XConf Lite | Vietnam

AI coding agents are becoming capable enough to write, refactor, and operate real systems—but capability alone does not make them reliable. This talk introduces harness engineering: the specifications, tools, tests, architectural rules, permissions, persistent memory, evals, and feedback loops that turn agents from clever assistants into trustworthy contributors.

We will examine the mental model of “agent = model + harness,” how guides and sensors protect maintainability, architecture, and behavior, and how durable traces, failure records, and self-improving workflows help teams move from vibe coding to verified, observable software delivery.

Agents that ace a benchmark can still unravel over a long run, quietly drifting from the goal, compounding small errors, or clinging to a broken plan.

This talk breaks down long-horizon agent design as a problem of depth, not duration: How many decisions an agent can chain before it loses the thread.

Drawing on three recent case studies: a frontier open-weight model, RLMs dividing and delegating longer context into sub-agents, and a payments company's fire-and-forget coding agents, we will surface the three failure modes behind most agent breakdowns, and the deterministic guardrails, forced goal-rereads, and judge patterns that fix them.

Production incidents are costly, chaotic, and deeply manual. Engineers sift through logs, metrics, configs, and code under pressure, often with no clear starting point. Agentic AI changes that.

In this talk, Vijay presents a live demo of First Responder, an agentic system that investigates production incidents end-to-end. Starting from an alert, automatic triaging and closes the loop with automated fix generation and pull request creation, with engineers staying in control throughout.

AI isn’t just making software delivery faster. It is fundamentally redefining what delivery means. As AI becomes deeply embedded across the development lifecycle, traditional metrics around capacity, team size, and effort are quickly becoming obsolete. The core question for tech leaders is shifting from how many FTEs we need to what outcomes we can create.

This talk breaks down the transition from scaling headcount to scaling capability and business impact. We will examine how the next generation of delivery combines human expertise with AI capability to create smaller, hyper-capable teams.