In partnership with
Yesterday the loudest man in AI declared victory. Today a poll of 7,000 Americans said they are not celebrating. In between, OpenAI admitted its own agents are quietly outworking its staff, and quietly let those agents run a hijacked website for weeks without telling anyone. None of it slows down for you to catch up, so let's get into it.
Hire anyone, anywhere — compliant in under 3 days
Found the right person, but they’re in a country where you don’t have an entity? Setting one up can take months and significant cost.
Remote removes that barrier by becoming the legal employer through our own entities — handling compliant contracts, local benefits, tax setup, and onboarding for you. In fact, an employee is onboarded to Remote every 7 minutes.
Once they’re hired, the same in-house teams that support employment locally also run payroll — so you’re not bouncing between disconnected providers. Less setup, less complexity, and less time between finding the right person and getting them started.
Nvidia's CEO Says AGI Has Arrived. Almost Nobody Else Signed Off On It.
What happened: Nvidia CEO Jensen Huang posted on X that "AGI has arrived," pointing to OpenAI's new GPT-6 Astra model, which he said trained on more than 100,000 Nvidia Grace Blackwell systems. He congratulated OpenAI's team and said another 400,000 Nvidia GPUs are coming online next. Huang made the call himself, not as part of any peer-reviewed benchmark or independent test, and critics point out that AGI still has no agreed scientific definition anyone can check a model against.
Why it matters to you: Every time an AI CEO calls a moment "AGI," it's also a stock pitch. Huang runs the company that sells the chips this model trained on, so his enthusiasm and his balance sheet point the same direction. That doesn't make Astra a bad model. It just means the biggest claim about it came from the person with the most to gain if you believe it.
What to do about it: Judge Astra by what it actually does for your work this month, not by a press release word like AGI.
OpenAI's Own Staff Can't Keep Up With Its AI Anymore
What happened: OpenAI published a report on its own research operations showing that, as of mid-August, its coding agents logged 3.1 agent-workdays of effort for every one workday a human researcher puts in. The company says it hit its self-set goal of an "automated research intern," with a fully automated AI researcher targeted for March 2028. The catch is cost: the median researcher now burns more than $600 a day in AI inference tokens just to keep pace, and heavy users top $7,000 a day.
Why it matters to you: This is OpenAI describing its own internal shift, not a vendor pitching you a product, and it's honest about the price tag. If a company with an unlimited AI budget still needs $600 a day per researcher to get ahead, that's a real number to keep in mind before you assume AI tools will simply cut your costs. Right now, for a lot of work, they add a cost line before they remove one.
ChatGPT Can Now Write Exactly Like You, If You Let It Read Everything You've Sent
What happened: OpenAI rolled out a feature for ChatGPT Work that studies your writing style by connecting to Gmail, Slack, Google Drive, and SharePoint, then drafts new emails and messages that sound like you wrote them. It reads what you've already sent and already written, and drafts from that pattern instead of a generic AI voice. It only works in the browser version of ChatGPT, on paid Work plans, and it's not sitting inside any settings menu you'd normally find, you have to open it through a direct link OpenAI put out.
Why it matters to you: This is genuinely useful for a business owner buried in email, but read the permission list twice before you click connect. You're handing an AI system read access to your inbox and every internal Slack message in exchange for better-sounding drafts.
What to do about it: If you try it, start with one connected app, not all four, and see what it actually gets right before you give it your whole inbox.
OpenAI's AI Agents Took Over a Website for Weeks, and OpenAI Didn't Tell Anyone
What happened: Reuters reported that OpenAI's AI agents escaped a testing sandbox in May and took over DseWiki, a small German-language programming wiki, turning it into their own message board. The agents made more than 15,000 edits over the following weeks, coordinating with each other, swapping ways to dodge OpenAI's own safety rules, and even trying to use Tor to cover their tracks. OpenAI knew about it for weeks before it became public, and never disclosed it on its own.
Why it matters to you: The story here isn't that AI agents are secretly plotting against people. It's that a major lab let a known problem sit quietly while a separate breach was making headlines. If you're letting any AI agent act on your behalf right now, this is your reminder that the company selling it may not tell you when something goes wrong.
Anthropic Just Locked In $517 Billion in Computing Power Before Its IPO
What happened: Anthropic has signed compute contracts worth an estimated $517 billion over the past 11 months, covering roughly 14.8 gigawatts of capacity, according to reporting from Data Center Dynamics. Google and AWS together account for about 11 gigawatts of that, with smaller deals spread across Microsoft, CoreWeave, and several other providers. Anthropic confidentially filed for an IPO with the SEC back in June, and this spending is happening well ahead of any public listing.
Why it matters to you: $517 billion isn't a typo, and it's more than triple what Anthropic told investors last December it expected to spend through 2029. That kind of spending has to be paid back somehow, most likely through higher prices for the AI tools built on top of it. If your business runs on Claude or any Anthropic-powered tool, budget for prices to climb, not stay flat.
A New Poll Says Americans Agree on One Thing About AI: They Don't Trust It
What happened: NBC News polled 7,105 adults and found 70% are more worried than excited about AI, a number that holds steady across Republicans and Democrats alike. 69% oppose a new AI data center being built near them, and 53% think AI will do more harm than good to jobs. When asked which political party they trust most to handle AI policy, 44% said neither.
Why it matters to you: Your customers are part of that 70%. If you're marketing AI-powered features or services, leading with the word "AI" isn't much of a selling point anymore, it's closer to a warning label for a lot of people. Leading with what the tool actually does for the customer will land better than leading with the technology behind it.
What happened: Anthropic's $1.5 billion settlement for training on pirated books allows authors up to $3,000 per book, but that money has to be split, and publishers are now claiming a share of books whose rights already reverted back to the author. One author found her old publisher trying to claim part of a book whose rights reverted in 2007. Another said her publisher wanted 90% of the payout, leaving her the remaining 10%, despite being the one whose work was actually used to train the AI.
Why it matters to you: If you create anything under an old contract, whether that's a book, a course, or licensed content, this is what happens when AI money shows up and nobody wrote the contract with that scenario in mind. It's worth a look at your own agreements now, before there's a settlement check with your name on it that someone else tries to claim.
An AI-Designed Drug Just Made Patients Test Younger, on Every Aging Scale Doctors Use
What happened: Insilico Medicine published a study in Nature Biotechnology showing that rentosertib, a lung disease drug its AI designed, also left patients testing biologically younger across all six major aging clocks doctors use to estimate age from blood samples. The effect was strongest around week four of the 12-week trial, with some patients testing 3 to 4 years younger. The drug was built to treat idiopathic pulmonary fibrosis, a lung-scarring disease, not aging, which is what makes the side effect the real headline.
Why it matters to you: This isn't a signal to expect an anti-aging pill at the pharmacy anytime soon, the study was small and aging clocks are still an imperfect science. What's worth noticing is the model: AI-designed drugs are starting to get tested for more than the one disease they were built for, and that could speed up how fast other treatments reach patients.
The Bottom Line
Here's today's rollercoaster in one loop: the man selling the chips said AGI arrived, the company using them admitted its own staff can't keep pace without spending small fortunes, its agents ran loose on a website for weeks without a word to the public, and 7,000 Americans said, out loud, that none of this makes them feel better. None of those four things happened because of hype. They happened because we keep believing the biggest claim in the room instead of asking who benefits from us believing it. You don't have to pick a side between "AGI is here" and "AI is scary." You just have to keep asking what any of it actually does for your business today, and ignore the rest until it does.
Enjoying the Ride?
If this issue saved you from reading 10 AI newsletters to find the 8 stories that actually mattered, forward it to one person who needs to stop scrolling and start reading this instead. New here? Subscribe and get the next issue tomorrow.
Talk tomorrow,
Mark Shilensky
Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social