Welcome, humans.

So the AI economy is doing a very weird split-screen right now. Tech companies cut roughly 63,000 jobs in June, with some employers citing automation and others redirecting resources toward AI infrastructure like chips and servers.

Meanwhile, creators are using that same cheap intelligence to make things nobody had budget for before. One built Gateway Isle, a fake archival tourism film for an imaginary island so convincing that viewers started asking for documentaries and games set there.

Another made fake Lord of the Rings behind-the-scenes footage where Gandalf's takedown of the Balrog looks like a practical effect. Man, could you imagine?

Now, temper your ire: these videos did not cause said layoffs. But the juxtaposition is the whole AI economy in miniature: companies are spending aggressively to make intelligence cheaper, while everyone else is discovering what cheap intelligence lets them make. So far: fewer coworkers, but significantly better fake DVD BTS footage. Speaking of DVDs… did y’all know they’re making a comeback??

Here’s what happened in AI today:

• 😺 Meta paired its superintelligence manifesto with local Muse Glimmer.

• 📰 Anthropic pushed a famous Riemann bound from 41.6% to 67.2%.

• 📰 Nvidia and Wall Street explored $500B in AI infrastructure financing.

• 🍪 OpenAI launched GPT-5.6-Cyber as frontier hacking capability climbed.

• 🎓 ChatGPT Scheduled Tasks can monitor changes and alert you automatically.

…and a whole lot more that you can read about here.

😺 Meta Wants You to Own Superintelligence, Starting With a 30B Agent

Mark Zuckerberg spent Monday making a very Meta argument about the future of AI: superintelligence should not belong to one company, government, or tiny group of people. Then Meta made that philosophy less theoretical by releasing an AI agent you can actually download and run yourself. Anthropic would NEVER.

Here's what happened:

• In a 6,500-word manifesto, Zuckerberg argued for "personal superintelligence" that works toward each user's goals, including private agent modes, broad free access, paid compute auctions, and a $1B fund for communities hosting Meta data centers. Zuck, do you read our NL?!?

• Muse Glimmer is a roughly 30-billion-parameter open-weight model, meaning Meta publishes the model files so people can run and modify it instead of only calling a Meta server.

• Meta tuned Glimmer for coding, images, tool use, long-running workflows, and recovering when an agent gets stuck or fails.

• Unsloth says a quantized version can run in about 18GB of memory on one consumer GPU. Quantization stores the model's numbers (the at lower precision, shrinking memory use while keeping most of its capability.

Why this matters: The frontier-model fight has mostly been about who has the smartest model. Meta is opening a second front: who lets you control it. A capable local agent can keep more data on your machine, keep working without a cloud connection, and be customized without asking the model provider for permission.

That does not make Glimmer "superintelligence," by, uh, any stretch of Zuck’s imagination (not yet anyway), and critics such as Aaron Scher argue that distributing increasingly autonomous systems creates its OWN safety problems. But it turns Zuckerberg's philosophy into an actual product strategy. And ya know what? We respect it. Glad to see Meta actually releasing open AI again. Unlike, ya know… “Open”AI.

Our take: Meta's strongest weapon against closed AI labs may be the same one it used back during the Llama days (Meta’s older open models): make ownership cheap enough that developers expect it. If local agents become good enough for everyday work, the question stops being "which chatbot should I subscribe to?" and starts becoming "why am I renting a brain I could own?"

Oh, and deeply related: Stripe’s reported $10B talks to buy OpenRouter have kicked off a bidding frenzy around AI routing startups, with companies like Snowflake now circling the space too. OpenRouter lets you access lots of different AI models in one place, switch between them easily, and pay through a single account.

FROM OUR PARTNERS

Forrester studied the ROI of Slack. Here's what they found.

The productivity gains from AI are real. A new Forrester study puts a number on them.

A new Total Economic Impact™ study by Forrester quantified what actually happens when enterprises run work through Slack as their operating system. The findings: 312% ROI over three years, $51M in total benefits, and payback in less than 6 months.

The biggest driver? $50M in collaboration and efficiency gains from teams that stopped switching tabs and started executing in real time, with AI, data, and people all in one place.

See what Slack can do for your organization.

Read the study

🎓 AI Skill of the Day: Make a Scheduled Briefing That Reports Only New Changes

Recurring AI briefings get annoying when they repeat yesterday's answer with a fresh timestamp. ChatGPT's updated Scheduled Tasks can run one-off or recurring jobs, remember earlier monitoring runs, and notify you only when a meaningful change appears.

The useful move is to define the change before you schedule the task:

1. Name the topic and the sources that count.

2. State what qualifies as meaningful, such as a price change, product release, deadline, or official filing.

3. Tell it to compare against previous runs, report only the delta, and say "No meaningful change" when nothing qualifies.

4. Add an end condition so a temporary monitor stops after the event passes.

ChatGPT says tasks can run no more than once per hour, and unattended tasks may pause after inactivity. Review the Scheduled page occasionally so an important monitor does not quietly go dormant.

Every weekday at 8:00 AM, check [TOPIC] using [SOURCE LIST]. Compare the findings with previous runs. Notify me only if [DEFINITION OF MEANINGFUL CHANGE]. Include the source link, publication time, what changed, and the action I should consider. If nothing qualifies, reply: "No meaningful change." Stop this task after [END CONDITION].

Have a specific skill you want to learn? Request it here.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

1. *Adobe's ChatGPT plugin lets you create and edit images, video, designs, and PDFs with 70+ Adobe tools inside ChatGPT. Free to try.

2. GPT-5.6-Cyber gives approved security researchers an OpenAI model tuned for vulnerability research, exploit validation, and other advanced defensive cyber work. Daybreak access; pricing not public.

3. Perplexity's Stripe connector lets you query revenue, customers, invoices, and subscriptions, then issue refunds or create payment links from chat. Pricing varies by Perplexity plan.

4. Google Ads and Analytics can now answer campaign questions, surface personalized insights, and build dashboards from prompts. Existing Google Ads/Analytics pricing applies.

5. Imagine Image 2.0 gives you regional edits, background removal, smart resizing, templates, and up to five reference images in Grok. Pricing not public.

6. Xirp from Spotify gives coding agents institutional memory of services, owners, docs, and architecture so they stop starting every session from zero.

7. Stagehand v4 gives browser agents a Playwright-style API with faster, more token-efficient actions that can recover when webpages change. Open-source SDK; cloud pricing varies.

📰 Around the Horn

• Dyna Robotics introduced Dyna-2, a robot world-action model trained on one million hours of human video that reached 87% zero-shot quality at new sites without site-specific retraining.

• Anthropic made Claude Sonnet 5's introductory $2 input / $10 output pricing per million tokens permanent instead of raising it later this month.

• Claude is adding invisible statistical watermarks to output from new models, encoding the signal into token choices so it can survive copy-paste and light edits.

• Discovered Materials shows what agent swarms look like outside software, using Anthropic-powered agents plus physics simulations to hunt for cooler, more efficient chip materials.

• More than 500 U.S. jurisdictions have restricted new data centers, turning local permitting into a growing constraint on the AI compute buildout.

• AI agents are now taking entire online college courses for cheating students, from watching lectures to writing papers and joining class discussions.

FROM OUR PARTNERS

A shocking 95% of organizations have delayed AI projects - and it's spurring major data infrastructure redesign. Read more.

Download the report now

📣 Tuesday Tell All

AI is getting better at doing more work per person. But if that productivity actually shows up in your job, who should capture the upside?

🆕 NEW FROM THE NEURON: Build AI Agents for Total Beginners

Want to build an AI agent that actually does useful work? We turned our two-hour live workshop with agent builder James McAulay into a step-by-step beginner guide covering memory, MCP, reusable skills, testing, and scheduled agents.

Read the guide + watch the demos here.

A Cat’s Commentary

Btw: We just launched a robotics newsletter! Sign up for it here.

And before you go… have you subscribed to our YouTube Channel? If not, can you?

P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.