Nine agents behind one newsletter

Why this conversation

Agentic AI moves fast enough that keeping track of it is a job in itself, and, as the host pointed out, agents are not only for writing software. At AGNTCon + MCPCon Europe in Amsterdam the host met Alex Salkever, who came from San Francisco and is Editor-in-Chief of the Agentic AI Foundation. In practice, he says, that means he does all the newsletters, and he uses agents of his own to do it.

The story of the week it was filmed is in AGNTCon + MCPCon Europe 2026: who we met.

Nine agents and one editor

Most Sundays Alex reads everything that happened in agentic AI over the past week and decides which 90 pieces of information go into the newsletters, which run to about 3,000 words. He does not read every item all the way through. About nine agents do the curation for him, bringing the week down to three or four hundred news items. He looks through those, skims most, and reads the whole thing when an item is controversial.

That last step matters because the AI makes mistakes. “It will tell you it’s one thing, but it was lazy”: it did not read the whole site or all the information. His example from the day before was a GitHub project that looked interesting: something can trend in stars not because it is new, but because somebody said something about it. In the news business, he says, you have to check these things very carefully, because the AI’s taste and its understanding of context and time are still not always accurate.

Scripts, feeds and changelogs

Behind the newsletter is software he built himself: scripts that pull in different RSS feeds and analyse what comes in, in several stages. Some stages translate, into Asian languages for example, when he wants to read back and forth. He also uses AI to pull out or summarise what he needs from very large or hard-to-read documents. When a project has a long changelog, the AI tells him what is interesting and where to focus.

Part of each newsletter is a sweep of all the protocols affecting agentic AI. If something changes in MCP apps, for instance, the foundation wants its members to know. The AI helps him parse the comments and the pull requests and cover what he estimates at about 50 things, across the IETF, the protocols the foundation follows, and projects such as AGENTS.md. That is what makes it doable to look at all of it and condense the most interesting pieces.

When the host called him a 10x engineer, he declined the title: he writes code, but he is not a coder.

Subscribe

The newsletter is published on LinkedIn, where anyone can subscribe. He also writes a second, internal newsletter, which is much longer.