Editorial policy

Weekly Dose of AI is a ranked weekly briefing, not a firehose. The editor publishes by hand. Nothing auto-publishes.

How stories are selected

Candidates come from a fixed source list: lab announcements, primary documentation, and a small set of independent newsrooms. Stories rank by consequence for people who build or run systems — APIs, models you might actually run, infrastructure, security, and cost. Virality is last. Celebrity, routine funding, and consumer toys stay out unless they change how software gets built.

A human editor marks each candidate approve or skip. The briefing is drafted only from approved stories.

How sources are evaluated

1. Official product announcements and documentation
2. Research papers and public repositories
3. Regulatory or standards documents
4. Direct statements from the organization that shipped the work
5. Reliable independent reporting, when a primary page is missing

Unsourced social posts are not enough. If a claim is reported rather than confirmed, the briefing says so. Rumors, leaks, and previews are labeled.

Skip, Skim, or Act

That label is editorial judgment, written for a technical reader, not a score.

• Skip — no repo, API, infra, or on-call consequence this quarter for most readers.
• Skim — worth knowing if you already work in that stack; nothing to implement today.
• Act — a model id, API, price, control, or security default you should review this quarter.

Facts sit under “What happened.” Interpretation sits under “Why it matters” and the care label. They are not the same thing.

Research and drafting

Research may be assistant-aided. The editor owns approve/skip, the wording, and the publish click. The public briefing is the record, not a chat log.

Sponsored and affiliate content

There is none at launch. If that changes, it will be labeled on the page and will not replace a ranked news briefing.

Corrections

How to report an error, and how a fix appears, is on the corrections page.