A quick BlinkLabNews scan to roll with coffee first: Because a few readers are arguing for me to work harder not smarter. That’s a tale worth telling.
First, though, let’s suck up some coffee and run the global systems check.
News Compressor: ON
What has changed is the Plaskett Fire: Cal Fire mapped it at 8,241 acres and 1% contained Saturday evening, up from 5,924 acres Friday night, with the fire backed to Highway 1 and firing operations around residences and corridor infrastructure. . Cal Fire: https://www.fire.ca.gov/incidents/2026/8/26/plaskett-fire
NASA’s live Roman coverage is running and the 45th Weather Squadron gives only a 50% chance of favorable weather for the 7:26 a.m. EDT Falcon Heavy liftoff, citing the cumulus-cloud rule and surface electric fields. NASA: https://science.nasa.gov/blogs/roman/2026/08/30/nasas-roman-space-telescope-launch-updates/
Nepal’s official death count rose to 734 on the Nepal side and 16 in Tibet, 750 combined, with about 2,498 missing in Nepal and 3,044 combined; the State Department Saturday listed about 100 Americans unaccounted for and five rescued, still with no confirmed U.S. deaths. https://www.nbcnews.com/world/asia/nepal-floods-death-toll-climbs-missing-rescue-efforts-rcna594990
Columbia, S.C., Police Officer Christopher DeLong, 29, was shot and killed Saturday at Southeast Park; Officer David Dymock was wounded and listed stable; suspect Adam Dowdy, 33, also died. Chief Holbrook said it is the department’s first line-of-duty death in more than 50 years. https://www.thestate.com/news/local/crime/article317057853.html
Inside Pages
Weather Cluster
NWS Phoenix still has a Flood Watch Sunday evening through late Sunday night for south-central Arizona; Tucson/Flagstaff add watches into Monday for central and northern Arizona. Vector: Newly scheduled.
Oklahoma and northeast Oklahoma Heat Advisories run Sunday afternoon into evening with heat-index values to 106–107. Vector: Newly scheduled.
NOAA SWPC’s 0030 UTC Sunday product forecasts no G1-or-greater storming Aug 30–Sep 1; peak expected Kp is 3.67. Vector: De-escalating.
War Cluster
AP: a Russian strike on a Ukrainian warehouse killed 37, described as the war’s deadliest attack this year.
NYT: a Postal Service rule that could bar millions from voting by mail remains under a temporary judicial block. Expect appeals to splash this coming week.
NYT: California growers are plowing under millions of heads of iceberg lettuce after demand collapsed following the cyclospora outbreak.
Ontario Premier Doug Ford answered the U.S.-side “Lake America” rename with a shoreline sign reading “Lake Ontario. Now and Always.”
ICE deported British national Milo Yiannopoulos, per Sunday wires.
Burning Man is underway beside the Hawk Fire, previously reported near 95% contained.
Venezuela oil announcement still has no posted contract or named field list; markets are closed Sunday. Vector: Unchanged but consequential.
Related: Monday cash oil opens more than $2 off Friday on the unsigned Venezuela headline. Window: Monday session. Odds: 35%. Raise on a filed JV or named major; lower if Caracas or Houston lawyers punch holes. Evidence: AP “if realized” language still in force; weekend is dead tape.
Now, on to the main event of the Sunday coffee ponder:
The Product Was Always Time
People piss away time like they aren’t really going to die. Well, lookie-here: You are. Whether you own up to it or not, every waking second is a choice at the “Smorgasbord of Life” between living a purposeful life and pissing it away. Like social media. Ego exercises. Styles and fashion. Today I’m focused on purpose because that’s how I have rolled and likely will keep rolling.
See, there is a tendency whenever a new technology arrives for people to divide themselves into camps. One side becomes evangelists, the other becomes critics, and then both sides squander enormous amounts of time arguing about whether The New Thing is wonderful, dangerous, overrated, revolutionary, fraudulent, job-destroying, civilization-saving, or some combination of all of them.
I’ve never been terribly interested in joining either camp. My question has generally been simpler: Can I use TNT (The New Thing) to blow up my to-do list faster or better than I could yesterday? My TNT attitude didn’t start with artificial intelligence; it goes back more than forty years, and when I look backward, the recurring theme isn’t really computers at all. It’s leverage.
Early “Personal Explosives” Work
In 1982 I was already fooling around with sending wireless data into Seattle using AM radio. This was at a time when “wireless data” was hardly a normal household phrase, there were no smartphones, and most people weren’t walking around imagining that someday nearly every object they owned might communicate with something else. 276 PNW techies heard that first god-awful scratchy noise and decoded 300-baud modem tones off KMPS AM and FM.
Two years later, in 1984, I was the only guy on most transcontinental flights carrying a portable computer. I loved my HP 110C portable (aircraft certified) because while the rest of the cabin scrawled on yellow pads, lugging briefcases and voluminous reports, I could work on spreadsheets and memos that were print-ready by Nebraska on the SEA-ATL legs. Was I smarter than everybody else on the airplane? Hardly, but I had acquired something useful: a time advantage.
The guy making notes on a legal pad still had to arrive somewhere, type them, calculate things, revise them and distribute the results. Then proof the work and supervise staff. I could already be doing some of Monday morning’s work at 35,000 feet, and that distinction mattered.
Software Was Never a Religion
The same thing happened with accounting software. I started with VisiCalc, and people raised on modern spreadsheets probably cannot appreciate how astonishing the electronic spreadsheet originally seemed. This was on a Commodore business PET, not the C-64.
Before that, changing one assumption in a financial model could mean recalculating entire sections manually. The spreadsheet changed the economics of thought because you could alter an assumption in one place and immediately see numbers somewhere else change with it. Think two graphing calculators that could talk. Now make 10,000 of them work in unison.
Then along came Lotus 1-2-3. It was better for what I wanted to do, so I moved, and later when Excel became the better workbench, I moved again. My ALFMS (airline financial modeling system) was a first complex portable airline daily-operations tool in the mid-sized jet carriers. No IBM heavy iron or punch cards needed.
There is a useful lesson in that progression because I didn’t usually stop using software because it had suddenly become bad. Usually it was still doing precisely what it had always done, but I had learned enough about the problem to understand what the existing tool couldn’t do.
Even working with BoeingCalc — first and in some ways best of the 3-D spreadsheets — and Javelin, which could backfill to a projection, software was still the thinking geek’s speed bump. Still, I was able to manage a vocational college using spreadsheets, but it wasn’t yet “right.”
Then Came Databases…
Eventually my spreadsheets became increasingly elaborate: lookup tables, cross-references, student information, financial information and academic progress information. At some point I realized I was trying to make spreadsheets behave like databases, which meant the nature of my problem itself had to change with it.
A spreadsheet essentially asks, “What’s in this cell?” A database can ask, “What (else) do we know about this person, account, class, transaction or event?” and that is an entirely different way of organizing information.
So I moved into flat-file databases, including Jim Button’s FileExpress. It was a wonderfully useful program, and inevitably I eventually discovered its boundaries, too. That discovery pushed me toward the next rung rather than convincing me the old one had somehow failed.
Next came dBase III, and after building a dozen-user Novell LAN (my first “Electric Railroad” for data) dBase IV was ready. By the mid-1980s I had designed and built a school-wide student tracking and academic-progress system in dBase, and this wasn’t computing for the sake of computing. The only thing close at the time was RGM Software and Rafael Gonzalez did great code. Heavier iron, but best of class and we shared ideas around it.
The point was getting to best-of-class management. Which students were progressing, who was falling behind, which programs were working, where were the exceptions, where was the (federal student loan) money going, and what required management’s attention? Details matter with the feds.
Instead of information being scattered through paper files, instructor memories and separate office processes, the institution began asking proactive questions. That same conceptual architecture followed me years later as the technology evolved into SQL, servers, networks, interactive reporting and financial tracking.
By the early 2000s I was helping a Microsoft Gold Partner (Campus Management) apply expanded versions of some of those ideas in higher education. The software had become enormously more sophisticated, but the basic management idea hadn’t changed very much: Get the right information into a structure where it can be found, compared and acted upon faster than the other guy can do it.
VisiCalc wasn’t the destination, and neither were Lotus, Excel, FileExpress, dBase or SQL. Each was simply another rung on a ladder, and every rung bought some combination of speed, reach, memory and time.
Sometimes the Vocabulary Comes Later
There is another clue that you’re working close to the edge of a technology: sometimes the vocabulary hasn’t settled yet. You wind up needing words for things because the normal language doesn’t quite describe what you’re trying to build.
I ran into that as far back as 1985 while working around what we would now comfortably describe as speaker-independent voice recognition. Back then, I remember conferences where the language was still floating around things like “user-independent voice,” because even the terminology was evolving along with the technology.
I ran into the same problem again four decades later while building my new Over-the-Horizon project. I needed a term for a discrete, forward-looking piece of information that could be extracted from present information and placed onto a future time field, so I wound up inventing one: “Future Atom.”
A Future Atom is simply a bounded piece of information carrying some claim about a future condition, event or interval. The fact that I needed a term for it doesn’t make the concept mystical; it means I was trying to describe an information object for which the existing vocabulary wasn’t particularly useful. The evolution of a technology usually requires language distillation.
That experience feels very familiar. Sometimes when you get to the tool early enough, you’re not only learning the tool—you are still learning what to call the things the tool lets you do. The vocabulary often arrives a little later than the capability.
Then Along Came AI
Which brings us to artificial intelligence. I didn’t encounter AI after a lifetime of avoiding computers and then suddenly decide chatbots were miraculous; AI arrived after I’d already spent four decades watching information technologies replace one another.
So I treated it the same way. Get in, use it, push it, find its limits, figure out where it saves time, figure out where it wastes time, and figure out what it can do reliably versus what still needs a human standing over it with a club. In other words, use it as a tool, not a religion.
That research work has now produced four books touching AI from different directions: Mind Amplifiers, Co-Telligence, Theomachines, and The Judgment Engine. There is also a substantial amount of free learning and theoretical work at HiddenGuild.dev, including material aimed farther out toward the edges than the usual “how do I write a better prompt?” discussions.
Concepts like declaring a Session Framework Experience (SFE) evolved along the way. Some of that work asks how machine reasoning might be structured differently, how local AI changes the economics of intelligence, how information becomes reusable organizational capital, and how alternative compute architectures might accomplish useful AI work with substantially less energy. A couple of those architectural ideas have already gone far enough for me to file two provisional patent applications around novel compute architectures aimed at lowering the energy cost of AI.
Developers are reading some of those edge papers right now, which is exactly where I like research to be. Not because every idea will turn out to be right, but because leading-edge work is supposed to involve building, testing, discarding, revising and keeping what survives. The failures are part of the tuition.
Careful Use Note on TNT (The New Thing)
The operating rule hasn’t changed much since 1982. Get your hands on the new thing early enough that you have time to discover what it really does before the herd catches up, and find out what it doesn’t do while you’re there — preferably first.
AI makes mistakes, certainly, but so did spreadsheets when the operator wrote the wrong formula. A database could store garbage flawlessly, and SQL could return a perfectly accurate answer to a stupid question, yet nobody therefore concluded that databases should be outlawed.
We learned how to operate them. AI is the same problem moved farther up the abstraction ladder, except this time the leverage can be enormous. Oh — and never judge what you think of AI on the free crap. There’s a world of difference between a frontier model and something for wannabes to play judgy over: that’s the difference between testing a technology and testing the cheapest implementation of it. Frontier models are simply a different class of machine, and their reliability is miles better.
The Real Asset Is Time
The great benefit of computing isn’t necessarily computation. It is often time recovery, which is something I’ve gradually come to appreciate more than any benchmark or feature list.
The spreadsheet recovered recalculation time, databases recovered lookup time, SQL recovered organizational search and reporting time, networks recovered transmission time, and the Web recovered discovery time. AI is beginning to recover something larger: cognitive assembly time, the hours spent gathering, sorting, comparing and assembling information before serious thinking can even begin.
Let me give you a lived example. Suppose you wake up one morning and have a strange thought: Could I use the Web itself to look ahead into the future?
Not crystal balls, and not predicting Tuesday’s Dow close to six decimal places. Instead, could we collect the enormous number of statements already being made about the future and determine where they point? That’s a very different problem from fortune-telling.
Government agencies forecast, energy agencies forecast, farm agencies forecast and shipping companies discuss coming capacity. Weather services publish outlooks, companies publish guidance, news articles say next month, later this year, expected in September, by winter and within six months, while prediction markets put money against future outcomes and human forecasters make their own claims.
Suppose we collected all of those forward references, timestamped them, classified them and assigned their future windows. Then suppose we separated original sources from copies, collapsed duplicate ancestry, down-weighted known calendar events, tracked which information source moved first, and preserved contradictory evidence instead of throwing inconvenient observations away.
Finally, suppose we laid all those forward-looking statements onto a time axis and asked a simple question: Where are they clustering? That is the basic idea behind my experimental Over-the-Horizon software, and it turns the Web from a pile of pages into a dated field of forward-looking signals.
Yes. A Software Time Machine
It may work because (like in bullets and ballistics) initial states degrade predictably over time. And so, out on the edge, we get thinking… The dated forward-looking pieces are the Future Atoms described earlier; once extracted, weighted and laid onto a common time base, the question becomes whether independent sources begin clustering in the same future neighborhood.
Now a mind-stretch: imagine doing one serious OTH run manually. You visit the sources, read hundreds of pages, summarize them, copy the useful passages, record publication dates, decide which sentences actually contain future information, extract future dates, normalize fuzzy time language, separate facts from forecasts, identify geography and classify domains.
Then you determine whether five stories represent five observations or five rewrites of one Reuters story. You work out whether a frightening headline is actually new or merely a twelve-day-old event being recycled, and then you build the source ledger, forecast ledger and working spreadsheet.
Next come the formulas. You normalize the data, collapse duplicates, calculate weights, calculate the temporal field, compare it with the previous run, determine what moved, and then go back into the evidence to figure out why it moved. And then there are the graphics.
Running the “Software Time Machine”
Oh, it’s real, and I will be doing it now and then on the Peoplenomics subscriber side. Since you’re a natural-born skeptic, here’s the run given subscribers last Wednesday. You don’t have to take my description of the process on faith:
Over_the_Horizon_Report_v0.98_2026-08-26_COMPLETE
Scan the whole thing, read the Executive Summary, look at the glossary, and eyeball the charts. Focus: Look at the graphs in Wednesday’s report. Sure, I can make every one of those manually; I’ve been making spreadsheets and graphics for decades, but first I have to build the spreadsheet correctly, define the ranges, build each graph, establish the scaling, label everything, check it and then repeat the exercise when the data changes. The color and style, and….
Or I can offload much of that mechanical process into a few dozen lines of reusable tasking code. The machine can take structured output (speak JSON!), calculate what needs calculating, produce the graphics, preserve consistent scaling, place them into the report and prepare much of the publication package without requiring me to rebuild the process every time.
That isn’t eliminating my ability to make a graph. It is eliminating my need to spend part of my remaining life making that graph again, and that distinction is enormous. Capability stays with me; repetition goes to the machine.
A complete manual OTH run could easily consume something like 30–60 hours of human work. There is a practical absurdity hidden there because by the time you’ve spent 30 hours looking over the horizon, part of the horizon has already arrived. Besides, I will be turning 78 early in the new year… no time to waste!
Information Has a Temporal Base
This is where another Ure-world obsession comes in. When I look at information, I don’t see only its content; I also see its temporal base. Word-Frequency Analysis has been around for a long-ass time, and intelligence shops have been counting changes in language and themes since at least World War II. OTH simply takes that old shop-floor trick, puts it onto a much larger dated corpus, and asks an additional question: Where in the future is the changing language pointing?
When did it become knowable, when did somebody first recognize it, when did markets recognize it, and when did ordinary news recognize it? Then comes the most important question: when did everybody recognize it, at which point knowing it may no longer produce much informational advantage?
Having a five-day head start in almost anything can be worthwhile. It doesn’t automatically make you right and it certainly doesn’t necessarily make you rich, but five days gives you five additional days to think, check, investigate, prepare, hedge or dismiss the signal if further work says it’s nonsense. The objective isn’t supernatural knowledge; it is earlier useful structure, which is why we’re experimenting with OTH in the first place.
Can public information already sitting in plain sight produce temporal structure before the eventual news narrative becomes obvious? Maybe it can and maybe it can’t, which is exactly why the work remains an experiment rather than a religion. The audit is there precisely because interesting isn’t the same thing as correct.
Now look again at the actual OTH report Peoplenomics readers received Wednesday. Don’t just skim the prose; look at the inputs, classifications, Future Atoms, competing evidence, audit layers and especially the graphs. That’s the proof-of-work part of this little sermon.
Think about what would be required to produce that manually. Gather everything, read everything, build the databases and spreadsheets, construct the graphics, establish the scaling, lay out the report, check the arithmetic, write the explanatory material and prepare the whole package for publication.
Then ask yourself one question: How long did that run take once the reusable tasking modules had already been written? The answer was 27 minutes of compute time, not 27 hours. Compared with something on the order of 30–60 manual hours, you’re looking at roughly a sixty-to-one compression or greater in the mechanical work surrounding the analytical process. Now we’re getting to the real reason we use AI.
Our Reports Aren’t Our Product
ChartPacks, deep-dives into theory, and things like the OTH view: The obvious output is a written report and/or summary. But the report isn’t really the product. The report is evidence that the process worked; the product is the time I didn’t have to spend producing it.
That recovered time can be redeployed into another paper, another experiment, building something in the shop, learning something, working outdoors, talking with somebody I care about, or simply thinking longer about the handful of decisions where human judgment still matters most. Those are all higher-value uses of my finite time than repeatedly resizing chart axes.
The machine can help gather evidence, classify, sort, compare, calculate, graph, draft and audit. What I increasingly want my own brain working on are the questions What does this mean? What are we missing? What would prove it wrong? And what should we do next? Those are better uses of human life than repetitive information assembly.
The Information Revolution Had Assembly Lines
But at least now, with AI tooling, we can automate a great deal of it. More important, the technology is finally getting good enough that we have a genuine choice about how much of that repetitive work we want to keep doing ourselves.
Which brings us all the way back to those transcon flights in 1984. The laptop wasn’t really the point, and neither were wireless data, VisiCalc, Lotus, Excel, FileExpress, dBase or SQL. Today AI isn’t really the point either. The product was always time, and the technology simply kept getting better at clawing back some of it.
Because every human gets only a finite allocation, I have become increasingly reluctant to spend mine doing manually what a machine can now do faster. I’m not asking anyone to worship AI, trust it blindly or pretend it is always right; I am suggesting something much more practical, which is to learn how to make it work for you. We audit like hell because machines make errors, and even humans make errors, remember.
This isn’t fundamentally a contest between humans and machines. It is increasingly a contest between humans who know how to use machines and humans who don’t, and that contest is already well underway. So here’s the “Why?” of this morning’s in-depth and free Sunday discussion: If you’re not actively mastering ways to reclaim your personal time with ALL TOOLS, including AI, the front wheels of the Bus that is the Future may already have run over you.
And no, we didn’t go through all this merely to tell you that you’ve been flattened. We just don’t want the back wheels to get you, too. That’s the whole point of the freebie.
One for the Road
Apparently, the heat will come off at some point and America will head into fall. From our “For the Man Who Has Everthing” folder: Drill Pumpkin Power Gutting Tool 6 Blade 304 Polished Stainless Steel ~Rotary Power Speed & Will Not Damage Seeds ~Pumpkin Wall Shaver Attachment ~Inside Cleaning. Well, I’ll be…a tool I don’t have! Echoes of ShopTalk.)
Off to the shop now, where I have 3D printers and CNCs to supervise. Yeah, we use them and an Evolution table saw, too.
Write when you find the right tool,