Week of June 22, 2026 · Updated weekly

Pulse

A weekly attempt to isolate real trends from mere conversation — across AI & society, AI & cognition, AI & work, and health AI. Bubble size and signal strength come from a transparent four-part rubric — source authority, breadth, primacy, and persistence — not metrics theater.

Meta put a number on the restructuring — ~8,000 jobs — while Brussels prepared to blink on the AI Act's hardest deadline and students told RAND they're using AI more even as they trust it less with their own minds.

What's bubbling up

Bubble size scales with signal strength, relative to this week; each card shows the exact 0–100 score. Click a bubble for the full read.

The timeline

How strength is scored

Each trend's strength is a transparent rubric, not a gut number. Four inputs, each scored 0–3, summed and scaled to 100:

Authority — the peak source: a peer-reviewed journal, arXiv paper, government text, or court filing (3), down through a major outlet or established analyst (2) to a trade blog or aggregator (1).
Breadth — independent outlets covering it, not reposts of one wire story: 6+ (3), 3–5 (2), 1–2 (1).
Primacy — whether a primary document anchors the story (the study, the ruling, the filing) rather than secondary commentary.
Persistence — building across multiple weeks (3), a fresh spike (2), or fading (1). The one input that separates a trend from a Tuesday.

strength = (authority + breadth + primacy + persistence) ÷ 12 × 100

Momentum is simply this week's score against last week's. The method is reproducible, not objective — the 0–3 calls are still human judgment, and breadth rewards whatever the press chooses to chase. Each card below shows its four sub-scores, so you can argue with the math rather than the mood.

This week's trends

AI & Work ▼ Cooling

Meta puts a headcount on the AI pivot

Meta began ~8,000 layoffs as part of an AI-focused restructuring, with roughly 7,000 more employees reassigned to AI teams — an explicit reallocation of people toward the technology reshaping their work, not an efficiency trim dressed as strategy.

Why it matters Reassigning ~7,000 people while cutting ~8,000 makes 'reshaping, not replacement' hard to sustain at scale. The reassignment is the underexamined part: thousands told that proximity to the model now outweighs years of craft — a disruption severance doesn't address. Related reading →

Authority 2 Breadth 3 Primary 2 Persistence 2

AI roundup (Jun 22)Yale Budget Lab PwC Jobs Barometer

AI & Society — Steady

Brussels prepares to blink

With the EU AI Act's high-risk obligations due August 2, the Commission's 'Digital Omnibus' proposes deferring them to December 2027 — adoption expected in July. But GPAI-provider enforcement and Article 50 transparency duties (disclose AI interactions, label synthetic content) still land August 2. Penalties reach €35M or 7% of global turnover.

Why it matters The split in what gets deferred is revealing: the expensive obligation — proving a high-risk system safe before deployment — slips to 2027, while the cheap one — disclosing you're talking to AI — holds for August. Honesty about what AI is is far easier to mandate than competence at what it does. Related reading →

Authority 3 Breadth 2 Primary 3 Persistence 3

EU AI Act tracker Latham & Watkins DLA Piper

AI & Cognition — Steady

Students use it more, trust it less

A new RAND American Youth Panel report finds more U.S. students using AI for homework and more of them believing it harms their critical thinking — the two curves rising together. The self-awareness is real; the behavior doesn't follow it.

Why it matters Rising usage alongside rising concern shows the problem isn't ignorance — students can name the cost and reach for the tool anyway, because the friction it removes is the friction that felt like effort. That makes it a design problem more than an awareness one: where to reintroduce difficulty so the choice isn't left entirely to willpower. Related reading →

Authority 2 Breadth 2 Primary 3 Persistence 3

RAND Forum for Linguistic Studies Nature HSSC

Health AI — Steady

The clinical floor drops as agents arrive

Three currents converge on one calendar: the FDA's January guidance loosened oversight of AI decision-support; ARPA-H's ADVOCATE program is selecting teams this month to build the first FDA-authorized agentic clinical AI — including a patient-facing agent that can autonomously adjust meds, diet, and appointments; and hospital AI-disclosure enforcement begins June 30.

Why it matters Loosened oversight and increased autonomy are arriving on the same calendar, inverting the order in which trust is normally earned. The meaningful line is between an assistive tool that drafts for a human to approve and an agent that acts on a patient unsupervised — two different relationships to accountability, diverging just as the regulatory floor drops. Related reading →

Authority 3 Breadth 2 Primary 3 Persistence 3

KevinMD Manatt Health AI Tracker npj Digital Medicine

Worth your attention

Gemini 2.5 Pro 'Deep Think' lands as China commits $295B to AI infrastructure AI roundup Capability leaps and state-scale capital landing the same week the EU softens its rules. The frontier accelerates; the guardrails defer.
EU AI Act: the live countdown to August 2 AI Act tracker Whatever Brussels defers, the transparency clock keeps running. Worth watching which obligations actually survive July.

How this works: each week I sweep research feeds, essays, and the news across these themes, separate durable trends from passing conversation, and score each one with the rubric above — then explain why it matters. Sources are linked; disagreement is welcome — find me on LinkedIn.