Every fast-growing company accumulates technical decisions that were right at the time, but wrong later. If unaddressed, the legacy system can throttle growth. Affirm spent years re-architecting a system that makes a new credit decision on every transaction, in real time, while the business kept growing around it. Geddes Munson, SVP Engineering at Affirm, walks through what it took: the technical sequencing, the organizational changes, and the company buy-in that made it possible. He'll also cover how AI accelerated the work, while rewriting the old textbook tradeoffs mid-project, and which of those rules still hold. You'll leave with a way to decide when a fundamental rewrite is worth doing, how to build support for it before it’s too late, how to sequence one without stalling the roadmap, and how AI fundamentally changes the rearchitecture equation.

Engineering organizations are hitting a Productivity Paradox. AI tools are generating more code than ever, yet the time it takes to ship that code safely to production in large-scale environments remains stagnant. The bottleneck has shifted from writing code to reviewing, validating, and trusting it. At Roblox, we've moved beyond coding assistants to build an AI-Native Engineering Platform, shifting human judgment from checking every change to defining the rules, gates, and production signals that govern classes of changes. In this session, we'll break down the architectural requirements for a prompt-to-production development cycle by walking one real change up the autonomy ladder. We'll detail how we replaced manual bottlenecks with agent-safe infrastructure, agentic guardrails that encode twenty years of institutional memory, deterministic approval gates, and per-line provenance that ties merged code back to the session and model that wrote it. We'll dive into the technical framework, Exemplar Alignment and hybrid symbolic-vector representations, that raised acceptance of AI review findings from 21% to 73% by teaching AI to reason like our senior domain experts. We'll also cover how to measure what actually matters in the new world of AI-driven engineering.

As engineering organizations grow and AI-assisted development accelerates, the gap between teams that have solid internal platforms and those that don't is widening fast. This talk explores two converging forces - the classical scaling problem where product teams buckle under infrastructure burden, and the new AI factor where each engineer's output is rising but can only be sustained on strong platform foundations. We'll cover how self-service, enablement, and deliberate platform investment become the leverage layer that lets product engineers stay focused on building products, not plumbing.

Most companies resolve the tension between AI usefulness and data risk by locking the data down, quietly killing the use case. Handshake couldn't: it supplies training data to frontier labs moving at an insane speed while running agents across its own stack. Matt walks through the internal products they're building (Mordor, Heimdall, Charon, Tartarus, and Citadel) that make both possible — column-level masking so models query broadly without broad exposure, a single surface where risky agent actions stage for human approval, a proxy metering every model call that turns the cost record into a training corpus, and two forms of sandboxing for safe agent code execution.

Most AI ROI conversations collapse into one question: did developers ship faster? That is an incomplete balance sheet. Enterprise AI also changes risk exposure, infrastructure and token spend, software quality, and how reliably value reaches customers. Engineering leaders from 1Password, Auth0 (Okta), and Anthropic will pressure-test the same deployment through three lenses: defense, dollars, and delivery. They will explore where these goals reinforce one another, where they conflict, and what leaders should measure before deciding to fund, scale, or stop an AI investment.

Last year we discussed how a single AI Week moved 200+ builders to be 40% faster. This is the part that matters more: what happened in the 51 weeks after. The spike faded, the skeptics got louder, the token bills arrived, and the agents we built failed in ways we could not explain. A year later, Amplitude ships three times the pull requests with the same headcount, cycle time is under an hour, and our second AI Week pulled in the entire company, from sales reps opening a terminal for the first time to the CEO shipping his own demo. None of it came from the tools. It came from leaders willing to build live in front of 600 people and fail, from treating AI spend as leverage instead of cost, and from accepting that adoption is a treadmill, not a project. This is the sequel: what stuck, what didn't, and what it takes from the person at the top.

An agent can be built in two days, but its value depends on how well an enterprise adopts it, whether that enterprise is a customer or the company's own engineering organization. Engineering leaders from Salesforce, Atlassian, and HPE will share what has worked from both perspectives: delivering agents to enterprise customers and driving adoption within their own teams. Panelists will discuss how to assess organizational readiness, why implementation determines whether an agent is used, and how to expand agent scope within compliance and risk requirements. Attendees will leave with practical steps to improve adoption and outcomes for the agents they build, whether those agents serve their own teams or their enterprise customers.

Every engineering leader putting AI into production is making a set of decisions that will outlast any single model release. How much to build internally vs. buy, how much control to trade for convenience, who owns the model, the data or the token cost curve as the AI market shifts. Each of these decisions shapes team structure, vendor dependencies, and how quickly an organization can adapt. Drawing on his work helping teams at companies like Uber and Shopify deploy AI at scale, Benny will walk through the tradeoffs behind these calls, including where a general-purpose model makes sense vs. when customizing open models wins, and how to build an architecture and a team that keeps evolving instead of locking into one choice. Attendees will leave with a framework to reason through these decisions for their own organization, weighing flexibility against operational load and how to lead a team through the daily decision-making of AI.