3 things that happened while you were busy
1 ChatGPT moves into Chrome and can act across your open tabs
OpenAI's upgraded Chrome extension lets ChatGPT see the page you are on, and ChatGPT Work can now take actions across several tabs at once. Pro subscribers also get faster GPT-6 responses and a way to track each helper agent in multi-agent jobs. It arrives less than two weeks after Claude moved into Google Docs, Sheets, and Slides. Add the extension.
2 Anthropic's agents now log six times more hours than its staff
A new Anthropic report says that by August, agents working on behalf of employees outnumbered people 6 to 1 in hours worked. The company has also slowed hiring for junior technical roles. Separately, Anthropic plans to put 5% of its shares into a public benefit fund, pending shareholder approval.
3 AI cloud stocks fall more than 20% in a week
Shares of CoreWeave, Nebius, and IREN, the so-called neoclouds that rent out AI computing power, each fell by more than 20% last week. Goldman Sachs points to two causes: borrowing is getting more expensive, and investors are questioning whether the economics of AI infrastructure add up.
Gemini, Claude, or GPT? Each top model now has its own strength
Gemini 4 Argon for tools and visuals. Google's flagship stands out when a task needs several tools in a row. Sundar Pichai showed it handling CAD animations and 3D work, and one user had it build a detailed pagoda temple on its own. It even drew a pelican riding a bike out of spinning circles, set to AI-generated music.
Claude Opus 5.5 for long jobs and writing. Anthropic's model is known for sticking with long tasks and for writing that sounds natural. It is also good at building visuals from code alone: one user made an entire ancient Greek world with no image files, and this thread collects ten more builds, from motion graphics to games.
GPT-6 Astra for 3D and computer tasks. OpenAI's model shines in 3D modeling and hands-on computer work. One developer built a 3D game in 45 minutes by connecting Codex to Blender, and another had it create an interactive model of a V8 engine.
How to choose. There is no single best model anymore. Match the model to the job: Argon for multi-tool workflows, Opus for long projects and polished writing, and Astra for 3D and on-screen tasks. If you pay for only one, pick the one that fits the work you do most often.
📈 A Hidden Downturn? The Kobeissi Letter argues the S&P 500 is in a "stealth" bear market even near record highs, pointing to a narrow rally, energy-driven inflation, and rising rate expectations (4M views).
📺 Ad-Free TV. Ryan Doyle used Claude Code to strip ads and bloatware from his TV, which also made it faster (2.8M views). Others are following this step-by-step guide.
🪰 A Fly in the Machine. Odyssey showed a model of a fruit fly's brain flying through simulated worlds on its own (4.7M views).
🚨 Emails That Trick Your AI. Attackers are hiding instructions inside phishing emails to manipulate AI assistants that summarize your inbox. Malwarebytes explains what to watch for.
🎨 Free Photoshop Rival. PhotoCraft, an open-source photo editor, quickly drew thousands of contributors, fueling the debate over whether paid desktop software can survive AI.
Stress-test your business idea before you spend a dollar
Friends and family usually say your idea is great. This prompt plays the skeptical investor instead, scoring the idea across ten areas, separating fatal flaws from fixable ones, and naming the three to five tests to run first. It ends with a clear go, test, pivot, or kill call. Use it in Claude or ChatGPT.
Act as a top-tier strategy consultant and startup advisor. Stress-test my business idea as if you were evaluating it for an investment committee.
Business idea:
- Concept: [describe the idea]
- Target customer: [customer segment]
- Problem solved: [problem]
- Solution: [product/service]
- Business model: [how it makes money]
- Market: [country/industry]
- Competitors: [known alternatives]
- Current stage: [idea/MVP/launched]
- Resources available: [budget/team/time]
- Main assumptions: [assumptions]
Analyze the idea across:
1. Customer pain and willingness to pay
2. Market size and growth
3. Competitive advantage and defensibility
4. Revenue model and unit economics
5. Customer acquisition and distribution
6. Operational complexity
7. Regulatory or execution risks
8. Scalability
9. Founder/team fit
10. Biggest assumptions that could make the idea fail
Do not be overly positive. Challenge weak logic, identify hidden dependencies, and distinguish between fatal flaws, fixable weaknesses, and normal startup risks.
Then propose the 3–5 highest-priority tests I should run before investing significant time or money.
Finish with:
- A score out of 10
- Bull case
- Bear case
- Biggest unknown
- What would need to be true for this to become a strong business
- A clear Go / Test Further / Pivot / Kill recommendationTip: Be honest in the "Main assumptions" field, since that is where ideas usually break. A meal-prep service might assume people will pay $15 a meal, a B2B tool that teams will switch from spreadsheets, and a newsletter that readers will pay for premium issues. Read the bear case first: if you cannot answer it, start with the tests.
