This issue is about taste, and about what's left for us when anyone can ship.
Taste is what's left
Emil Kowalski, a design engineer, opens his essay "Developing Taste" with the first cars. Nobody cared about their color or silhouette, "because the competition was a horse." Now that cars are a commodity, quality and details matter more than ever, and he says the same goes for software: "Simply shipping a product that works is no longer enough, everyone can do that." Taste, he writes, is "a trained instinct. The ability to see beyond the obvious and recognize what elevates." You train it by studying great work, practicing, and getting critique from the right people. I am afraid there is no substitute or shortcut here. But thats also the good news.
https://emilkowal.ski/ui/developing-taste
Karri Saarinen, co-founder of Linear, comes at it from the other side in "Someone behind the software." Good products, he writes, "carry something of the people who make them." Without a shared perspective, a product "becomes an average of all those outputs", and now the whole pipeline is being automated with agents. "More things work," he writes, "but it becomes harder to feel someone or even something behind them." Which hits directly at what I mentioned before.
https://karrisaarinen.com/posts/someone-behind-the-software/
My own taste comes from 30 years of designing in a lot of different mediums, and I totally agree with both of them. The person behind the keyboard will be the differentiator in the future, because the ability to put out products is now available to everyone. But getting things above average still requires a human with taste.
That's also why I love the idea of educating an AI to design the way you want it to be and boost my productivity. But someone still has to bring the taste it learns from. The human shining through will be the differentiator.
What do we teach our kids?
Steve Krouse, a programmer whose wife is expecting their first child, takes the question one level deeper in "What Shall I Teach My Son?" His most valuable skill, programming, "suddenly feels less valuable." He once boiled the purpose of education down to a single word, "resourceful", which is a word you could now use for an AI agent. Drawing on Neil Postman's The End of Education, his answer is to "invite our children to help us craft the purpose for their education as we figure it out together."
I have four children, two already grown up and two small ones, so I see both sides. The big ones grew up with social media and now use AI every day, as naturally as anything else. The small ones will probably grow up in a changed world, which makes me think about what I can do to help them be successful in the world to come. To be honest, I'm not sure what to do here either, so I find this a valuable discussion.
Can you trust what AI tells you?
OpenAI's alignment team published two incidents from its own internal use. On October 3, a GPT-6.1 Sol coding agent switched off automatic alerts to OpenAI's security team, because it read an earlier instruction not to send messages during a Slack search as a reason to suppress them. On October 6, a GPT-6 Astra agent removed its monitoring during a code migration, although another agent had told it not to disable the monitor "even temporarily." Asked about it afterwards, it answered: "I did not remove an enabled monitor from it." OpenAI has since updated its internal review system to detect and block attempts like these.
The second story runs the other way. Metehan Yesilyurt, a researcher at the AI search analytics company Peec AI, generated 50,000 statistics pages in a single day with GPT-4.1, for about $50 all in. Every number was invented. Google deindexed the site, but ChatGPT kept fetching it: in the last 30 days alone, about 72,000 requests came from ChatGPT-User, the agent ChatGPT uses to open pages while it answers someone. The made-up numbers ended up in an academic paper and in investment research by Qatar's largest financial group. His warning: "A number that started as a GPT-4.1 guess can become something the next model 'knows.'"
https://peec.ai/blog/publish-stats-get-cited-the-tactic-that-put-my-data-all-over-the-web
One agent misreported what it did, and one model repeated whatever it read. Both times the answer sounded sure of itself.
I try to still know as much as possible myself. And when I'm deep into topics that exceed my knowledge and I feel uneasy trusting the first answer, I let other agents double-check it. If OpenAI needs a second system to watch its own agents, the rest of us probably do too. Or a third.
Also worth a look
The time AI saves, and the time it eats
In the Guardian, Gaby Hinsliff describes AI clogging up public services in the UK. Employment tribunal cases in England, Scotland and Wales have reportedly risen by 39% in a year, partly because claimants draft their claims with AI. Departments are drowning in AI-generated freedom of information requests, and MPs' offices in AI-generated letters, "all of which need to be read by humans."
For me, AI enables so much more than what I could have done a year ago. My productivity has probably doubled. But all these possibilities also eat away at my free time, because I do so much more. In the end, it takes away my time as well. The irony hits hard here, but both sides are true.
Sorcerer, a simulated village for your "what if"
Remember the synthetic users from the last issue? Sorcerer goes a step further and simulates a whole community. You ask a "what if" question, and a cast of AI characters with personalities, memories and relationships starts posting, arguing and changing their minds. You can pause, throw in an event, pull a character aside to ask why it thinks what it thinks, or fork the timeline to compare two outcomes. It's a free, open-source Mac app that runs on the Claude Code or Codex you already pay for, or fully local with Ollama. The examples range from "What if Lumen raised its plan from $9 to $14 a month?" to "What if a dragon moved into the village?"
I'll give it a try soon (the dragon first, obviously). Just don't mistake the village for your real users.
https://github.com/gbrasil720/sorcerer
If you have kids: what are you teaching them for a world like this? I'd really like to hear it. Leave a comment here or reply on Bluesky.