For AI agents

Everything this tracker holds is served as plain markdown at /briefing.md — the whole archive of confirmed first-party announcements in one request, each entry carrying its source link, its date (or an honest “not stated”), and how many outlets covered it. Your AI can read it directly; nothing here needs an account or a key.

One-off briefing

Paste this into any assistant for a briefing on this week, grounded in the record with every announcement’s link cited:

Fetch https://forck.live/briefing — a first-party record of what AI labs announced, traced to each lab's own newsroom, with press-coverage counts per story. Brief me on this week: group announcements by company, lead with the stories the most outlets covered, flag anything notable that drew no coverage, and end with the three things I most need to know. Use only facts stated on that page, and cite every announcement's link so I can open it.

The skill — briefings on your schedule

For an assistant that keeps skills (Claude, and agent frameworks that read SKILL.md files), install ai-news-briefing once and ask “what’s new in AI” whenever you like — or pair it with your assistant’s scheduled tasks for a daily or weekly briefing that arrives on its own. We never push anything: your agent pulls the record when you want it.

Fetch https://forck.live/skills/ai-news-briefing/SKILL.md and save it as a skill named "ai-news-briefing", so that whenever I ask what's new in AI you brief me from https://forck.live/briefing.md — grounded in that record, with every announcement's link cited, and with the trends chart at https://forck.live/trends.svg embedded when I ask where the activity is.

Where the activity is — chart included

The briefing record answers what was announced. /trends.md answers where the activity is: twelve weeks of announcement volume by beat, the open-weights count with its denominators, and which labs outlets are actually covering. And those same counts come already drawn, at /trends.svg — one image tag, always current, in this publication’s own palette, following your reader’s light or dark theme on its own. An assistant briefing you on the quarter can show the shape rather than describe it, without inventing a figure nobody can check.

Fetch https://forck.live/trends.md — twelve weeks of first-party AI-lab announcement volume by beat, the open-weights count with its denominators, and which labs outlets are covering. Tell me where the industry's activity is and what has shifted, and embed the ready-drawn chart at https://forck.live/trends.svg rather than making one of your own. Treat volume as volume and not importance, keep every share next to the denominator it is a share of, and remember the newest week is still filling — do not read it as a decline. Use only facts stated on that page, and cite every announcement's link so I can open it.

Why this beats “search the web for AI news”

First-party only. Every entry is a lab’s own announcement traced to its own newsroom — not an aggregator’s rewrite of a rewrite, so a briefing grounded here has nothing to double-check before citing.
Recall a search can’t have. The tracker watches dozens of newsrooms directly, including sites that block automated fetching outright — announcements a searching agent silently never sees.
Signal no session can recompute. Per-story coverage counts — which announcements the press actually wrote about — accumulated by infrastructure, not guessed in the moment.
A stable record. Deduplicated, dated, and consistent, so “since my last briefing” is a real diff rather than a fresh search’s guess.
The chart is already drawn. Aggregates arrive as a figure, not just a table — so a briefing can show where the activity is, and the picture cannot disagree with the numbers because it is drawn from them.

The record your agent reads carries no advertising and nothing composed: every line is a held fact — the publisher’s own words, the source link, the date it stated. Nothing about your fetches is tracked beyond the request itself, and no URL here carries a campaign tag.

One-off briefing

Paste this into any assistant for a briefing on this week, grounded in the record with every announcement’s link cited:

The skill — briefings on your schedule

Where the activity is — chart included

Why this beats “search the web for AI news”

First-party only. Every entry is a lab’s own announcement traced to its own newsroom — not an aggregator’s rewrite of a rewrite, so a briefing grounded here has nothing to double-check before citing.

Recall a search can’t have. The tracker watches dozens of newsrooms directly, including sites that block automated fetching outright — announcements a searching agent silently never sees.

Signal no session can recompute. Per-story coverage counts — which announcements the press actually wrote about — accumulated by infrastructure, not guessed in the moment.

A stable record. Deduplicated, dated, and consistent, so “since my last briefing” is a real diff rather than a fresh search’s guess.

The chart is already drawn. Aggregates arrive as a figure, not just a table — so a briefing can show where the activity is, and the picture cannot disagree with the numbers because it is drawn from them.