How goose led to the AAIF

Why this conversation

Every platform team adopting agents is choosing a stack that is barely a year old: a protocol for tools, a format for instructions, a harness that writes code. Who owns those pieces decides whether you can rely on them. The Agentic AI Foundation was created to answer that question for the agentic layer, and Manik Surtani is its CTO and co-founder. I sat down with him at AGNTCon + MCPCon Europe in Amsterdam, after his keynote, and asked for the short version: why does the foundation exist?

This episode is a two-minute cut of that conversation. The rest of it, and the keynote, are written up on lucaberton.com (linked below).

The origin story starts at Block

Manik answered with the origin story. A couple of years ago he was head of open source at Block, at the very start of what he calls the large language model era. His team began experimenting with LLMs for coding and built a harness to help them write code with them: goose. It predates Claude Code, he points out, and it found a lot of attention inside the company.

So Block open sourced it, to see what the community would do with it, and it took off: people started using it, forking it, changing it and improving it. That is the moment the foundation’s story starts. In the full conversation he explains what came next. Some organisations would not contribute their changes back while one company still owned the trademarks and the roadmap; the existing foundations did not quite fit; and Anthropic was looking for a neutral home for MCP for the same reasons. So they created one.

Sovereignty

The cut ends on sovereignty, which Manik calls a big deal for him personally: “I’m a believer in people need to own their own AI destinies, and this is where open source has a huge role to play.” For a European team choosing where its agents run and whose protocols they speak, that is the practical argument for building on projects with a neutral owner.

The rest of the conversation

The cut leaves out most of what we covered, and three points are worth carrying with you. Member companies want working groups on robotics and embodied AI, and those come once the foundation has a lifecycle that can also close a group. On careers, he separates programming from software engineering: the programming can now be delegated, as assembly once was, but someone still has to understand what is being built and how the pieces fit. And the way in is open: use the projects, come to the events, start a local chapter, contribute a project or propose a working group.

What he said on stage

His keynote that morning, “Catch Them Early”, was about the foundation’s entry bar. The AAIF had inherited a bar built for a mature industry, and it was turning away projects that were technically real but too young to meet it. The fix is a Sandbox phase, a second door before Growth and Impact: a working implementation and a named, active maintainer get a project infrastructure and a neutral flag, nothing more. For a platform team, that is the signal to read before building on an agentic project: which phase it is in, and whether two unaffiliated organisations already run it in production.

Big thanks to Manik for joining me in Amsterdam.