Weekly Review: Oct 12–18, 2026

TL;DR: Five posts this week, and in most of them the part that decides the outcome sits outside the frame you were looking at. The sender stares at the silence, while the real cause is the receiver’s pile of two hundred messages. The asymmetry series ends by asking the question its nine playbooks never asked: should a gap exist at all? The classic explanation of lift leaves out the air a wing throws downward. Ads in AI answers keep the search ad playbook but move money away from publishers in every scenario I modeled. And an agent can book an electrician through calls that all succeed while the landlord never hears about it. If you only click one link, make it The Matryoshka of Humans in the Loop, because an agent that can commit money or work needs an explicit path to whoever holds authority over it.

The asymmetry series ends this week, in two parts: one on attention, one that looks back at all nine earlier posts. The other three posts this week come from different corners: physics, ad economics and agent design. They kept landing on the same move. The useful answer only turned up once you looked at a party, or a piece of the system, that the usual picture leaves out.

This Week on the Blog

You Remember Every Word. They Got 200 Messages.

You spent an hour on the message, and they gave it four seconds somewhere between message 87 and message 88 of their day. The post leans on Herbert Simon’s line that a wealth of information creates a poverty of attention, on the spotlight effect (we overestimate how much others notice us), and on research linking workplace telepressure to burnout and poor sleep. Put them together and you get a loop. The sender reads the delay as a verdict, follows up, adds to the pile, and slows the reply down even more. The fixes sit on both sides. Senders write for triage and send one follow-up that adds something, and receivers send a cheap “got it, Thursday” and publish their response rhythm. This is Part 9 of the asymmetry series.

How Do Planes Actually Generate Lift?

The textbook story says the air over the longer top of a wing has to hurry to meet the air underneath at the back edge. NASA has a whole page on why that is wrong: on a lifting wing the air over the top reaches the trailing edge first, and the two parcels never meet again. A wing lifts by turning a large mass of air downward, and the air pushes back up, which is why a flat paper plane flies and why planes can fly upside down. The lower pressure on top is real too, as the other half of the same flow. The post has an interactive demo with an angle-of-attack slider, a little momentum math, and the circulation picture behind the Kutta–Joukowski theorem. This is the fifth Lunch Break Physics.

What Doesn’t Close

A friend’s line sets up the ending: asymmetries, like ambivalence, mostly don’t resolve. The closer puts every gap from the series on two axes. One asks how far a gap can be narrowed with a named intervention, the other how well the gap that remains is justified. And the burden of justification sits with whoever holds the authority. That gives four regions, each a place to start: close it; keep the function and cut the excess; hold it and discipline it; or compensate, constrain and create recourse. It also argues that performed equality (“we’re all equals here, speak freely,” from the person who signs your review) is worse than honest hierarchy, and it withdraws Part 7’s framing of a parent’s authority as mainly a moral-hazard problem to reduce. This is the tenth and last part.

Same Ads, Different Room

Ads in AI answers are bought much like search ads, with auctions, cost-per-click and outcome bidding, pixels and attribution partners. Three things change. The system reads the whole conversation instead of a short query. In the format I model, one ad sits next to one synthesized answer instead of ten links. And no publisher stands between the platform and the user. A small Python simulation shows publishers losing in every scenario I tried, and total ad spend only grows if the answer engine converts about 1.45 to 1.67 times better than search. The numbers are my assumptions, and the code is on GitHub so you can change them.

The Matryoshka of Humans in the Loop

Amazon Mechanical Turk closed on 30 September 2026, and its old tagline, “artificial artificial intelligence,” turns out to be a good lens for what came after it. Each layer hid another kind of worker: a chess master inside an automaton, crowd workers behind an API, and by 2023 language models inside some of those workers. Agents now call people again, through MCP elicitation and confirmation steps. A hypothetical wall-box booking shows the gap: every API call succeeds, the tenant taps confirm, and the landlord is never asked while the grid operator is never notified. The post argues that the nesting has to stop at whoever holds authority over the work, and it ends with the AWS primitives for building that, from Bedrock Agents user confirmation to temporal policies in AgentCore.

The Thread This Week

Look at where the answer was in each post. In the attention post it was in the receiver’s inbox, which the sender can’t see. In the lift post it was in the air behind the wing, which the textbook diagram stops drawing. In the ads post it was in the publisher’s referral traffic, which drops out of the chain, and in the Matryoshka post it was at the landlord’s front door, which no API call reached. And the series closer adds the question the earlier nine posts never drew: whether a gap should exist at all. Sound familiar? The habit is the same each time, and it’s a cheap one. Before you trust the picture, ask who or what it leaves out.

Until Next Sunday

If there is one habit to take from this week, it is to ask who is missing from the picture before you act on it. That goes for the unanswered message, the gap you are trying to close, the explanation you learned in school, the ad slot, and the agent that is about to book something. Who is outside the frame of a decision you are making this week? Tell me in the comments on LinkedIn.

This is the Weekly Review, and it also goes out as Sunday’s newsletter.