Talk
Getting agents running is half the battle. Making them performant and continuously improving across thousands of sessions is much more challenging. This talk covers how to do it with Claude Managed Agents running on Vercel's agentic infrastructure, including best practices for optimizing agentic performance in production.
Demo
Companies use Ora to find out whether their website works for AI agents, and the way Ora answers is by sending real agents onto the site to sign up, integrate, and pay, then tracing every step. Hear how the Ora team built an agent fleet on eve, what hundreds of real agent journeys show about where websites break, and see Is-Agentic score a live site.
Talk
How do you serve 168 models through one integration and stay reliable at 3M+ requests a day? Adapta, the team behind Brazil's largest generative AI platform, did exactly that on AI Gateway and the Vercel Platform. Hear how they built an app-level router that picks the right model per request in under 200ms, wired up automatic failover and per-user budgets, and kept reliability climbing as traffic spiked. A practical look at the architecture decisions behind running frontier models at scale.
Talk
Mercor connects domain experts with frontier AI labs, generating the human data used to train and evaluate leading models. Behind the platform, hundreds of business users build the apps their work depends on, and engineering teams tap a wide range of open-weight and closed models. The challenge: doing both without the business losing track of what's shipping, who has access, and what it costs. See how Mercor, with Vercel, built a managed identity layer every app inherits by default and gained full observability into who is spending and where, on a foundation that scales to the agents it's building next.
Talk
Most agents handle one prompt and a quick burst of work. Hyperagent runs agents that wake on a schedule, work for hours in real browsers and code environments, and escalate to humans. Built on Vercel Sandbox with AI Gateway, we tested 14 models on 42 real jobs and found a 50x cost gap on identical tasks.
Talk
Building a custom agents platform is one of the key pieces to unlocking AI for complex business workflows. Monte Carlo is the sales-specific system we've built to help scale Monaco's revenue: a platform where revenue teams encode their own playbooks as agents that run against everything we've ingested about their business, including pipeline, outbound, calls, and CRM. Sales is a particularly complex vertical for the following reasons: no two revenue orgs look alike, the data is large and heterogeneous, and the expertise that matters lives in per-rep workflows nobody has written down. Making it work at scale meant rebuilding the agent runtime on per-run Vercel Sandbox microVMs, a decision that propagated much further than we expected into how we handle context, how much of a playbook ends up as prose versus code, and what an agent is permitted to do at all. In this talk, I'll break down the components a verticalized agent platform needs in order to be fast and correct, including an execution substrate, memory and context, how expertise gets represented, and the evaluation loop, along with what is still unresolved.
Talk
AI coding tools make it easy to track tokens, licenses, and model spend. The harder question is what that spend actually produces. In this session, Emilie Schario, VP of Engineering at Anaconda and Co-Founder of Kilo Code, introduces AI Cost to Ship, a metric that connects AI spend with merged pull requests, and explores how leaders can use it alongside adoption, quality, and model choice to understand whether AI investment is translating into shipped engineering output.
