Agents Are Distributed Systems — What Navan and TikTok Learned
A timeout on a refund tool doesn't mean failure. It means unknown. Two AI Engineer talks on running agents in production, from the teams doing it.
Conference
9 stories · prices and picks kept current
A timeout on a refund tool doesn't mean failure. It means unknown. Two AI Engineer talks on running agents in production, from the teams doing it.
Best Buy's Ahnaf Prio maps the agentic commerce stack — MCP, A2A, ACP, UCP, AP2 — and explains why browser-driving shopping bots failed at checkout.
Two AI Engineer conference talks in depth: how one agent cleared Turnstile and reCAPTCHA v2 alone, and why Aviator's co-founder says the diff is dead.
Instagram's co-founder on porting 200K lines over a weekend, why code review is really a comprehension bottleneck, and how Anthropic Labs kills projects every two weeks.
Clare Liguori of AWS on Amazon's 50-team agent pilot: a median 4.5x gain in deployment velocity, the five habits behind it, and why half the teams missed.
Varun Shenoy on why AI still hasn't reached a 200-person property firm — and what owning 35 services businesses taught Long Lake about deploying agents.
Jesse Lumarie on shipping Figma's MCP server as a 20% project: why React and Tailwind beat the scene graph, and the eval that started in Excel.
Metronome co-founder Andrew Garvin demoed an agent building a Lovable-style credit pricing model from one sentence — and drew a hard line at the production boundary.
Uber built a multi-agent code review engine because review time tripled. Figma's lesson: verification beats prompting. Two AI Engineer talks, unpacked.