Adrian Cole

How to break up with your Agent: Goose architectural shifts towards interop

The biggest change developers had in the last year or so, wasn’t the rise of agents.. Rather the rise of agentic subscriptions. We switched from focusing on choosing LLM platforms and counting tokens, to choosing an agent CLI and counting hours of use for a fixed monthly price. We began thinking each agent is more or less the same coding agent.. until they weren’t. Today, we have comparable big cloud agents like Claude, Codex and Cursor, but we also have specialized ones like OpenClaw for social media and StakPak for SRE automation. There’s also mixed used customizable agents like Goose. There are dozens of agents out there, if not more. What limits us is that some agents imply a specific frontend, like a VScode integration or a nice terminal UI. Others have mobile apps, shell integrations and other neat features. So what if you like the frontend, but the agent itself is too expensive? What if you love the agent, but it doesn’t integrate with your favorite JetBrains IDE? Can you break up with your agent without breaking your development process? This session walks you through how to switch agents as often as you like, through the lens of the Goose project. You’ll learn how Goose developed features for parity with coding agents, but eventually opened up to be powered by other agents via a community standard called ACP. You’ll learn how Goose uses this to let users keep the UI but swap the engine, or keep the engine, but swap the UI. You’ll also learn where ACP stops and what is still agent-specific as of today. Goose has always been open source and was recently donated to the Agentic AI Foundation (AAIF) inside the Linux Foundation. Every slide links directly to the PR or GitHub discussion so you can go look at the exact code yourself afterward. Come for the unfiltered story of how to break up with your agent or use as many as you like at the same time!