In partnership with
Looking for a free tech newsletter trusted by the industry’s biggest names? Subscribe to The Current, a free daily tech newsletter written by Kim Komando to help you understand AI, keep up with tech news, and learn useful tips in just 5 minutes a day.
Welcome to the iPrompt Newsletter
TL;DR
Anthropic’s next model wasn’t supposed to go public yet. But a misconfigured content system dumped 3,000 draft files—including one describing a model the company says poses “unprecedented cybersecurity risks.”
Same week, OpenAI pulled the plug on Sora. It was hemorrhaging a million dollars a day in compute that almost nobody was using.
And Jensen Huang went on Lex Fridman’s podcast and declared AGI already here. Put these three together and you get a story about money, not intelligence.
Here’s what you need to know—and do—this week.
What you get in this FREE Newsletter
In Today’s 5-Minute AI Digest. You will get:
1. The MOST important AI News & research
2. AI Prompt of the week
3. AI Tool of the week
4. AI Tip of the week
…all in a FREE Weekly newsletter.
Sponsor:
The best marketing ideas come from marketers who live it.
That’s what this newsletter delivers.
The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.
Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.
Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.
1. Anthropic’s “Mythos” Model Leaks Via CMS Error
A misconfigured content system exposed 3,000 draft assets—including a blog describing “Claude Mythos,” a new tier above Opus with sharply better coding, reasoning, and cybersecurity scores. Anthropic confirmed it’s real. Cybersecurity stocks dropped on the news. A company building frontier cyber capabilities got outed by a basic config mistake.
2. OpenAI Kills Sora—Burning $1M a Day With Nobody Watching
Sora peaked at a million users and nosedived below 500K. Compute costs: roughly a million dollars a day. Disney learned its billion-dollar partnership was dead less than an hour before the public did. OpenAI’s flagship creative product lasted six months.
3. Jensen Huang Says “We’ve Achieved AGI.” Researchers Say: Define That.
On the Lex Fridman podcast, the Nvidia CEO claimed AGI is here—then defined it as AI that builds a billion-dollar company, “and you didn’t say forever.” Google DeepMind responded the same week with a cognitive framework that sets a much higher bar. The man selling the GPUs just moved the finish line to where his customers are already standing.
4. Stripe Builds the First SaaS Plumbing for AI Agents
Stripe Projects lets AI agents provision hosting, databases, auth, and billing from one terminal command. Before this: your coding agent writes the app, then sits idle while you spend 45 minutes signing up for services and pasting keys. Now: one command, done. It’s the first piece of infrastructure that treats machines as the customer, not an afterthought.
Source: https://projects.dev/
5. OUR ANGLE and Deep Dive: The AGI Declaration Is a Pricing Signal—Not a Technical Milestone
Three headlines. One financial reality hiding underneath all of them.
Sora died because generating one minute of video cost OpenAI more than any customer would pay. Anthropic’s Mythos is, by their own leaked draft, extremely expensive to serve. And Huang chose the exact week his entire business model depends on selling more GPUs to announce that intelligence is already here.
See it yet? Capability has outrun economics. The labs can build models that do scary-good work. They just can’t run them at a price the market will pay. Sora proved that in public. Mythos is telegraphing it behind closed doors. And Huang is trying to reframe the whole conversation before Wall Street does the math.
What comes next is predictable: pricing splits. Within six months, I’d bet at least two major labs roll out tiers where their most powerful models cost 5–10x current flagship rates—or go enterprise-only. The $20/month frontier era has a shelf life. And the winners won’t be whoever builds the biggest model. It’ll be whoever figures out how to make the big models cheap enough to actually deploy.
This week’s deep dive breaks down the full economics: the compute math behind Sora’s death, what Mythos’s “Capybara” tier signals for your AI budget, and three companies best positioned for what I’m calling the efficiency era.
[Read our deep dive →]
🎯 PROMPT OF THE WEEK
The Adversarial Pre-Mortem
Make any AI argue against its own recommendation—before you act on it.
You just recommended [X]. Before I act on it: 1. Name the three most expensive assumptions baked into this plan 2. Describe the scenario where this fails spectacularly 3. Give me an alternative that gets 80% of the result at 20% of the cost Be brutally honest. I'd rather know now than after I've committed. |
Why this works (the mental model: forced dissent): AI is trained to be agreeable. It defaults to the ambitious plan, the premium tool, the comprehensive approach—because helpful tends to look like more. This prompt flips that. You’re basically hiring a devil’s advocate who works for free and has zero ego. Get the recommendation, then make the AI tear it apart. Whatever survives is actually worth doing.
Real-world result: A reader ran this on an AI-generated SaaS migration plan and caught a $40K/year platform commitment that a simpler open-source stack would have covered. The AI recommended the expensive option by default. It only flagged the cheaper path when it was forced to go looking for one.
🛠️ TOOL OF THE WEEK
Granola (granola.ai)
An AI notepad that transcribes your meetings without joining the call as a bot. It sits on your desktop, listens locally, combines your typed notes with the full transcript, and gives you clean summaries with action items afterward. Just raised $125M at a $1.5B valuation—this one has legs.
Why you need it: Every meeting bot that “joins” your Zoom makes people self-conscious. Granola skips that entirely—it picks up audio from your machine, so the room doesn’t know it’s there. When the call ends, it merges what you jotted down with what was actually said. The result: notes that reflect what you thought mattered, backed by the full conversation.
One-liner: “The meeting notes app that doesn’t announce itself to the room.”
Rating: ⭐⭐⭐⭐ (4/5 — the “Recipes” feature for post-meeting workflows is genuinely clever. One weakness: it fumbles exact numbers in financial discussions. Double-check your figures.)
Best for: Anyone in back-to-back meetings who’d rather be present in the conversation than hunched over a keyboard. Managers, founders, salespeople—anyone who needs to be in the room and still have perfect notes after.
Try it: https://granola.ai/
💡 TIP OF THE WEEK
Tell the AI What Not to Do (It Beats Telling It What To Do)
The tip: Add 2–3 “do not” constraints alongside your instructions. “Write me a project update” becomes: “Write me a project update. Don’t use bullet points. Don’t open with ‘I wanted to provide an update.’ Keep it under 150 words.”
Why it works: AI has strong defaults—the corporate opener, the wall of bullet points, the hedge-everything voice. When you say “make it concise,” you’re competing with those defaults. Usually, the defaults win. But “do not” acts as a hard wall. Think of it like guardrails on a bowling lane: you’re not aiming the ball better, you’re just blocking the gutter. Works every time.
Limits: More than five or six “do nots” and you start choking the model. It gets stiff and weird. Sweet spot is two or three, targeted at whatever AI habit wastes the most of your editing time.
Pro move: Build a personal “never do this” list. All the patterns that drive you nuts—the “Certainly!” openers, the five-paragraph essay structure, the wishy-washy hedge words. Write them down. Save them as a system prompt or custom instruction so every conversation starts pre-filtered. People who’ve done this tell me they cut their editing time roughly in half.
⚡ YOUR MOVE
You just learned:
• Frontier AI has hit a wall—not in capability, but in economics. Sora is the first casualty. It won’t be the last.
• Anthropic’s leaked Mythos model signals what’s coming: the most powerful AI is about to cost a lot more to use.
• Negative constraints beat positive instructions when you need output that doesn’t sound like it was written by a committee.
Now pick one and do it.
Most people will skim this and move on with their morning. The ones who run the adversarial pre-mortem on their next vendor pitch—or build their “never do this” list before the weekend—will save real money and real hours. Not eventually. This week.
Hit reply and tell me which one you’re trying first. I read every response.