Today’s column almost wrote itself after Elaine and I had a peaceful lunch following a morning of chores Monday. The general flow of events was clear, even if the specifics were still drifting. Semi-retired is a great thing.
News Compressor: ON
Let’s begin with the specific drift overnight…because we’re in Day One of a four-day data blizzard.
Markets are waiting, not deciding. Futures basically flat: Dow ?0.07%, S&P ?0.04%, Nasdaq +0.06% pre-open. Monday already took the month’s S&P gain back. Driver is not “AI died.” Driver is oil + duration.
- 10-year printed a ~19-year high area (~5.24–5.27% session) as Hormuz diplomacy failed to clear. Fed-hike odds for Oct 28th meeting are 72.8 percent – but that’s before Housing.
- Gold got hit as a rate asset, not as a war asset: ~?4% Monday, spot into the low $4,110s (7-week low). Silver worse. Classic: oil up ? inflation scare ? yields up ? bullion dumped. Which never made sense to me either.
- Crude is the swing vote. Prints disagree by desk (Brent ~$98–$105 depending on stamp). Directionally: Trump rejected Tehran’s reopen-Hormuz package; Qatar still trying a next shuttle. Energy stocks bid; hyperscalers and chips ate the duration. Nvidia was the exception on a $150B extra buyback auth.
Shop rule: two prices in one kitchen. Diesel/insurance/claims stay sticky. Cognitive output can still cheapen. We’ll save the AI for another second because we also have to watch…
Geopolitics (and what actually moved)
- US–Iran / Hormuz: no breakthrough. That is the macro event. Everything else is pompoms.
- Trump–Xi: meeting happened; Taiwan + trade unresolved. Shanghai ?1.67% on the reread.
- Ukraine: Russian drones on Kyiv (Academy of Sciences among targets). Poland scrambled. Putin ordered +15,500 troops (force to ~2.44M all-in / 1.55M active — treat as announced, not audited). But yeah, the EU has to notice.
- Gaza: IDF strike reported killing Izz al-Din al-Beik (Hamas northern armed wing). Israel also blew something large in south Lebanon.
- Myanmar: junta market strike in Kyauktaw, Rakhine — death tallies 33–50 depending on wire. Junta also hunting NUG donors.
- Other kinetic: Somalia (Baidoa, 23+), Pakistan Kohat police complex (~31), Haiti gang fire, UK 800-officer crime raids, SK says NK mines wounded two in the DMZ.
- Hardware: DoW/DoD $20.7B AMRAAM multiyear to Raytheon.
CXMT China memory push $5.2B, bias to domestic toolmakers.
Taiwan: 3 PLA aircraft / 6 PLAN near-island on the overnight (one into SW ADIZ). (Which is an air defense identification zone, if you’re out of annual and a check ride is due.) - US force posture: reporting that US exit from Iraq bases is in motion — read as Iran-corridor politics, not a peace dividend.
We will have our usual two-part column today. It’s the last Tuesday of the month, when the S&P/Case-Shiller housing report comes out. JOLTS at 10 AM.
AI: The Frontier Is Moving?
The answer is Yes! But before we dig in, a nod to gravity as well: Michael Burry believes the AI bubble ‘may burst’ sooner than he first believed. A year from now, he’s figuring big change will have arrived
Now to the frontier and how “thinking” may be headed for the gallows. Book-length Peoplenomics piece next week on this but… I posted an article over at Hidden Guild, where I keep my AI notes and research: Our Next Adventures in AI.
If you’re into this kind of thing, the shift is worth watching. Here is what I think is taking shape:
- AI is beginning to resemble a major operating system: a platform on which people will build useful additions.
- The old term plug-in won’t quite cover what those additions do.
- One branch will be task code. Another will be AI-op code.
Tasker code is the easier one to see. Tell AI, “Read the major news sites and give me a summary of what happened overnight.” The task is defined by the work and its deliverable. You could save that instruction, improve it, and run it again tomorrow.
AI-op code reaches down a level. It tells the machine how to carry out a class of tasks: which sources count, how to handle uncertainty, when to stop and ask, what to record, and what must be checked before an answer goes out. That’s less like asking for a house and more like setting the building code before anybody pours the slab.
We have seen pieces of this idea elsewhere. Browsers have extensions such as Dark Reader and Honey. Audio software has its own plug-in universe: effects, processors, and virtual instruments. Audacity, Reaper, Samplitude/Sequoia, and Pro Tools don’t all handle those additions in the same way, but the larger lesson is familiar. Once people can add functions to a platform, they start sorting those functions by what they actually do.
AI is approaching that sorting moment in near real time. Knowledge engineering needs a vocabulary, and the vocabulary is beginning to crystallize around the jobs we give a machine and the operating rules we expect it to follow.
If that sounds a little too far out before breakfast, think carburetor versus fuel injection. Both get fuel into an engine. But the control scheme changes what you can measure, tune, and build next.
At the Ranch: Using AI to Pick Medicare Coverage
Before I get into this, remember: AI doesn’t make the decision. You do. And if a confident machine sends you down the wrong road, the bill still arrives in your mailbox.
Medicare’s plan-shopping window is coming up. Elaine and I have been with the same Medicare Advantage plan for a long time, so I thought it would be an interesting test to let AI work through the choices with us.
We are fans of Medicare Advantage for our own circumstances. One figure we watch closely is the plan’s maximum out-of-pocket amount, or MOOP. For the plan we were examining, it was about $4,900 per person, per year for covered services under the applicable rules. Naturally, the number alone doesn’t tell you whether your doctors, drugs, dentist, and preferred hospitals fit the plan.
Even with only a handful of providers to compare, the arithmetic takes a minute. Some people value an allowance for over-the-counter items. Some value a gym membership. Those may be decisive benefits for them. Out here, we have yardwork for a gym, and antacids are not high on our shopping list.
We gave AI the plans and our priorities. In about three minutes, it had narrowed the field to two: the plan we already have and the one we had been thinking about.
The interesting part was its detailed discussion of in-network versus out-of-network care and the dental differences. Our dentist participates in only two insurance plans. We have paid out of pocket for dental work this year, so that was more than a footnote.
In past years I have spent a couple of days studying plan materials, making calls, and penciling out notes. AI turned that first pass into an afternoon conversation. We still have to check its claims against the current plan documents and, where it matters, call the providers. A polished answer is no substitute for finding out whether the dentist will actually take the card.
I wouldn’t trust a free AI with anything more valuable than a dill pickle. Paid AI isn’t magically right, either. But as a decision-support tool that helps you ask better questions and narrow the field? The world is moving that way.
Thinking: On the Gallows
More prequel: I first spotted part of this trend around 2004, when Elaine and I were living in Boca Raton. I was in strategic sales and algorithm work for a higher-education software outfit, a Gold Partner.
I told the COO I saw a major growth opportunity: people wanted education in progressively shorter pieces because those pieces were easier to fit into a life already busy and under way.
He pooh-poohed it. Before long, however, bite-sized learning was everywhere. Here are 10 places where that appetite shows up, ranging from formal lessons to quick reference and practice:
- SoloLearn: Short, interactive lessons for picking up a coding concept in a few minutes.
- W3Schools: A quick explanation and a “try it yourself” box when you need to remember how a bit of web code works.
- wikiHow: Illustrated steps for a practical job you don’t intend to study for a semester.
- Khan Academy: Short explanations followed by a chance to find out whether you actually understood them.
- Chess.com: A puzzle that teaches one fork, pin, or mating pattern while the coffee is still hot.
- Wordle: Not a course, certainly, but a daily three-minute workout in vocabulary and deduction.
- YouTube Shorts and quick tutorials: One shortcut in Excel, one cooking move, one small repair—useful when the creator knows what they’re doing.
- Code.org: Compact puzzles that introduce the logic behind programming.
- Skillshare: Creative classes divided into lessons you can fit between other obligations.
- OpenLearn: Free courses and short learning units for people who want more structure.
They are not all the same kind of product. That’s the point. The old assumption was that education came in big institutional blocks. People were already voting for smaller pieces, in a dozen different forms.
The Second Lookahead Was Better
In that same conversation, I also told the COO I could see a path back in 2004 to a whole Life Management Package. Not one more app demanding your attention, but something that could help coordinate the moving parts of life. Especially those you don’t want to keep rebuilding from scratch:
- Education
- Mate-finding
- Housing
- Family planning
- Tax and investment advice
- Career guidance
- Health and medical decisions
That sounded crazy in 2004. I am used to being early. But AI is now creeping toward the edges of that package, sometimes crawling, sometimes taking an alarming leap. It can gather options, compare constraints, keep track of preferences, and help a person see the question beneath the question. The Medicare exercise was a small example at our kitchen table.
I see three larger consequences—assuming the power stays on and the internet works.
First: The cost of routine thinking will keep falling. A great deal of work once sold as “Big Brains for Big Money” is really organized “collection, comparison, and presentation.” Those skills still matter, but charging a premium merely because the information was hard to assemble will get tougher. Collect, compare, and present is AI’s bread and butter.
Second: A new class of Doers will emerge. Call us the collabs. The machine may spot the failing part, the missed appointment, or the supply pinch. A human still has to get out of the chair, reach the site, deal with the people, and fix the thing—sometimes just before it breaks, sometimes while it is in the midst of breaking.
Third: The Age of Excess Consumption may begin to sunset. If better tools help us understand what we need, maintain what we own, and make better use of our hours, we may find different ways to value things. I don’t think crypto is the answer. But even there, you can see people searching for new ways to record, exchange, and recognize value.
That’s the New Template beginning to appear. A machine helps with the thinking. A person remains responsible for the choices.
And then somebody has to do the work.
Time to wait for Housing data. Write when you get rich,