In this issue
The AI boom has been an abstraction for most of us, something happening in datacenters we will never set foot in. This week it showed up somewhere we all look: the price tag on a new MacBook. Apple raised prices on Macs and iPads and pointed straight at the memory shortage the AI buildout is causing.
There is a thread running through this whole issue, and it is the one our name is built on. AI keeps proving it can do the work, then quietly proving it still needs us watching. Doctors lost skill when they leaned on it. Ford had to rehire the humans it tried to automate away. And the cheapest capable coding model in the world is suddenly open-weight and Chinese. Let us get into it.
Topics of the day:
Apple raises Mac and iPad prices due to AI crunch
AI use is eroding doctors' real skills
Ford rehires the humans it tried to automate
Curated reads, including a tool I built
China's GLM-5.2 beats GPT-5.5 on code
The Shortlist: Apple's M7, Midjourney's scanner, agent IDs
Apple raises Mac and iPad prices as AI eats the memory supply
What's happening: Apple raised prices on MacBooks and iPads this week, with the entry MacBook Neo jumping to $699 from $599, and blamed the memory and storage shortage driven by the AI datacenter buildout.
In practice:
Budget more for any hardware refresh this quarter, because the cheapest MacBook just rose 17% and analysts expect rivals to follow with their own increases.
Pull forward purchases you already planned, since memory prices are still climbing while datacenters buy up supply, not settling back down.
Watch your AI vendor bills too, because the same chip crunch that hit laptops feeds the compute costs your tools quietly pass along.
Bottom line: The AI boom just became something you can feel in your own checkout cart. Buy the hardware your team actually needs before the next price round lands.
AI is quietly eroding the skills of the people who use it
What's happening: A new study in The Lancet found that endoscopists who started using AI to catch precancerous growths got worse at finding them unaided, with their detection rate falling from 28.4% to 22.4% on the days the AI was switched off.
In practice:
Keep practicing the core parts of your job by hand sometimes, since the doctors lost ground precisely when the tool was taken away.
Treat AI output as a draft you verify, not an answer you accept, especially on the judgment calls you were actually hired for.
Measure your own work against the model now and then, so you catch your skills drifting before they show up somewhere expensive.
Bottom line: The tool that makes you faster today can quietly make you worse at the thing you do. Senior author Yuichi Mori put it plainly: there is no fix for deskilling yet.
Ford is rehiring the humans its AI was supposed to replace
What's happening: Ford has been rehiring veteran "gray-beard" inspectors and engineers to retrain younger staff and reprogram the AI quality tools that were not catching defects well enough on their own.
In practice:
Pair AI with the experienced people who can tell when it is wrong, since Ford's actual fix was human judgment correcting the machine.
Expect a quality gap wherever AI replaced headcount quickly, because the misses show up downstream on the line, not in the vendor demo.
Keep your senior people close to any AI rollout, as they are the ones who catch the confident mistakes a model learns to repeat.
Bottom line: Ford ran the experiment, and plenty of companies are running quietly, and found the humans were load-bearing. The cheapest rollout is rarely the one that ships good work.
Read Later
Model Fit - I built this one. It benchmarks different LLMs on your own codebase with repo-specific probes and blind rubric scoring, so you pick models on your real work, not a generic leaderboard.
Karpathy on a new way to work with Claude - Andrej Karpathy shares a more inline pattern for working with Claude that is worth copying into your own setup.
Why does everyone hate AI? - Paul Krugman on the booing of Eric Schmidt and why the public mood on AI has soured even as the tools improve.
Open Knowledge - An AI-native markdown editor and LLM wiki, part of the growing case for structured markdown over RAG.
China's GLM-5.2 beats GPT-5.5 on coding at a sixth of the cost
What's happening: Z.ai released GLM-5.2, an open-weight model that scored 62.1 on SWE-bench Pro against GPT-5.5's 58.6, runs at roughly a sixth of the cost of US frontier models, and is now the strongest model many teams can openly run.
In practice:
Test GLM-5.2 on a real coding task this week, since open weights let you run it yourself and it already tops several coding leaderboards.
Recheck your AI spending, because a capable model at one-sixth the cost changes the math on anything you run at volume.
Factor in where it runs, as GLM-5.2 is tuned for domestic Chinese hardware, which matters for some teams' data rules.
Bottom line: The cheapest capable coding model right now is open-weight and Chinese, which was unthinkable a year ago. Worth a serious test before you renew an expensive seat.
The Shortlist
Apple is skipping its high-end M6 Mac chips to jump straight to an AI-focused M7 line, a sign your next pro Mac will be built around on-device AI.
Midjourney unveiled the Midjourney Scanner, a full-body ultrasound machine, its first hardware bet and a strange pivot for a company known for AI images.
The Linux Foundation announced the Agent Name Service, a DNS-based identity layer to verify which agent is which, backed by Cloudflare, GoDaddy, and Salesforce.
This newsletter is where I (Kwadwo) share products, articles, and links that I find useful and interesting, mostly around AI. I focus on tools and solutions that bring real value to people in everyday jobs, not just tech insiders.
