A new stack is forming around enterprise agents. Not just models and prompts, but execution, isolation, governance, and deployment across customer-controlled environments. Waseem AlShikh shares how Writer approaches enterprise requirements like BYOC, security controls, and real implementation support, and why these constraints shape product architecture as much as they shape go-to-market. We connect the dots between sovereignty and runtime, and why sandboxes are showing up as a core primitive for agents that do actual work.

Databases were built for applications with predictable traffic and human-driven workflows. Agents change that assumption. They create databases on demand, fork them aggressively, spike concurrency without warning, and expect instant availability. In this fireside, Nikita Shamgunov shares what happens when AI-driven workloads collide with traditional database architecture. We explore why the old tradeoffs between PLG scale and enterprise depth no longer hold, how bursty agent traffic forces a rethink of storage and compute separation, and what it takes to build a database layer that can handle both millions of ephemeral workloads and mission-critical systems at the same time.

Modern cloud infrastructure was designed for stateless web applications. AI agents are not stateless web applications.

As AI capabilities commoditize, model ownership becomes a more durable advantage than model choice. In this keynote, Lin Qiao, CEO and co-founder of Fireworks AI will share hard-won lessons from building AI products: managing inference costs, finding product-market fit, and scaling reliably without ceding control to third-party platforms. She'll explain why owning your AI stack is the key to a defensible moat with autonomous agents. ᐧ

The world’s businesses first reorganized around the web, then around cloud. Now they’re reorganizing around agents. That shift isn’t just new apps. It’s a new infrastructure stack. In this fireside chat, Parag Agrawal breaks down the core primitives every serious agent system will need: LLMs for reasoning, execution environments, web access and search, and secure connections to private tools and data. We’ll explore what changes when search is built for agents instead of humans, how evaluation and reliability look different when the “user” is autonomous, and what new primitives will define the next AWS era.