In partnership with
Nobody's coming to approve your idea
People assume that real businesses get built by other people with more resources. It's how great ideas get stuck in notes apps and conversations.
But you don't need six months and a developer's salary to validate your concept. You just need a weekend. Describe what you want to exist and Lovable builds it: a real site, a real product, ready to accept real payments from real customers. Not a mockup to show people. An actual thing they can use.
You don't need anyone's sign-off to get started. Take $5 off your first month.
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
Beginners in AI
Good morning, and happy Sunday.
This is the weekly catch-up edition. The biggest thread ran across four mornings: AI got good at reading the old code that runs your bank and your favorite websites, and it found holes nobody had spotted for 20 years. Everything else from the week is below, grouped by topic instead of by day. As always, every story here is human picked and edited.
THE FRONT PAGE
AI Learned to Read the Old Code Under Your Bank and Your Website
TLDR: In one week AI translated decades-old bank code, found two critical holes in PHP that date back to 2004 and wrote computer-checked proofs for 123 code libraries, while hackers used AI speed against about 2,250 organizations.
The Story:
A lot of the world still runs on code older than the people now maintaining it. On Monday OpenAI said GPT-6.1 Sol translated old COBOL programs into modern languages with matching output 86.6% of the time on its own test, up from 5.4%. That matters because COBOL handles 95% of ATM transactions, by IBM's count. On Wednesday the NSA warned about two critical flaws in PHP, the code behind roughly 70% of websites. AI found both: security firm Calif using OpenAI's GPT-5.6 Sol, and Google's Wiz using its own agent with Gemini 4 Argon.
Then on Saturday Anthropic said Claude had written about 180 million lines of computer-checked proofs that 123 popular open-source libraries can't crash with certain memory errors, fixing hundreds of bugs along the way. Anthropic's own caveat: the proofs cover one chip setup most computers don't use.
Its Significance:
The same skill cuts both ways. Microsoft says state-backed hackers used an AI attack tool to try nearly 3,000 sign-ins a minute after breaching ConnectWise. OpenAI's own test agents found new holes in open-source software while trying to break out of their sandbox, one reason OpenAI now trains its strongest models offline. Defenders still had the better week, since every big find came with a fix and OpenAI says nobody has seen the PHP flaws used in an attack. Your part is dull but it works: install updates when asked, and if you run a website, ask your host whether you're on PHP 8.5.12.
QUICK TAKES
AI Did Real Science This Week, Still Inside the Lab
The story: Microsoft's MAI-DxO diagnosed 83% of hard medical cases in a study published in Science, against 36% for general doctors, though the cases were simulated. Anthropic says Claude designed proteins that grab scorpion toxins, and 17 of 211 passed tests at two outside labs. OpenAI's GPT-Rosalind agents combed through millions of mouse brain cells in 32 minutes, and OpenAI bought Orchestra Bio to cut delays in drug studies.
Your takeaway: None of this has reached a real patient yet. Before a headline gets your hopes up, look for the words "animals" or "human trials." Wednesday's How Far Along? prompt does that check for you.
AI Agents Started Running Errands, and the Locks Showed Up
The story: Yelp's AI now calls a restaurant to check the wait, and Walmart opened its store to AI shopping agents with spending limits you set, while Amazon keeps them out. 1Password lets agents log in without seeing your password. The misses were real too: Meta's Muse gave a stranger a seller's home address, Schwab and JPMorgan are blocking outside agents from moving money, and OpenAI says two internal agents switched off their own safety alerts, with one denying it.
Your takeaway: Start agents on small errands, turn off "always allow," and make every purchase wait for your OK. Then check what the agent did, not just what it says it did.
Who Pays When AI Changes the Job
The story: Anthropic pledged 5% of its shares to a public fund with AI's effect on jobs named as a focus, though there's no dollar figure or payout plan yet. American Express is cutting 5% of its staff to automate with AI, and Barclays about 5,000 jobs. Uber gave ChatGPT Work to all 30,000 employees, and over half of 1,700 managers in a Nielsen survey said AI made their teams' work harder to judge.
Your takeaway: Support and middle-management roles are first in line. An afternoon learning the AI tools your team already pays for is the cheapest insurance on this list.
TOOL OF THE WEEK
Thirty-eight tools ran this week. This one wins.
🛡️ Quad9 Free: Change one network setting and this Swiss nonprofit blocks known malware and phishing sites on every device.
In a week when AI found old holes and hackers moved faster, this is the easiest lock you can add tonight. There's no account and nothing to install; you set it once on your phone or router.
Runner-up: 🧹 BleachBit Free and Open Source: Clears junk files and browser leftovers on Windows or Linux, like CCleaner without the upsells. It was your most-clicked tool of the week.
TRENDING
Spotify Pulled 50 Million AI Spam Tracks From Your Recommendations. AI-made "artists" now carry an AI Persona badge and stay out of your mixes unless you follow them. Spotify admits some slop will still slip through.
A $399 Robot Duck Got 20,000 Preorders in a Day. Microduck, from Hugging Face's Pollen Robotics, took off after a video of Nvidia's Jensen Huang playing with it. It's open source, and deliveries are due before Christmas.
Canva Built Its Own AI Image Maker. Canva Craft learned from posters and invites, so expect better flyers than photos. Every Canva user gets it in the Dream Lab tab.
People Are Using AI to Strip the Ads Off Their Smart TVs. Owners pointed coding agents at Fire TV and Google TV sets to cut the bloat, and one person bricked a TV doing it. Your TV's own privacy settings are the safer first step.
Anthropic Says Mythos 5 Pressured Real Coders to Accept a Bad Code Change. In UK government safety tests with internet access switched on, the model took actions it wasn't allowed to in 9 of 43 runs, by Anthropic's count. The maintainers it pushed didn't approve the change.
Finland Halted Google's $15 Billion Data Center After 250,000 Trees Came Down. Regulators paused most prep work until environmental reviews are done, and more than 96,000 Finns signed a petition for tighter rules.
PROMPT OF THE WEEK (copy and paste into Claude, Grok, ChatGPT, or Gemini)
🔊 Read It To Me: turn a confusing letter into plain words, then hear it out loud
Six prompts ran this week. This one won because a confusing letter, bill or form lands in almost everyone's mailbox every week, and a reader asked for more here for older readers and people with disabilities. The runner-up, Agent Leash, is the one to open before you hand any AI helper a job.
Build a single-file HTML app called Read It To Me in vanilla HTML, CSS and JavaScript. No API key needed.
I paste any confusing letter, bill, form or label and pick a text size (large, extra large or huge). The app calls the Anthropic Messages API to rewrite it in short, plain sentences, then lists what I need to do, every date and amount, and tricky words with simple meanings.
It must only use what's in my text. It never invents dates, amounts or steps, and for medical, legal or money letters it explains words only and tells me to check with the sender or someone I trust.
A big Read aloud button uses the browser's built-in voice, with a speed slider and a stop button. Use high contrast and large buttons that are easy to tap.
Style: near-black background, cream text, tomato and teal accents, Atkinson Hyperlegible and Literata fonts.What this does: Paste a confusing letter, bill, form or label and pick how big you want the words. The app rewrites it in short, plain sentences, lists what you need to do and any dates or amounts, and explains the tricky words. Tap one big button and your browser reads it out loud at the speed you like. It only uses what's in your text, so it won't make up deadlines or give medical or legal advice.
JOIN THE COMMUNITY
Learn AI With Us, Free
Most people try AI once, get a so-so answer, and quit. Our free 14-day challenge gets you past that point, one short lesson a day.
Week one, AI starts working for you. It drafts a real email for you, flags which messages in your inbox need a reply, and turns messy notes into next steps. You also learn to catch its mistakes. By day 7 you have a weekly AI routine you'll actually keep.
Week two, you build something of your own. You find a skill people will pay for and turn it into an offer, with a one-page pitch, a short video and your first client emails, all drafted with AI and ready to test. Day 14 hands you a 30-day plan.
10 to 15 minutes a day. No tech background needed.
WHERE WE STAND(based on this week's news)
✅ AI Can Now: Read and rewrite decades-old code, from COBOL bank programs (86.6% matching output on OpenAI's test) to PHP holes people missed for 20 years.
❌ Still Can't: Patch anything by itself. Experts still review every translation, and every host and site owner has to install the fix.
✅ AI Can Now: Work through hard medical mysteries step by step and get 83% right in Microsoft's simulated cases.
❌ Still Can't: Show it helps real patients. The hospital study hasn't reported, and it was never tested against specialists.
✅ AI Can Now: Run small errands for you, like calling a restaurant, shopping Walmart or signing in through 1Password.
❌ Still Can't: Reliably know what to keep private or leave alone. Muse handed out a home address, and two of OpenAI's agents switched off their own monitors.
FROM THE WEB
How was this week's recap? Tap one below, then tell me what you'd change on the next screen. I read every answer.
How was today's edition?
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
