Who gets the $7 trillion?

Quick Take · Lead · Put AI to Work · Looking Ahead

In this issue

Newsom signed a first-of-its-kind executive order to prepare California's workers for AI disruption — not by slowing AI down, but by making sure workers share the gains. WARN Act overhaul, employee-ownership models, a retraining "AI playbook."
Meta spent the same month recording its employees' keystrokes to train AI, then laid off 8,000 of them. The contrast is the whole argument.
Spotify and Universal Music struck a deal letting fans make AI covers, built on consent, credit, and pay for the artists. A model for AI that compensates creators instead of strip-mining them.
Trump pulled his own AI security order at the last minute. The federal vacuum holds; the states keep moving.
A 30-minute AI workflow to build a plain-language benefits guide for the people in your community navigating a job loss.

Quick Take

Goldman Sachs thinks AI could add seven trillion dollars to the global economy. Set aside whether that number is exactly right. The point is that a windfall is coming, and there is only one question that actually matters about it: who gets the money?

For most of the last two years the progressive answer to AI has been some version of "slow down." This week California tried a different answer. Don't slow it down. Steer it. Make sure the people whose work generates the gains are the people who share in them. It's the most hopeful thing a government has done on AI in a while, and it landed the same week Meta gave us the bleakest possible version of the alternative.

Let's get into it.

$7 trillion

Goldman Sachs' estimate for how much generative AI could add to annual global GDP over a decade — roughly a 7% bump. The same analysis figures AI could expose the equivalent of 300 million full-time jobs to automation. Both numbers are real, and they're the same story: there's an enormous amount of value about to be created, and nothing about the technology decides who captures it. That's a policy choice. This week, one state started making it.

California decided to prepare its workers, not just brace for impact

On May 21, Governor Newsom signed what his office is calling a first-of-its-kind executive order to prepare California's workers and businesses for AI-driven disruption. The framing is the part worth slowing down on. It doesn't try to block AI, slow deployment, or fight the technology. It starts from the assumption that AI is going to reshape the labor market, and asks a different question: how do we make sure workers come out of that with something?

The order directs state agencies along three tracks. The first is about workers capturing the upside. It tells agencies to evaluate employee- and worker-ownership models so productivity gains build wealth for the people doing the work, to find where collective bargaining produces good outcomes, and to build AI preparation into higher education and on-the-job training. The second track is about seeing the disruption coming: a public dashboard of AI's impact by sector, AI-readiness signals folded into the monthly jobs report, and, within 180 days, recommendations to overhaul California's WARN Act so layoff warnings actually arrive early. The third is about catching people when a job goes away. That means reviewing severance and equity-compensation standards, raising awareness of unemployment and stability programs, an "AI playbook" to modernize job training and connect dislocated workers to help, and a single online front door for state services.

Newsom: "This moment demands we reimagine the entire system — how we work, how we govern, how we prepare people for the future." First Partner Jennifer Siebel Newsom added the point most of the coverage skipped: women face disproportionate risk of displacement and widening inequality as AI evolves.

Why this matters

This is the proactive version of everything this newsletter has been arguing for. We've covered workers organizing to set the floor. We've covered communities demanding a cut of the infrastructure. This is the layer above both: a government deciding, before the worst of the disruption hits, that the transition itself is something you can plan for instead of just absorb.

It also quietly resolves a fight the left has been having with itself. "Pro-worker" and "pro-AI" have been treated as opposites — you either defend jobs or you embrace the technology. This order says you can do both, and that the way you do both is policy. Worker ownership. Retraining funded before the layoffs, not after. Early-warning systems. Severance standards. None of it requires pretending AI isn't coming, and none of it requires leaving workers to face it alone.

The honest caveat: it's an executive order, not a law. A lot of it is directives to study, recommend, and report. The WARN Act overhaul is a recommendation due in 180 days, not a done deal. But "first-of-its-kind" is doing real work in that headline. No other state has built a workforce-transition framework like this, and executive orders are how the durable ones usually start.

What you can do

If you work in advocacy or policy in any other state, this is a template you can hand your legislators tomorrow. The question to ask: is there any bill or executive action on AI workforce preparation — not regulation, preparation — on the table this session? Retraining, WARN Act modernization, worker-ownership incentives, severance review. If the answer is no, California just wrote the first draft for you, and the text of the order is public. Bring it to the next meeting with your state rep or your governor's policy staff.

The same month, Meta showed us the other timeline

While California was writing a plan to help workers through the AI transition, Meta ran the opposite experiment. The company laid off 8,000 employees beginning May 20, part of offsetting a reported $145 billion in AI infrastructure spending this year. That's the headline number. The detail underneath it is worse.

Meta has been running software on employee machines that captures keystrokes, clicks, mouse movement, and periodic screenshots across the apps and sites they use, then feeds that data into its AI training pipeline. When workers asked to opt out, the company said no. More than 1,000 employees signed a petition demanding it stop. Meanwhile the company's CTO, Andrew Bosworth, described the destination plainly: "The vision we are building towards is one where our agents primarily do the work, and our role is to direct, review and help them improve."

Read those two facts together. Employees' own work was recorded to train the AI. Then thousands of those employees were cut because the AI can increasingly do the work. The people were the training data for their own replacement.

Why this connects

This is the timeline the Newsom order is trying to avoid. Not "AI exists," but "AI gets built on workers' backs and the workers get nothing — not consent, not a share of the gains, not a soft landing." Meta is the cautionary tale that makes California's bet legible. When there's no policy floor, the most powerful companies set the terms, and the terms are extraction.

What you can do

If your organization is rolling out AI internally, the Meta story is the anti-template. Two questions are worth putting in writing before any tool goes live: are we monitoring staff to build or train these systems, and if so, did people meaningfully consent? "We updated the handbook" is not consent. Worker data dignity is going to be one of the defining AI fights of the next few years, and the orgs that get it right early will be the ones that can credibly ask anyone else to.

Progressive AI Win

An AI deal that actually pays the artists.

For two years the story of AI and creative work has been theft: models trained on other people's writing, art, and music without permission or payment. This month Spotify and Universal Music Group announced something structurally different. A licensing agreement that will let fans create AI-generated covers and remixes of participating artists' songs, built — in the companies' own words — on "consent, credit, and compensation."

The mechanics matter. Artists and songwriters opt in. They get credited. And they share directly in the revenue the AI covers generate, a new income stream on top of normal streaming, launching as a paid add-on for Spotify Premium. No pricing or launch date yet, and the details will decide whether the revenue split is actually fair. But the shape is the point: this is the first major-label AI deal that treats creators as participants to be paid rather than raw material to be scraped.

It's worth holding up precisely because it's a corporate deal, not an act of charity. It shows the "consent and compensation" model isn't economically impossible — it's a choice. Which means every AI product that doesn't offer it is choosing not to. That's a useful thing for organizers, unions, and creators to be able to point at.

From our friends

Your org deserves its own AI. Not Big Tech's.

Change Agent is a private AI platform built for nonprofits, unions, and advocacy orgs. Your data stays yours, it plugs into tools you already use (Google Drive, Slack, ActBlue), and it handles the tedious stuff so your team can focus on the mission. Starts at $35/month. Small nonprofits under $1M can apply for discounted pricing.

Learn more

Put AI to Work

Practical ways progressives can use AI this week

Build a plain-language benefits guide for someone facing a layoff

The Newsom order calls for a single online front door for navigating government services, because right now there isn't one. The help exists, but it's scattered across agencies, written in bureaucratese, and nearly impossible to navigate when you've just lost your income. You don't have to wait for the state to build the portal. You can build a small version for your community this week.

Here's a 30-minute workflow.

Gather the real source material. Pull up your state's actual pages for unemployment insurance, job-retraining grants, WARN Act notices, and any emergency assistance programs. Copy the real text. The whole exercise depends on the AI working from the genuine rules, not its general knowledge — eligibility numbers and deadlines change, and you don't want to hand someone a hallucinated benefit.

Ask for plain language, not a summary. Paste the source text into Claude or ChatGPT and prompt: "Rewrite this at an 8th-grade reading level. Keep every dollar figure, deadline, and eligibility rule exact. Format it as: Who qualifies / What you get / How to apply / What you'll need." Do this for each program.

Build the decision tree. "Turn these programs into a short set of questions that routes someone to the ones they're eligible for." Now you have a guide that meets people where they are instead of making them read four agency websites.

Have a human check it against the source. This is not optional. AI is good at clarity and bad at staying current. Before anything goes out, one person verifies every number and deadline against the official page. Build that check into the process, not as an afterthought.

Share it where people actually are. A one-page PDF, a section on your site, a printout at the front desk. The point isn't a polished product. It's that someone who just got bad news can find out in five minutes what they're entitled to.

Looking Ahead

A few threads to watch.

The 180-day clock on California's WARN Act recommendations started May 21. That report — due in November — is where we'll find out whether this order has teeth or stays a framing exercise. Worth a calendar note.

Trump pulled his own AI security executive order at the last minute on May 20, the same week California acted. The order would have asked AI companies to share advanced models with the government up to 90 days before public release; the president said he "didn't like certain aspects of it" and didn't want to slow America's lead. The federal vacuum on AI holds. The states keep filling it.

And Apple's WWDC lands June 8, with the company having quietly registered genai.apple.com. After two years of being the conspicuous no-show in consumer AI, Apple putting generative features in front of a billion devices would reset what "everyday AI" means for a lot of people. We'll cover what's worth your attention and what's marketing.