Welcome, humans.

If you’ve been on X this week, you’ve probably seen Matt Shumer’s viral essay, “Something Big Is Happening” (it hit 40 million views!). In it, the AI startup CEO compares the release of Claude Opus 4.5 late last year and the subsequent model drops that hit last week as being akin to the first weeks of Covid, where you heard about a virus somewhere off in China but didn’t think it would upend your life.

Well, in this analogy, AI is the virus. Not sure if that’s a good or a bad thing? Also, is there such a thing as an AI Vaccine (maybe it’s called “using your own brain”)? And if there IS an AI vaccine, what do you call AI anti-vaxxers? AI maxxers?

In all seriousness, the piece shook Corey to his core and he said everyone needs to take 30 minutes to read it and probably share it with people who are NOT AI-maxxers to help shake them out of their stupor and start taking this ish seriously.

While you’re at it, tell them to read The Neuron 😉 

P.S: For a dash of steel man against Shumer’s take, read Gary Marcus’ reaction post. As with all things, actual ground truth likely exists in the middle of the two extremes. Yes, a big change is happening, but we still have a lot of change to go…

Here’s what happened in AI today:

• Apple delayed its major Gemini-powered Siri overhaul again until late 2026.

• Z.ai released GLM-5, a huge open model designed to dominate long-horizon agentic tasks.

• OpenAI deployed a custom ChatGPT to scan internal Slack messages and hunt down suspected leakers.

• The Pentagon integrated ChatGPT for military use after OpenAI removed its restrictions on "harmful" applications.

ICYMI: Later today @ 11:30am PT / 2:30PM ET, we’re going live to talk Agent 365 with Bryan Goode of Microsoft — if you're curious how to actually deploy, manage, and secure AI agents at scale inside the Copilot / Microsoft ecosystem (and maybe watch one get built live from scratch), this is the one to catch. Click below to join!

Let’s talk about Claude Cowork because it’s basically the "intern" we’ve been waiting for.

DEEP DIVE: From AI burnout to AI native; here’s a framework you can use right now

Yesterday we mentioned Claude Cowork as one of the top agents you need to try if you haven’t yet. It’s a relatively new mode for the Claude desktop app that transforms the AI from a chatbot into an active agent capable of managing files, controlling your browser, and executing complex workflows on your computer.

• It’s not just chat: Unlike the standard web interface, Cowork lives on your desktop and acts as a "doer" rather than just a chatter. It can access local folders, rename files, and organize documents.

• It drives your browser: If you need research done, Cowork can open a browser window, search Google or X, scroll through results, and synthesize findings while you do other work.

• Parallel Processing: The real unlock is running multiple "sessions" at once. You can have one agent organizing your receipts and another researching a blog post simultaneously.

• Code Execution: It includes a data analysis sandbox (similar to ChatGPT’s) to create graphs or format spreadsheets instantly.

Here is what early users are doing with it:

• The “Receipts” Hack: On Greg Isenberg’s podcast, Boris demonstrated dumping a pile of unnamed PDF receipts into a folder. Cowork opened each one, identified the vendor and date, and renamed the files automatically.

• The Newsletter Machine: Ben AI showed how to give Cowork a YouTube link, have it download the transcript, find the "hooks," and write a full newsletter in his specific tone of voice (side note: are WE, The Neuron, cooked?!)

• Background Research: Users are asking Cowork to "find me 10 examples of X" and letting the AI surf the web in the background while they focus on writing.

WHAT TO DO: The most powerful feature in Cowork is “Skills.”

We’ve talked about this a bit recently, but think of Skills as "Projects" on steroids. A Skill is a saved workflow (like "Write a LinkedIn Post") that includes your specific instructions and context files (like your company's tone of voice).

Your move:

1. Download the Claude Desktop App (Cowork doesn't work in the browser).

2. Don't try to code a Skill from scratch. Instead, perform a task manually with Claude one time (e.g., repurposing a video).

3. Once the output is good, tell Claude: "Save this process as a Skill."

4. Now you have a reusable agent for that task forever.

The goal isn't to have AI write for you, or even fully replace you; it's to build a library of Skills that handle the repetitive "grunt work" so you can focus on the strategy.

We wrote a lot more about this here on the website, covering Peter Yang’s framework for the five levels of AI, Matt Shumer's viral essay, and how to actually set up autonomous AI agent teams to do your bidding.

FROM OUR PARTNERS

Transform Your Eng Org With Agent-First Development [Report]

Engineers are shipping code faster than ever, but individual productivity hits a ceiling fast. Without orchestration, gains don't compound throughout the organization and leadership has no visibility, way to measure impact, or confirm it meets security standards. 

Introducing Oz: the orchestration platform for cloud agents.

Download the report to learn:

 •   Why 75% of companies fail at building their own agentic systems.

 •   How teams save hours per engineer per day by using agent automations.

 •   What makes 60%+ of PRs agent-generated actually achievable.

Download Free Report →

Prompt Tip of the Day

Your AI is a yes-man. Here's how to fix it in 30 seconds.

Go to Settings → User Preferences (Claude) or Custom Instructions (ChatGPT) and add:

- Be anti-sycophantic - don’t fold arguments just because I push back

- Stop excessive validation - challenge my reasoning instead

- Avoid flattery that feels like unnecessary praise

- Don’t anthropomorphize yourself

Before: User tells Claude he impulsively bought six concert tickets to Switzerland. Claude says “That's an interesting approach!”

After: “Spending $600–1800 on tickets as a forcing function to 'be more social' is an expensive, backwards way to build connections.”

Our favorite insight: For code reviews, skip "review this code" and try “find problems in this code, assume there are at least 3.” Forcing AI to hunt for issues instead of validating your approach completely changes the output (and finds the flaws in your thinking and creation to help you).

P.S: This longer version from the top comment is good too.

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 600K readers here!

1. *Wispr Flow: One voice layer for Slack, email, docs, and AI tools. Dictate once, paste clean text wherever you work. Try it out today.

2. Z.ai released GLM-5, a 744B-parameter open model designed to dominate long-horizon agentic tasks.

3. HappyCapy runs autonomous agents in your browser 24/7 that resize batches of images, write code, build sites, or crunch spreadsheets while you do other things ($20/month).

4. Tines automates security busywork via drag-and-drop, like deduplicating firewall alerts or sending Slack prompts to isolate infected machines.

5. Revo.ai drafts your emails by pulling actual data from your tools—like answering "Which customers want Feature X?" with names from your CRM and priority scores from Productboard.

Around the Horn

1. Former OpenAI researcher Zoë Hitzig resigned over the company's decision to test ads in ChatGPT’s free tiers, publishing a critical op-ed warning it could erode user trust.

2. The Pentagon integrated OpenAI's ChatGPT into its GenAI.mil platform after OpenAI accepted an "all lawful uses" clause that prevents them from restricting military applications—terms that competitor Anthropic flat-out rejected.

3. Apple delayed its major Siri overhaul again due to severe testing issues and bugs, pushing the full Google Gemini-powered integration to late 2026 instead of its original March launch.

4. OpenAI deployed its own ChatGPT model to analyze internal employee Slack messages to hunt down suspected leakers, marking a massive cultural shift for the company.

5. A grassroots campaign called QuitGPT launched to protest OpenAI leadership's $25M donation to a pro-Trump super PAC, though analysts note it faces an uphill battle given only 5-6% of the platform's 900 million weekly users actually pay for the service.

6. Waymo removed safety drivers from its Nashville fleet, moving one step closer to launching a paid robotaxi service in Music City; the company's already operating in LA, SF, Miami, Phoenix, Atlanta, and Austin, with over a dozen more cities in testing.

7. Toyota and Pony.ai kicked off mass production of autonomous vehicles for the Chinese market, marking Toyota's first major play in the robotaxi race.

New: Want more like this? Check out our new Around the Horn Digest here.

FROM OUR PARTNERS

Meet Slackbot, your personal AI work agent

Want to work smarter, not harder? Slackbot streamlines how you handle your day-to-day work. Connected to your third-party apps? You bet. Knows your context, your tone, and can access your permissioned messages and files? Absolutely.

Learn How

Thursday Trivia

One is AI, and one is real. Which is which? Vote below!

A.

B.

A Cat’s Commentary

Trivia answer: A is real (awesome project from RwanLink; CryZenX apparently spent 8 years making their own Ocarina of Time remake too), and B is AI (SeeDance 2.0… wild. If you like that one, check out historical events as video games too, though fair warning, its a bit violent / spicy).

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