Engineering AI 25 Session 2

Michael Neale, creator of codename goose, an on-machine, open source AI agent built to automate your tasks explores the emerging landscape of ambient AI—agents that don't just respond to commands, but actively observe, learn, and work alongside you throughout your day. Drawing from his experience at the intersection of developer tooling and AI, Michael examines three critical shifts happening right now: <ul> <li>Ambient Agent Adoption: The radical trust required when AI agents have access to everything—your code, communications, calendar, and creative process. It's not just about project-specific AI anymore; it's about agents that understand your entire digital life and can act on your behalf.</li> <li>The New Economics of Intelligence: When AI token costs can represent up to a third of a creative professional's salary, we're not just buying software, we're accessing a new kin economy, much like when we got internet access in the 90s. Michael breaks down why this economics makes sense, even at twice the price, and what it means for the future of creative work.</li> <li>Universal Agent Architecture: Why the most powerful general-purpose agents aren't dumbed-down consumer apps, but sophisticated tools originally built for developers. Computers are for computation, and the future belongs to agents that embrace this complexity rather than hide from it.</li> </ul> This isn't a distant future—it's happening now. Come learn how to live productively with AI that knows you better than you know yourself.

If you've ever wished there were three of you (or maybe just a quieter version who gets stuff done), this talk is for you. As Head of Engineering at Relevance AI, I’ve been quietly offloading parts of my job to agents - and somehow still kept my job. In this session, I’ll share how I use AI to automate the boring parts, delegate without a team, and build agents that make decisions so I don’t have to. You'll get real examples, battle scars, and a practical playbook for working smarter - or at least pretending you are.

Some of the most important work in the software development lifecycle can also be some of the most boring! Keeping a codebase healthy requires a bunch of small repetitive tasks like keeping up with the breaking changes in dependency updates, or switching from an old architectural pattern to a new one in the hundreds of places that followed the old way. Large Language Models work best with clear instructions, examples to follow, and just the right amount of context. With a bit of context engineering they can be remarkably effective in these kinds of repetitive and tedious changes - allowing you to make sweeping tech health improvements across either your gigantic monoliths or sprawling micro-services (or both if you’re really lucky). AI might be coming for our jobs, but it’s still not quite there yet, so let’s have fun giving it the boring jobs.

Everyone's shipping AI features. Nobody's talking about how to test them properly. When AI becomes part of your application stack, traditional testing approaches break down fast. Non-deterministic outputs wreak havoc on CI pipelines, token costs spiral out of control, and teams struggle to build confidence in systems that make decisions they didn't explicitly program. Let's dive into test practices used in Envoy AI Gateway and Codename Goose. You'll see recorded response, local model serving, and LLM evaluation/benchmark techniques in non-theoretical applications. This isn't about testing AI models themselves, but about maintaining software quality when intelligence becomes infrastructure. Whether you're integrating with external LLM APIs or running models locally, you'll walk away with concrete strategies for testing AI-powered applications that teams can implement immediately.