The Download

Morning Brew Inc., Getty Images

TL;DR: Yesterday, Meta CEO Mark Zuckerberg laid out his dream for the agentic future with a one-minute ad: “Call us optimists. Call us dreamers. Just as we’ve always done, we’re betting on people.” The ad scrolls through a montage of happy human moments, set to the David Bowie song “Five Years”—which just so happens to be about the end of the world. But for Meta, “betting on people” has looked different in practice.

Here are the contradictions so far this year:

• Meta told staff in April it would cut 10% of its workforce—about 8,000 jobs—and scrap 6,000 open roles to offset AI spending.
• As for the workers who stayed: Roughly 6,500 engineers were reassigned to a new AI unit. Just a few months later, many described themselves as “draftees,” with one calling the team “literally the gulag,” per Wired.
• Also in April, Meta said it would start logging US staffers’ clicks and keystrokes for AI training data—but then scaled back its plan in June following weeks of backlash from employees.
• Dozens of employees sued the company this month over claims that it used AI tools to rank worker performance without making exceptions for those who were on leave. (A Meta spokesperson said the claims lack merit.)
• Instagram acknowledged the “creep problem” the Meta glasses have created and said it will remove videos of those harassing strangers in public—but didn’t say how it’ll enforce the ban. Meta is also reportedly testing always-on “super sensing” glasses that capture continuously and may even drop the recording light that tells people when they’re on.
• A new Facebook Verified badge, announced today, aims to prove a human is behind a Marketplace listing—on a platform (and now, a standalone Seller app) where Meta AI can write the listing, set the price, and answer questions.

Chasing TikTok (again): Amid the announcements is another move nobody asked for—Facebook will test a “reimagined experience” that takes users straight into full-screen video when they open the app. It’s also considering changes that sound suspiciously similar to it just becoming TikTok.

Bottom line: If Meta wants people to buy into this latest ad—a Herculean feat considering it landed at 96 out of 100 on a 2026 reputation poll (just four spots above Spirit Airlines, which has the excuse of being dead)—it will have to start aligning its products with its message. —LC

      

Sponsored By Smartsheet

Most AI tools operate in a vacuum, disconnected from what’s actually happening around you. When interaction starts without context, your teams end up making decisions without the full picture.

Smartsheet is a work management platform designed to unite people, data, and content at scale. Ask its AI what needs attention and get an answer. Create summary reports, update statuses, analyze risks, or build a plan, all in plain language.

Describe a dashboard and it’s ready before Friday’s review. No more building from scratch. Smartsheet AI doesn’t just surface information—it can think through your actual project data and act on it.

See what’s possible when your AI actually knows your work.

How AI can find your weekend plans for you

I spend an embarrassing chunk of my time finding fun things to do in NYC. The city has so much going on that at this point I’m basically doing a very lengthy research project that never ends. So I decided to make AI do the research instead—and you can too.

Depending on which tool you use, you may see quite varied results (Gemini, for example, can pull from live ticketing and event data, which is good for surfacing hyperlocal underground stuff). The results will also only be as good as the info you give it (and, candidly, aren’t always perfect), so here’s the full prompt, and you can decide how detailed you want to get:

Act as my local events scout. Find events happening in [neighborhood/city] between [dates] that match these interests: [e.g., live indie music, food festivals, outdoor markets]. My budget is a maximum of [$] per ticket. I’m free [days/times], and I’m willing to travel up to [X miles]. For each event, give me: name, date and time, venue, price, why it fits, and a link to verify or buy the ticket. Double-check that each date and event are still accurate before listing them. Flag anything that might be sold out or unconfirmed, and tell me which you’re least sure about.

One thing to note: Chatbots still occasionally get a date wrong or list something that’s already sold out, so double-check the details before you hit purchase. And if you do like the results, you can always schedule the prompt to run on a certain day each week, so you don’t have to go digging for your weekend plans. —SM

The Zeitbyte

Elementary, dear human

Shannon May

AI has a disturbing party trick: It can do a mean Sherlock Holmes impression—deducing a lot about you from clues you unwittingly let slip in conversation, as a New York Times columnist found out. AI correctly guessed his income (in part because he once mentioned a German car he likes to fix up). It also knew he was in his 40s, largely due to him never asking “questions about late-night activities.” And then it went for the jugular, diagnosing him as “addicted to the illusion of control”—all because he asked how long it might take homemade sourdough bread to cool. (The reporter did not dispute this read.)

This kind of profiling will probably only get more unnervingly precise (and judgy) as AI models roll out beefier memory. But it’s handy if you don’t want to preface every request for restaurant recs with another reminder of your shellfish allergy. —WK

Chaos Brewing Meter: ☕☕/5

• Opus 5 is finally here, with some users claiming it’s nearly as powerful as Fable.
• OpenAI just rolled out ChatGPT Health to every US adult. But be warned: HIPAA rules likely don’t apply to what you share with it.
• For the first time since SpaceX’s IPO, Elon Musk sat down for an extended interview—and by some accounts, it’s his first genuinely tough one in over a year.
• You’ll start seeing Yelp reviews in ChatGPT soon. Let’s just hope it goes better than its last content-licensing deal, which didn’t exactly end so well…
• We’re driving right into the next RAMageddon-related price increase: Automakers are saying cars could be next.
• And now it’s Anthropic’s turn: Weeks after OpenAI’s voice update, Claude’s voice mode is getting access to its smarter models, too.
• Huang just posted on X for the first time ever—to share a three-page letter from some of tech’s biggest names.
• This “old-school” search engine wants to be an ad-free, pre-AI alternative to Google—and makes an interesting case for why you should pay a small fee for it.

^{*A message from our sponsor.}

This time last week...

Readers’ most-clicked story was about Tesla fans, who have been begging for an e-bike. Instead, they got a $225 toddler push-bike that looks like something out of Tron.