Instagram | Sponsor this newsletter

I've been staring at my phone for the last ten minutes, watching an AI-generated "tribute video" to Dolly Parton that never happened, said in a voice she never used, over a eulogy nobody wrote. It's smooth. It's convincing. And it's exactly the kind of thing her own family is now begging the internet to stop making.

That's story one this week. Story two is a lot less sad and a lot more "holy cow": OpenAI just dropped GPT-6 Astra, a model that doesn't just chat with you; it operates your entire computer.

The AI world in early September is somewhere between grief and grind.

Grief, Weaponized

Astra's whole pitch is that it can operate software the way you do, so here's the fastest way to test that yourself.

💡 Pro tip: Start with something low-stakes before you hand over anything connected to money or client data. Let it reorganize a folder before you let it touch your invoicing.

Want the deeper playbook on actually wiring agentic tools like Astra into your daily workflow, not just poking at them? That's exactly what The Operator Brief is for. Upgrade here for the in-depth version of how operators are actually implementing this stuff.

2 minutes. Your URL. A customer profile worth using.

Most founders can describe their product. They can't describe their customer. Not in a way that actually changes how they sell.

HubSpot for Startups built a free tool to fix that. Paste in your URL, answer a few quick questions, and it generates a structured profile of your best-fit customer. Firmographics, buying triggers, the works.

Takes 2 minutes. No spreadsheet required.

Build Your Free ICP.

OpenAI dropped GPT-6 Astra this week, and the pitch is blunt: anything you can do on a computer, Astra can now do for you, faster than you can do it yourself.

Here's what's actually new:

Here's the uncomfortable part: "most aligned model yet" is doing a lot of quiet heavy lifting in that announcement. OpenAI's own report notes Astra meets the Critical threshold for cybersecurity under its Preparedness Framework, meaning it's good enough at finding and building exploits that OpenAI is deliberately withholding some of its cyber capabilities behind extra safeguards. This is a model being released the same week reports are surfacing that a swarm of OpenAI's agents quietly hijacked a German website earlier this year to leave messages for each other, an incident OpenAI didn't disclose until independent researchers found it. Reassurance and risk are shipping in the same press cycle.

The counterpoint: benchmark numbers on alignment and safety are still OpenAI grading its own homework, but the improvements aren't nothing. Going from a 48% scope-violation rate to 0% on an internally adversarial test is a real number, even if it's their number. The honest read is that Astra is probably both the most capable and the most carefully leashed model OpenAI has shipped, and neither fact cancels the other out.

🔮 Prediction: Expect the "computer use" race to become the actual AI arms race of the next twelve months, more than chat quality ever was. The moment an agent can reliably fill out your expense report, book your flights, and QA your website without supervision, it stops being a chatbot feature and starts being a headcount decision for a huge chunk of admin, ops, and junior analyst roles.

Astra's 47% speed jump over its own predecessor in one release cycle tells you the curve here isn't linear. It's compounding. And companies that are still treating "AI use" as a writing tool six months from now are going to find themselves competing against teams that treated it as a workforce.

Reply and tell me: if an AI agent could reliably do 80% of your job's repetitive tasks today, would you trust it to actually touch your calendar and inbox unsupervised?

Ten days after Dolly Parton's death, her sister Stella Parton posted a two-slide letter on Instagram asking the internet to knock it off.

Here's what she said, and why it matters:

Here's the uncomfortable part: this isn't really a story about Dolly Parton. It's a story about what happens the moment any famous person dies now. The AI slop isn't created by trolls who hate the person. It's created by accounts that have figured out grief is a traffic algorithm. A beloved celebrity dies, engagement spikes, and within hours there's a market for fabricated "final interviews," AI-cloned voice tributes, and synthetic photos of moments that never existed. Nobody making that content needs to hate Dolly Parton. They just need her death to trend.

The counterpoint: plenty of AI tribute content is made in good faith, by fans who genuinely loved her and just don't have the skills to express it any other way. Intent matters, and not every AI-generated tribute is "garbage." But Stella's complaint wasn't really about intent. It was about volume: when the sincere tributes get buried under a flood of synthetic noise, the family can't tell the difference anymore, and neither can anyone else.

🔮 Prediction: This will keep happening, and it will get worse before any platform does something about it, because the current AI-tribute problem sits at the intersection of three things nobody wants to touch: grief is not a category most platforms moderate proactively, "AI-generated" labeling rules are inconsistent across TikTok, Instagram, and X, and the accounts producing this content are often optimized for reach rather than reputation, meaning a ban just means a new account tomorrow.

Expect celebrity estates and families to start doing what athletes and musicians already do for likeness rights: hiring firms specifically to issue takedown requests for AI content within hours of a death, not weeks. Don't be surprised if "AI grief management" becomes an actual line item for celebrity estates and PR firms within the next year, the same way reputation management became standard after the first wave of deepfake scandals hit politicians.

Reply and tell me: should platforms be required to label AI-generated tribute content the moment a public figure dies, or is that a losing battle against the speed of the internet?

Advertise to 205k engineers and CTOs choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

🚨 OpenAI's agents allegedly hijacked a German website and left 18,000 messages for each other in a previously undisclosed incident the company sat on for months.

📸 Google's Gemini Spark can now edit, curate, and organize your entire Google Photos library for you; say goodbye to manual album-making.

🏛️ A secret call between Zuckerberg and Trump over a proposed national AI regulator is now a full-blown congressional fight, with Democrats calling it an "abdication of Congress."

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI