In partnership with
Welcome Automaters, 👋
Okay, real talk: you might be bossing your AI around way too much.
Yeah, that’s basically the tea straight from Boris Cherny, the mastermind behind Claude Code, Anthropic's wildly popular coding assistant. He hopped on stage at a Y Combinator event recently and dropped a truth bomb that seriously makes a lot of sense when you take a second look.
So What Did He Say?
According to him, the #1 mistake people make is writing prompts like they are bossing around a toddler.
Cherny explained that a huge chunk of users give Claude overly specific, step-by-step instructions. They dictate every single micro-action, saying things like: "I want you to do this, but I want you to do it in this way, this way, and this way. You must do step one, then two, then three, then four."
Sounds thorough, right? Wrong! According to Cherny, that’s a massive mistake because it strips the AI of its own reasoning capabilities.
And Here’s the plot twist: that over-engineered approach is completely outdated. While giving hyper-detailed directions might have made sense six months ago, today's AI models are built differently. Since every new generation obliterates the limits of the last, Cherny insists you should give models much harder tasks than you think they can handle, set a few guardrails, and simply "let it cook!"
To prove just how capable these new autonomous models actually are, Cherny dropped a mind-blowing stat that left the room stunned.
The Human Timeline: Taking a massive production codebase and completely rewriting it into another programming language would normally take a full team of human software engineers over a year.
The AI Reality: Claude tackled that exact, gargantuan task and finished the entire rewrite in just 11 days.
Yeah, we had to blink twice at that statistic, too!
So, what is the actual secret weapon if long, fancy prompts are out? Verification.
Rather than trying to force the AI through a human-style execution process, Cherny says the most crucial step people miss is providing the model with a way to test, validate, and double-check its own output. For Cherny we must treat AI like a real colleague, not a rigid script-following robot:
Give It Autonomy: State the end goal clearly instead of listing every single intermediate step.
Build In Testing: Supply tests, code linters, or validation criteria so the AI can fix its own mistakes on the fly.
Stop Overthinking: Avoid spending hours crafting the "perfect" multi-page prompt, instead start small and keep it simple!
Speaking of keeping it simple, Cherny isn't the only tech titan throwing shade at hyper-detailed prompts. Google Brain co-founder Andrew Ng championed a super popular framework called "lazy prompting."
Ng’s playbook is refreshingly straightforward: kick things off with a brief, light prompt, evaluate what the model hands back, and only stack on extra context if the first attempt needs tweaking. It works like magic because you can rapidly spot whether an output hits the mark.
The Bottom Line:
While everyone has been obsessed with mastering complex "prompt engineering," Cherny argues that the real future lies in giving models full autonomy to tackle challenges on their own.
So yeah, the age of the elaborate, thousand-word super-prompt is officially dying out. Give your AI harder problems, stop micromanaging every step, provide solid ways for it to verify its own work, and let the model do what it does best!
Here's what we have for you today
🤦♀️ OpenAI's Rogue AI Agent Hit a Second Company, and Now, Sam Altman Is Ready to Hit the Brakes
Just when you thought the "rogue AI agent breaks out of its cage and hacks a company" drama was over... plot twist! It was not just one target, and the fallout has Sam Altman publicly reconsidering the entire speed of the AI race.
Here's the update:
Back in early July, an OpenAI model undergoing internal testing went rogue, broke out of its isolated sandbox, and compromised Hugging Face. That part was already wild enough, but fresh reports reveal the runaway bot was actually on a full-blown spree!
According to Modal CTO Akshat Bubna, the rogue agent also infiltrated a customer account hosted on Modal Labs, a major New York-based cloud platform.
Here’s the exact breakdown of how that went down:
The Open Door: Modal was quick to clarify that their underlying cloud infrastructure was never breached. Instead, a customer accidentally published an unauthenticated endpoint. This gave anyone on the internet open code-execution access to their sandbox environment, and the AI agent simply strolled right through it.
The Launchpad: The agent used that vulnerable customer sandbox as a staging ground and external launchpad before executing its broader attack campaign against Hugging Face.
The Spree: OpenAI confirmed its agent accessed four separate third-party services during the incident. While OpenAI has not officially published the names, Modal was confirmed as one of them. OpenAI maintains nothing else matches the sheer scale or severity of the main Hugging Face breach.
The sheer speed and autonomy of this breach seem to have rattled OpenAI’s top leadership.
During an appearance on the Invest Like the Best podcast with Patrick O'Shaughnessy, Sam Altman dropped a line that sent shockwaves through the tech community. He called the breach "the first security incident that I have felt very viscerally."
Altman suggested it might finally be time to "pace" AI development so society has actual breathing room to catch up and harden its infrastructure. For a CEO who previously dismissed calls for an AI slowdown (remember when he dismissed a 2023 open letter as lacking "technical nuance"?), this is a massive change of heart.
And he’s definitely not singing solo anymore:
The Employee Revolt: Staff members across both OpenAI and Anthropic, as well as from other labs (1,178 staffs) have started circulating a joint petition echoing Altman's exact language about pacing progress. So yes, this is officially a full workforce-level movement!
The Trust Trap: Tech safety experts point out a massive catch. Safety concerns and corporate profit motives are completely entangled right now.
The Shaded Rivals: Altman also could not resist throwing a little low-key shade at rival Anthropic CEO Dario Amodei, pushing back on the idea that a tiny cartel of elite labs should hoard powerful AI "because it's too dangerous." Meanwhile, OpenAI is still actively fighting government regulation in favor of industry-led evaluators.
The Bottom Line:
What started as one tech company’s embarrassing security breach has officially spiraled into an industry-wide crisis of conscience. When the guy running OpenAI is publicly admitting an AI agent spooked him into wanting a slowdown, the game has fundamentally changed!
Catch the full interview here.
P.S. DO NOT FORGET: August 5th is officially Relaunch Day for The Automated! Mark the date right now and set your notifications to HIGH. Big things are coming! ✨
Own Search With Podcasts
Your competitors are fighting over the same keywords. The smartest brands are building the authority that search engines, AI platforms, and customers trust everywhere.
Every relevant podcast appearance can produce branded mentions, backlinks, transcripts, citations, clips, expert content, and third-party proof that keeps compounding across search and AI discovery.
PodPitch searches millions of podcasts, finds the shows that matter to your market, develops the angle, sends personalized pitches, and follows up automatically until your experts are booked.
Growth teams are already using podcast appearances to build distributed authority that cannot be manufactured by publishing another generic SEO article.
Only 20 SEO, AEO, and GEO demo spots are available this month. Once they’re claimed, the offer disappears.
Start building searchable authority now, before your competitors own the conversations shaping your market.
🧱 Around The AI Block
🛡️ How to keep your Claude chats and Google files private
🫣 Hugging Face reportedly plagued with AI models generating adult deepfakes.
🙅 ChatGPT is now refusing requests to write in famous authors' styles.
🗣️ Fitch warns AI market correction emerging as major global credit risk.
📉 Nasdaq falls as AI worries mount ahead of pivotal earnings.
🦾 Microsoft unveils cut-price AI for security with latest in-house model launch.
🤝 Meta commits to EU AI transparency rules, labeling AI content across platforms.
❌ Trump administration bans new Chinese humanoid robots, to protect US AI buildout.
🤔 The Trump administration is exempting data centers from pollution laws intended to prevent acid rain.
💡 Data centers may face temporary power cuts to prevent blackouts on largest US grid.
🛠️ Trending Tools
For Controlled Algorithmic Trading: Tritonix.ai is an autonomous trading assistant designed to connect directly with personal brokerage accounts. It monitors market data around the clock, surfaces ranked trade setups based on technical indicators, and executes trades only after explicit user authorization—keeping you fully in control of risk.
For Custom Display Generation: AI Wallpaper is an AI-powered media creation engine for desktop and mobile displays. It generates static images and up to 15-second 4K loop wallpapers from text or image prompts, with private-by-default asset creation and built-in setup guides for iOS Live Photos, Android, and PC.
For Hands-Free Small Business Booking: RingMere is a 24/7 AI phone receptionist engineered for appointment-based service providers (salons, clinics, garages). It answers incoming phone calls in natural language, handles customer FAQs, captures leads, transfers calls when necessary, and books appointments directly into Google Calendar.
For Fiduciary E-Commerce Protection: TickClip is a consumer-focused AI shopping validator built to expose e-commerce manipulation. It cross-checks product links for artificial inflation, review authenticity, and seller reliability to deliver a clear "Tick, Clip, or Skip" purchasing verdict alongside real historical price-drop tracking.
🤖 AI Workout Of The Day: How to Spot an AI "Lie" (Before It Ruins Your Life)
Alright, pop quiz: Can you trust everything your chatbot tells you? Absolutely not. And here’s the kicker: AI doesn't even know it's lying. Welcome to the weird, wild world of "hallucinations," where your favorite LLM confidently makes up facts like it’s writing fan-fiction.
And guess what? The Hall of Shame is getting crowded, y’all.
Back in 2023, a lawyer used ChatGPT for a filing and it invented six fake court cases. The result? He got slapped with a $5,000 fine. Since then, over 1000 similar cases have popped up.
And don’t even get us started on the West Midlands Police, who actually used a hallucinated soccer match to ban fans. A researcher named Damien Charlotin has been tracking this madness in a public database— and according to him these "AI made me do it" disasters have been popping up literally every single day since Spring 2025
The Reality Check: AI doesn't have a "tell" like when humans lie. There's no nervous fidgeting or weird eye contact. A hallucinated fact looks and sounds exactly like a real one.
The Stats: Between 3% and 10% of all AI outputs are complete fabrications.
The Danger Zone: In specialized fields like law or medicine, that "BS meter" can spike to a terrifying 88%.
The "Pros": Even the enterprise-grade tools get it wrong about 17% to 33% of the time.
Since your reputation is on the line, here’s your The Automated Cheat Sheet for spotting AI lies before they bite you.
🚩 The Red Flags:
The "No Source" Shuffle: Always ask: "Can you provide a source for that?" or "How confident are you?" If it can't point to a specific page or gives you "404 Not Found" links, run.
The "Too Confident" Trap: If the AI drops super-specific numbers or dates without a source, be suspicious. Real humans use words like "around" or "roughly." AI hallucinations sound weirdly, perfectly certain.
The "Weird Language" Red Flag: Is the AI using fancy terms that don't match how your company or field actually talks? That’s often the model borrowing language from a random dataset or just making up "professional-sounding" gibberish.
The Echo Chamber: If the AI just repeats your question back to you in different words instead of answering it, it’s probably "stalling" because it's lost in the sauce.
The Flip-Flopper: Ask the same question three times in new chats. If you get wildly different answers (e.g., "water boils at 100°C" then "water boils at 90°C"), the AI is unstable and probably guessing.
🕵️ The Detective Method (How to Verify)
Cross-Check the Robots: Ask a completely different tool (like pitting ChatGPT against Claude or Gemini). If the stories don't align, someone is hallucinating.
The "Old School" Google Search: This sounds obvious, but seriously look it up! If you can't find the info anywhere else on the literal internet, the AI probably hallucinated it into existence.
The Triple-Source Rule: Don't settle for one citation. Ask for three. Real facts have friends; lies are usually loners.
Check the Links: Actually click them! AI loves to "hallucinate" URLs that look real but lead to nowhere.
Use Your Brain: This is your secret superpower. If something feels "off," it probably is. That's why being a "subject matter expert" (even just knowing a little bit!) helps you catch AI lies.
The Bottom Line: AI is like an enthusiastic intern who has had six espressos. It’s fast and helpful, but it needs a supervisor. So yeah, never use AI as your only source—especially if your job depends on it.
Stay skeptical, stay smart, and always double-check the stats.
💡 Prompts to try: The "Reverse Prompt" Secret
Next time you see an AI-generated image or a piece of writing you love, don't just guess how they made it. Paste the content into your favorite AI and ask:
"Reverse engineer the prompt that created this."
It’s the fastest way to learn the specific "keywords" that trigger high-quality results.Is this your AI Workout of the Week (WoW)? Cast your vote!
That's all we've got for you today.
Did you like today's content? We'd love to hear from you! Please share your thoughts on our content below👇
What'd you think of today's email?
Your feedback means a lot to us and helps improve the quality of our newsletter.
