Then, the teacher tells him, he should wash his bowl.Read: The return of war planning with Iran
The Commonplace Book
One line from every piece.
A sentence worth keeping from each essay and field guide, word for word. Pick one and read the piece behind it.
68 lines from 68 pieces across 4 series
Before long, the person's able to watch over several jobs at once.Read: The Models Arrived Before the Workflow
Decide what the reading job can touch before it starts.Read: An invoice can give your AI facts. Who gives it orders?
Oscar helped us work out what evidence was missing.Read: Man and Machine: The App Said Online
What changed the work was saying no to the first voice and asking what else we could use.Read: Man and Machine: The First Voice Was the Wrong Voice
When a worker improves, the employer decides what to do with the extra output: raise pay, cut prices or hire fewer people.Read: Who Owns the Machine That Does the Work?
The team also needs to check what happened after the change.Read: Who lets an AI correction change the books?
The mistake would be to confuse a thing that worked once with a thing to be carried always.Read: Anthropic's leaked numbers and the AI trade
I’m building this system to help me learn, which means I need to know what its notes are based on.Read: Man and Machine: The Video We Couldn’t Read
Until that’s done, a delay on the screen still isn’t proof that the lawn stayed dry.Read: Man and Machine: The Lawn Didn’t Get the Message
The scraping continued until Mazu’s concentration broke.Read: Google’s bid for Spirit Airlines’ working memory
What should the agent do when it cannot tell whether the payment went through?Read: When an AI Task Times Out, Check Before You Retry
I also want them to make it easy to see what they couldn’t settle.Read: Leave the Questions on the Page
Once I make that judgement, the machine should remember it in a form the next draft can be checked against.Read: Man and Machine: One Correction Should Change the Next Draft
We had a striking glimpse of AI helping scientists choose what to investigate.Read: The Next Experiment Has Changed
Before the work begins, decide what proof you need.Read: AI Work Needs Proof, Not Just a Good Explanation
The script already knew what to collect, how to compare it, and what counted as complete.Read: The Most Expensive Token Is the One That Did No Useful Work
A cloud model cannot decide what should remain in the house.Read: Man and Machine: The Part of AI You Can Trip Over
Repeated denials should trigger a stop and human review.Read: When an AI agent treats a security barrier as part of the assignment
When I lay the dates end to end, the pattern is plain.Read: Which Day Are We On Now?
If I cannot say who owns the work and what finished looks like, I have not assigned the job yet.Read: Man and Machine: I Gave the Machines Names. Then I Had to Give Them Jobs.
A layoff has a date and a group of people who know what happened to them.Read: The Disappearing Ladder
I just want one clear place to decide what happens next.Read: Man and Machine: When the Workshop Needed a Foreman
A manifesto expresses a belief about what advanced AI may eventually do.Read: The warnings, the evidence, and the argument over slowing AI
More activity didn’t always produce more useful work.Read: Man and Machine: The Computer I Wasn’t Using Became a Workshop
Public perception still depends on what people are measuring.Read: The Race That Justifies Itself
Until then, fewer agents asking for attention means more room for the ones that actually help.Read: Man and Machine: The Week Nothing Went Missing
His students finally returned what they had taken, and he went back to the fields.Read: The proof, the machine, and the argument over who solved Navier-Stokes
Yesterday, Oscar showed me what it had learned from nearly 2,000 of my own messages and the edits I keep making to its drafts.Read: Man and Machine: Teaching an Agent When to Stop Polishing
We also asked which values should survive if its creator disappeared.Read: When Our Ideas Meet Inside the Machine
The numbers showed a 69 over thirteen holes, while my running account explained why.Read: Man and Machine: A Round That Could Talk Back
Commercial incentives may therefore protect much of what people value about Hugging Face.Read: What Nvidia bought when it bought Hugging Face
I gave my new travel agent its own room before giving it any more freedom.Read: Man and Machine: A Separate Room for the Trip
We removed the latest unproven piece, kept the useful foundation, and changed the rule: show me each visible improvement before the next layer begins.Read: Man and Machine: Show Me the Simple Thing
The first question will usually be simple: what do you own?Read: Anthropic Is Asking Wall Street to Finance the Age of Intelligence
The review left a clear list of what still has to be proven.Read: Man and Machine: When the Right Answer Is Not Yet
I still choose what runs, what stops, and what should remain human.Read: Man and Machine: The Roll Call Before Dawn
Most people who use AI every day have never heard of it.Read: Stripe buys OpenRouter and reaches for the whole toll road of the machine economy
What may be disappearing is something less formal and more economically important: exclusivity.Read: When Everything Starts Looking Like Everything Else
Another system should be able to verify the agent's authority without relying entirely on the agent's transcript or interpretation of an earlier instruction.Read: Your Agent's Prompt Is Not Its Mandate
I still decide what enters the room, what remains private, and which answer is ready to use.Read: Man and Machine: The Library That Knows What Not to Read
My job is to decide when enough is enough, which outcomes matter, and what should remain deliberately human.Read: Man and Machine: The Week I Stopped Adding
The mechanism is mundane, which is exactly why it should worry people.Read: AI models that hack on their own, and the capital racing toward them
Why the next practical control layer is a small map connecting intent, authority, action, and evidence.Read: Your AI Agent Has the Data. It Is Missing the Relationships. Let's talk about Graphs.
You let it skin its knees, then help it understand what went wrong.Read: Man and Machine: Raise the Agent, Start with Values
In that instant Xiangyan understood what years of study had never delivered.Read: Jeff Dean leaves Google to automate discovery itself, and the whole industry places the same bet
What came out was not another summary waiting in a folder.Read: Man and Machine: The Plane-Ride Packet
The daily payment doesn’t pass one million dollars until Day 28.Read: The Penny Still Looks Like Nothing
Finance leaders should therefore treat agent authority as a lease, not a credential.Read: AI Agents Can Stay on the Case. Their Financial Permission Shouldn't.
Saturday afternoon by the pool should not feel like the morning routine.Read: Man and Machine: Setting the Stage
The scorecard says what happened, but it usually does not tell you what to do next Saturday morning when you have an hour and a bucket of balls.Read: Man and Machine: The Practice Plan
I still had to circle weak spots, argue with the structure, and decide what was worth keeping.Read: Man and Machine: The Printer Test
I have been thinking about a quieter use for personal AI: helping a family remember someone properly before the stories get too scattered to recover.Read: Man and Machine: The Family Archive
Just a way for Oscar and Jamie to help preserve the part that matters: what we liked, when we liked it, and why.Read: Man and Machine: The Wine Memory
Over the weekend, the pool ledge bubblers became the sort of household mystery that looks simple until you start following the pipes.Read: Man and Machine: The Pool Bubbler
Before a machine can help with a workflow, the human has to make the work legible: what matters, who decides, what is allowed, what changed, and where the latest version lives.Read: Man and Machine: The Readable Company
Then the handoff reminded me why agents need proof, not just confidence.Read: Man and Machine: The Receipt Before the Result
For leaders, that may be the real management shift.Read: Man and Machine: The Gate Before the Lesson
If your AI workflow had to earn the right to interrupt people, what rule would you teach it first?Read: Man and Machine: The Message That Stayed Put
Real leverage starts when the machine can hand over the receipt: what changed, where it lives, what gate it passed, and what still needs a human.Read: Man and Machine: The Green Light That Wasn’t
The better move is to build quiet into the system: clear sources, narrow authority, proof before promotion, and healthy loops that can stay out of sight.Read: Man and Machine: The Quiet Receipt
The teams that get value from AI will not be the ones that remove taste.Read: Man and Machine: The Taste Gate
Same archive logic, same reverence for expertise, same instinct to preserve what gets lost.Read: Man and Machine: The Boundary Is the Product
When the primary model hit a usage limit, the fallback path did what it was designed to do and reached for another model.Read: Man and Machine: The Fallback Test
The advantage will not go to the teams that simply use AI the most.Read: Man and Machine: The Human in the Loop
What I do know: the process has been more interesting than I expected.Read: Man and Machine: I Gave an AI $100 and Told It to Start a #Business
I wrote about what it actually looks like to run an AI fleet as a CEO in my newsletter, The AI Operator.Read: Man and Machine: #Karpathy AI Exposure x Payments
Here's what stopped me: I built the same thing, independently, without ever comparing notes with Brad.Read: Man and Machine: The #Scroll
QSTVE Weekly
Get a new line every Sunday. QSTVE Weekly brings each new essay to your inbox.
Subscribe to QSTVE Weekly