2 years ago, the engineering team's structure was fairly standardized, 2-pizza cross-functional teams and an engineering manager leading the team. These days, everyone is figuring it out as they go, making adjustments and seeing what may work and what not.In this talk, Gregor will share how companies like OpenAI, Anthropic, Shutterstock and others build engineering teams in the AI era. He will also share his recommendation on the best structure for a specific type of organization.

We used to explain our bugs to rubber ducks. Now, the duck talks back, writes code, and occasionally lies to our faces.LLMs are incredible tools. They can cure "blank screen syndrome," instantly write boilerplate, and explain weird legacy code. But treating them like senior engineers is a massive trap.

Teams often start agent adoption by adding tools: MCP servers, API wrappers, chat interfaces in developer portals. But once coding agents move from local help into real delivery workflows, they usually fail earlier than the tool call. They read stale ownership, conflicting runbooks, missing deploy history, and policies that were written for humans.This talk introduces company memory as the missing layer for production coding agents: trustworthy, permission-aware engineering context that agents can read, cite, and act on safely. We’ll look at what belongs in that memory, how to expose it through capability registries and machine-readable tool contracts, how to use on-behalf-of identity instead of shared AI service accounts, and how to grow trust from read → recommend → act.The goal is simple: before giving agents more tools, make sure they can trust what they read.

AI makes individual engineers faster - and quietly more isolated.As coding agents become everyone's default pair, the collaboration that keeps an org healthy erodes: pairing fades, context gets trapped in private chat histories, knowledge stops diffusing, teams drift into silos.Velocity dashboards look great, so nothing seems wrong - until onboarding slows and bus-factor spikes.This talk names the failure mode, explains why standard metrics miss it, and shows what leaders can do to keep teams connected as AI adoption accelerates.

I built an RSC renderer in Rust by reverse engineering React's Flight protocol. The wire format, the serializer, the streaming machinery. I got it wrong in ways that taught me exactly how React internals work. Then I integrated the canonical Fizz renderer and discovered why React's approach exists.

In this data-driven talk, we will bypass marketing hype and unmask the hidden creatures behind Next.js., Using massive, real-world datasets from the HTTP Archive and the Chrome User Experience Report (CrUX), we will review the chronological evolution of Next.js performance and contrast its promised performance features with the real-world metrics of massive sites struggling with performance debt. Don’t dream your web is fast. Face the reality before it's too late.

Almost a decade ago, React’s rendering algorithm was rebuilt, and with that it brought about new tools, ideas, and ways of building UI. But those tools arrived incrementally and were adopted in isolation, often framed as “optimizations” rather than part of a unified model. So are they truly independent ideas, or are they different expressions of the same underlying model made possible by the new rendering algorithm?Join us as we re-introduce "Async React" as the mental model for modern react and show how we can write more declarative code, embrace async-first as the default and let React handle the rest.

Next.js 16 introduced cache components, a fundamentally new approach to caching that replaces implicit behavior with explicit, composable control. In this talk, I'll build a dashboard from scratch to show you how cache components work in practice: how the static shell gets prerendered and served instantly from the edge, how 'use cache' caches expensive data on your schedule, and how Suspense streams dynamic content in real time, all in the same route. You'll see the three 'use cache' variants in action, learn the donut pattern for maximum performance, understand why generateStaticParams matters more than you think, and walk away with practical patterns for cache invalidation that work in production.

Frontend teams are increasingly using AI to speed up UI development by generating code directly from designs. While the results can look impressive, they often fall short in real production environments: incorrect components, missing accessibility, and broken design-system rules are common failure modes. In this talk, I’ll present a real-world Figma-to-Code case study that treats AI as part of the frontend infrastructure rather than a code generator. By teaching AI how a design system works and orchestrating a structured workflow, we moved from “cool demos” to production-ready UI code - an approach we now rely on in real product development, saving significant time while improving consistency, quality, and user experience.

At TAIT, we build the full automation stack behind the world's biggest live experiences, from the systems that move stages and fly performers to the React interface that operators use to control them. Our frontends can't lag, freeze, or crash while hardware is moving, or someone can get hurt. This talk covers the architecture and performance patterns behind that interface, and what it teaches us about building any React app where failure isn’t an option.

Next.js is a great framework - until you want to deploy it somewhere that isn't Vercel. vinext is a Vite plugin from Cloudflare that reimplements the Next.js API surface from scratch, so your existing app runs everywhere with a single command. In this talk, we'll look at why this exists, how Cloudflare built it in a week using AI, and how close to "drop-in replacement" it actually gets.

Frontend and backend are often treated as opposite worlds, each with its own tools, constraints, and claims about which side is harder. But are they really so different?Modern frontend applications increasingly resemble distributed systems. Multiple tabs drift out of sync. Shared caches require invalidation.

React has changed a lot in the last year: React 19, 19.1 and now 19.2 brought a stable React Compiler, new hooks like useEffectEvent, the <Activity /> API, and better SSR primitives such as Partial Pre rendering. And moreIn this talk we’ll take a demo React app that’s full of effects, memoization and “old school” patterns, and modernize it step by step.

We have been supporting builders of AI products with over a dozen of features in TanStack AI. Today I want to show you how most AI workflows work under the hood, what it takes to run a harness, what is important after the initial implementation of a chat experience and how to grow your functionalities and scale to the needs of your users, and most importantly, how TanStack AI enables you to do all of this with ease and joy.

After six interviews with people with disabilities about their daily struggles and opportunities they see in AI, it became clear that while standards and patterns have brought accessibility far, there are hard limits on how well one interface can adapt for all users.This led us to a different approach: instantly generated UI based on personal needs and context. Not adapting one interface, but exploring what AI makes possible when launching an app takes the same time as creating one on-the-fly.