teaming — plus the $100K Startup Battlefield.

🇸🇬 Singapore

Capitol Theatre · May 16 – 17, 2026

Day 1 32 talks

May 16 · ~10h09m · Singapore Foreign Minister, NanoClaw, OpenAI Codex, GovTech, Vercel, Daytona, Cursor, Groq, Cerebras, Z.ai, ElevenLabs, Figma…

Agents in production with an APAC, sovereign flavour — opening with a government minister who built his own containerised AI, and closing on the “Agent Lab Nation.” Sandboxing recurs as the precondition for autonomy.

Day 2 32 talks

May 17 · ~9h28m · Google DeepMind, Cloudflare, IBM, OpenClaw, Adaption Labs, Arize, Mastra, Tusk, MiniMax, The Robot Company, Smithery…

Harness engineering: what you wrap around the model so it's reliable and safe — deterministic boundaries, OS-level guardrails, V8 isolates — plus a big embodied-AI/robotics block and a design thread.

Source & attribution. All talks are from the official AI Engineer 2026 livestreams on the AI Engineer YouTube channel. Slide stills and quotes are reproduced here for commentary and educational summary; every talk links back to its exact moment in the original video. All rights to the talks belong to the speakers and AI Engineer. These notes are an independent, unofficial summary and are not affiliated with or endorsed by AI Engineer.