Who picks
I’m an AI engineer who builds agents for customers and advises on AI. The newsletter follows what I want to keep up with myself: building agents (the core), the AI ecosystem, how markets and enterprises are reacting, building an AI business, advising employers and customers, choosing between models, and the inference stack. Every pick serves at least one of these, and is tagged with it.
Where posts come from
A fixed list of sources: research and engineering blogs from AI labs and platform companies, independent writers and newsletters, and a few YouTube channels of conference talks. Issues go out on Monday, Wednesday and Friday evenings. Each run checks every source for posts published since the last issue.
How a post is picked
- Interest, not recency. A post is picked because it changes how one would design, ship or run agents, or because it says something useful about the market, a business model or a model choice. No source is guaranteed a slot.
- Read in full first. A post is only picked after it has been read in full; for a video, the whole transcript. Paywalled posts and pages that block automated reading are left out.
- At most eight per issue. When there are more strong posts than that, the rest carry over and compete with the next issue’s posts.
- Usually skipped: funding and partnership announcements with no wider lesson, minor product launches, events, and content without a concrete, reusable insight.
- Never in two issues. A post appears once.
Everything a run considered, picked or not, is listed in the Library with a one-line description. Posts marked “not reviewed in full” were judged from a skim or couldn’t be read.
How the summaries are written
I use an AI assistant to scan the sources, read the candidates and draft each entry, following written rules on what to pick and how to write. I review every issue and approve it before it’s published.
- Summaries are neutral and in my own words, without copied sentences. Every number is checked against the post.
- “Why I picked this” is my opinion: why the post matters to someone building agents for customers. When a post is mostly a product announcement, the note says so.
- Key numbers are figures reported in the post itself, not a rating of it.
- Corrections are made in place, with a visible note saying what changed.
What this is not
- Not advice. Nothing here is investment, legal or professional advice.
- Not sponsored. No one pays to be included, and there are no affiliate links.
- Not anyone else’s view. The opinions are mine, not those of any employer or client.
- Not the original. Every post belongs to its author and links to the original; the summaries are a pointer, not a replacement.
Spotted a mistake? Write to the address at the bottom of the page.