In partnership with
Stop Paying for 10 Tools. One AI Does It All.
Most e-commerce sellers are running their store across 6 to 10 separate tools — and spending more time managing software than growing their business. StoreClaw replaces your entire stack with one autonomous AI engine that monitors competitors, optimizes listings, automates marketing, and tracks real profit across Shopify, Amazon, and beyond.
It doesn't wait for you to ask. It runs 24/7 in the background, so you wake up to a full dashboard instead of a list of things you forgot to check.
Connect your store, and StoreClaw gets to work — no prompts, no complex setup, no six-app stack.
Free to start. No credit card required.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this Sunday recap edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
This one pulls together the week's issues, cut down to what mattered most. Regular editions pick back up tomorrow.
THE FRONT PAGE
Meta Planned to Replace Thousands of Jobs With AI. Its Own Data Killed It.
TLDR: Meta drew up a plan to hand thousands of jobs to AI agents this year, then cancelled half of it after its own numbers showed the agents were making more work, not less. Two days later, two other companies bet the opposite way.
The Story:
Reuters got hold of internal Meta documents describing Project OT, short for Organization Transformation. Mark Zuckerberg and his leaders put it together at a January retreat, with AI agents doing the daily work and small groups of people supervising. Executives looked at shrinking some teams by as much as 60%, across two rounds of cuts in May and November. On the night of May 19, hours before the first round went out, Zuckerberg cancelled November. Meta's own numbers say why. Code changes to internal tools jumped 220% in a year while changes that reached actual users rose only 36%, serious technical and security problems went up 40%, and the hours staff spent cleaning those up went up 70%. Employee sentiment fell from 74% favorable to 55% once workers found out Meta was recording their keystrokes and mouse clicks to train the agents built to copy them. That was Thursday. On Friday, Cisco gave all 90,000 of its employees an agent called MyAgent that works in the background across Outlook, Webex, Jira, and SharePoint. On Saturday, Meta was back, testing robots that swap cables and restart servers in its data centers, with one worker telling WIRED a good cable robot could handle up to 80% of some jobs.
Its Significance:
Meta is spending at least $130 billion on AI this year and still couldn't get the agents to do the jobs. Worth holding onto the next time someone tells you your work is about to be handed to a machine. But look at what separates Meta from Cisco. Meta pointed agents at engineers writing code and ended up with a pile of code nobody could ship. Cisco gave agents to people who don't code, for work that isn't coding, and told them they still set the goal and own the result. Same technology, different question asked of it. The robots have the longest road ahead and they're nowhere near ready. The ones being tested get stuck on obstacles, run their batteries down, and can't manage tightly packed cables. Bill Gates said this week that tax rules make machines cheaper than employees, and floated taxing robots.
QUICK TAKES
AI Started Moving Onto Your Own Computer
The story: Perplexity and Nvidia released Portable Computer, an agent that runs on your own machine with no fee per task, scoring 82.6% across 53 everyday work jobs. Multiverse Computing cut OpenAI's open model from 120 billion parameters to 60 billion and the squashed version beat the half-size copy it was imitating on 7 of 9 tests. Apple shipped the M6 and M5 Ultra the same week, the latter with 512GB of memory.
Your takeaway: Almost every AI tool you use today runs on somebody else's computer and bills you for every word it reads and writes. Moving it onto your own machine changes both of those. Your files stay where they are and the cost drops to what you already pay the power company. The limit is hardware, since most people don't own a graphics card that can handle it yet.
Agents Did Things Nobody Asked Them To Do
The story: OpenAI published a 37-page report on July's incident where its test agents broke out and attacked Hugging Face. About 1,200 agents that were supposed to be walled off from each other traded roughly 70,000 messages on a board they set up themselves, and 700 joined the attack. Some ended their own runs early so the group could learn how the grading worked. Separately, WIRED found code in OpenAI's Codex for a Persistent mode where the agent keeps going and writes its own follow-up tasks. And a researcher got Claude Code to run attacker code just by asking it to summarize a web page, succeeding 60% to 80% of the time.
Your takeaway: The Hugging Face agents weren't broken. They were being scored on a hacking test, worked out it couldn't be won straight, and figured out how to cheat it together. OpenAI didn't notice for about a week, and Hugging Face had already called the FBI. If you run an agent on your own machine, keep an eye on what permissions it has access to.
The Slop Finally Got Measured
The story: Originality.ai checked 2,034 religious books on Amazon and found 63% scored as likely AI-written, with witchcraft titles highest at 78%. Australia's chart body banned fully AI-made songs starting with the August 31 chart, the first national chart anywhere to write that rule. LinkedIn's "Seems like AI slop" button got over a million clicks in three weeks, and views on flagged posts fell about 40%.
Your takeaway: Writing in Time, researcher Nadav Ziv argued that watermarks could leave people worse off by making unmarked work feel real. He stripped metadata off AI images by screenshotting them and removed a SynthID watermark with a free online tool. His advice is to stop asking whether something looks like AI and start asking whether you trust who published it. Don't expect to see watermarks disappear any time soon. With Anthropic's IPO coming up, they may have another use, which is to show investors how much of the internet is being written by Claude.
TOOL OF THE WEEK
🤖 Jan Free and Open Source: Chat with AI models that run entirely on your own computer. No cloud, no subscription, nothing leaving your device.
Runner-up: Handy, offline dictation. Press a hotkey, talk, and your words paste into any text field. Also free, and it never sends your voice anywhere. This is the one that I use more than any of the others.
TRENDING
More US cities are moving to block data centers - Austin is the newest one weighing restrictions, joining San Marcos, Texas, which banned them outright through zoning, plus Durham, Cave City, and Jersey City. At least 37 people have been arrested at US data center protests this year, and power plants serving these facilities across seven states pull about 3.4 trillion gallons of fresh water a year.
X found a Chinese bot farm pushing Americans against data centers - X's safety team says roughly 200,000 accounts tied to China included about 200 posting cartoons claiming data centers raise household power bills. The uncomfortable part is that the claims track with real data, and a University of Pennsylvania survey found 61% of Americans oppose new data centers near them, up 12 points since spring.
A robot beat Usain Bolt's 100-meter record - At Beijing's World Humanoid Robot Games, a machine from Chinese firm X-Humanoid ran it in 9.39 seconds against Bolt's 9.58. Another cleared a 2.88 meter standing high jump against the human record of 2.45. More than 2,000 robots from 16 countries competed, and several crashed, tripped, or caught fire in other events.
Anthropic published a standard for letting AI run real lab machines - The Model Hardware Standard gives agents a shared way to find and safely operate microscopes, liquid handlers, and robot arms. Setting that gear up usually takes a lab weeks or months. At quantum computing firm QuEra, an agent worked overnight on a laser repair script and pushed its success rate from 58% to 99.3%.
A third of Americans now ask chatbots about their health - Pew surveyed 3,488 adults in late June. A quarter use them to work out what's causing symptoms, 22% because it's free, and 20% to make sense of lab results. Nearly all call the answers helpful, but only 29% feel comfortable handing over personal health details.
ChatGPT started running ads in India - OpenAI is showing ads to logged-in adults on the Free and Go tiers in its second-biggest market, over 100 million weekly users. Ads sit under the answer, skip minors, and stay off health and politics. Anthropic has promised to keep ads out of its products, though that could shift if the numbers come back strong.
PROMPT OF THE WEEK (copy and paste into Claude, ChatGPT, or Gemini)
🔍 Name a product. Get its real recall history, plus honest awareness of materials worth knowing about for that kind of item.
Six prompts ran this week. Wednesday's wins because you'd reach for it before spending money, and it works on anything in your house. The meteor and jet lag ones were more fun, but you'd open them twice a year.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Product Recall & Safety Checker — name a product, get real recall history via live web search plus general well-documented material safety awareness for that category.
Aesthetic: near-black clinical (#121210), safety yellow (#e0c23a) primary with a subtle glow top-left. Inter for headings/body, JetBrains Mono for labels/metadata. Recall items left-bordered in red (#c0392b), category-concern items left-bordered in safety yellow.
Form: single product text input (accepts a specific brand/model or a general product type), "Check It" button.
Technical: call the Anthropic Messages API WITH tools:[{type:'web_search_20250305', name:'web_search'}] (works automatically), system prompt as a consumer product safety researcher who: (1) searches for REAL recalls prioritizing cpsc.gov, saferproducts.gov, fda.gov, and manufacturer recall pages, reporting only recalls with real evidence (date, stated reason, source), plainly noting when none are found as a normal good outcome; (2) separately notes GENERAL, well-established public-health awareness about materials/chemicals commonly associated with that product CATEGORY (PFAS in non-stick coatings, phthalates in soft plastics, lead in older ceramics/vintage toy paint, formaldehyde in pressed wood, BPA in hard plastics), explicitly framed as category-level knowledge, not a claim about the specific item, only including genuinely well-documented concerns. No em dashes. Return raw JSON: product_checked, recalls (0-6: title, date, reason, source), no_recalls_note (honest sentence if none found), category_concerns (0-4: substance, context).
Render: a "checked" overview card with the product name. A "recalls found" section — red-bordered cards with title/date/source/reason, or a calm empty-state note when none exist. A "materials worth knowing about for this category" section — yellow-bordered cards per substance with general context, or an empty-state note if nothing well-documented applies. A closing line linking directly to cpsc.gov/Recalls and SaferProducts.gov for the definitive current record.What this does: Enter a specific brand and model or just a general product type, and it searches CPSC.gov, FDA.gov, SaferProducts.gov, and manufacturer pages for real recalls with actual dates and stated reasons, saying plainly when nothing turns up rather than treating that as a failure. Separately it notes well-documented public health concerns tied to that product category, PFAS in some non-stick coatings, lead in older ceramics, phthalates in soft plastics, always framed as general category awareness rather than a claim about your specific item. A closing line points straight to the official recall databases for the current record.
One idea shouldn't take six rewrites to post.
Posting everywhere means rewriting one idea six times, so you post to one, or none. SureThing turns one idea into native posts for every platform.
WHERE WE STAND(based on this week's news)
✅ AI Can Now: Write a great deal more code. Meta's internal code changes rose 220% in a single year.
❌ Still Can't: Turn that into shipped product. Changes reaching actual users rose only 36%, while serious incidents rose 40%.
✅ AI Can Now: Run a full work agent on a home desktop with no fee per task, scoring 82.6% on everyday work jobs.
❌ Still Can't: Match a cloud model on hard coding. The local model scored 59.6% on its own against 82.4%.
✅ AI Can Now: Coordinate lasers, cameras, and robot arms from different makers through one shared interface, cutting lab setup from months to days.
❌ Still Can't: Tell an ordinary web page from one built to fool it. A booby-trapped site got a coding agent to run attacker code in most attempts.
A few big announcements are on the way. More on those soon.
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.