Maybe that’s why the word “taste”, ironically, left a bad taste in my mouth. It feels like it’s used in the wrong quantities (too much) and at the wrong time (too imprecise). It’s like the word “interesting”, a word that can often feel li…

Read more

For example, today I just saw a Tweet where a guy used AI to perform surgery on a mouse! This could transform ALL lab experiments in a profound way (at scale, with precision, and repeatability).

Of course, transformation doesn’t have to be this groundbreaking. Sometimes a simple reconfiguration of a team org structure or a workflow can improve efficiency and effectiveness. This too is transformation, even if it’s on a completely different scale of effort and output.

To me, this is another opportunity to make things Sesame Street Simple.

Here’s my attempt to define 3 different options with 3 different illustrations.

So You Want to Upgrade a House

Let’s start with the infographic, then address each one.

If you want to transform the feel of a home, you have at least three different levels.

And summarizing:

1. Redesign: Everything from new furniture to a fresh coat of paint.

2. Renovation: Ripping down walls while keeping the bones intact.

3. Reimagination: Tear the house to the ground, start from scratch.

Redesign

My wife loves changing how the house feels with the seasons. Sometimes it’s just a matter of switching out pillows or putting out different decorations. Sometimes we get bored with a room’s color (or it’s starting to chip), and we put up a coat of paint.

This is equivalent to nuts-and-bolts AI transformation.

• Summarize a document

• Draft an email

• Run detailed search

• Translate X to Y

• Speech to Text

These can be huge gains! But they aren’t rocket science anymore. They make everything different and better (transformation), but for now, much of it is table stakes.

Renovation

Sometimes an organization is constrained by the literal structure of their building. Maybe there aren’t enough conference rooms. Maybe there is no soundproofing (something that didn’t matter before remote work became a thing). Maybe the wiring sucks, and that is limiting the internet speed.

Here, you can’t just get away with an off-the-shelf solution. You might need a bit more planning, more budget, more focus, and more effort.

In AI, you might rip apart a key system. Perhaps your team uses Excel, and there’s a tremendous amount of manual back-and-forth updating. Here is where an AI agent with connectors might be able to replace multiple steps of an SOP with a single, holistic pass. What if you could process the intake form, pull CRM data, and create a proposal all at once? This might require setting up triggers and agents, plus some testing and training. But you can eventually throw out an old way and introduce a significantly better, new way.

Reimagination

What if you had no constraints?

You’re not limited to the existing floor size, and you have no historical baggage.

You bulldoze the building and build it big, with the exact configuration that supports your goals.

This is what AI-first from day 0 looks like. You can use “1st principles” reasoning to identify the goals and the means to achieve them. Existing SOPs may be useful for inspiration, but to hell with them if that’s how we used to do it.

You’re no longer making choices based on what can fit where. You build the building around what and how you want to operate.

So there you have it. 3 different ways to scope your AI transformation. I hope it’s a useful illustration and metaphor for your organization!

The annual ritual and its missing half

Every year between November and January, people who think for a living write down what the next year will bring. Investors, founders, newsletter writers, the person at your company who runs the all-hands deck.

Then the year happens, and almost nobody goes back to check. The few who do grade themselves, and self-grading has a known bias.

Here’s the bias in one example. Mike Ippolito, who co-founded Blockworks, posts 27 crypto predictions every New Year.

On a podcast in December he said he got 21 of his 27 for 2025, which is 78 percent. A trade publication had graded the same list a few weeks earlier at 48 percent. Both grades were done in good faith. Neither is wrong in any way you could prove. They differ because the two graders held the predictions to different bars, and nobody wrote the bars down.

That gap is the subject of this piece, and of the tool we built to measure it.

Three bars

Every argument about whether a prediction came true is an argument about the bar. “Flying cars arrive in 2026” is met if a car hovered for 30 seconds in a demo, and missed if you meant something you could commute in. The facts are the same either way. The disagreement is over what the sentence promised.

So instead of picking one bar, write 3 of them, before looking at any evidence:

• Strict is what a hostile critic would require before conceding. Usually the literal claim with its numbers and dates intact.

• Standard is what a fair reader would accept as “this happened.”

• Lenient is the weakest reading a critic would still concede counts. Lenient about the claim’s scope, never about the evidence. A single anonymous post meets no bar.

Grade all three. Report all three. The number that matters is the distance between them, which we’re calling the generosity gap: how much the grade depends on who is grading. A prediction with hard numbers in it has a small gap because the bar is in the claim. A prediction built from adjectives has a large one.

(The name is Claude’s. We went looking for an existing term and found “generosity gap” only in philanthropy, where it means the distance between what people say they’ll give and what they give. Nobody seems to use it for forecasting, so treat it as ours until someone points us at prior art.)

Two more rules did a lot of work. The bars get written before any evidence is fetched, because writing the bar after you know the answer is exactly how self-grading goes wrong. And the author’s own later words count as evidence with an asymmetry: an admission that a call missed is strong, a victory lap is weak.

Why this works now

The reading, the searching and the bar-writing are judgment calls, and a strong reasoning model does them. The grading is a different kind of work: the same yes-or-no question, asked over a small pile of evidence, hundreds of times, where what you want is a number that means the same thing every time you ask.

For that we used Jev, TypeSafe’s decision classifier. You give it a state (the claim, the bar, the evidence rows) and typed questions (is the bar met, does the evidence show it was missed, how strong is the best evidence), and it returns calibrated probabilities instead of a paragraph.

We re-asked 40 of the grading requests live, hours after the first pass, across two subjects. 120 answers moved by 1.1 and 1.4 points on average, 97 percent of them by 5 points or less, and two crossed our decision threshold (both on cells a neighboring bar had already settled). The one answer that moved 17 points was a vault-AUM cell where three trackers report three different numbers, and it stayed inside the “unresolved” band either way. Drift concentrates where the evidence is ambiguous, which is where you’d want it. For a grade that gets published with someone’s name on it, a number that reproduces is worth more than a clever rationale that doesn’t.

Everything a run touched is committed with it: the exact text graded, the exact questions asked, the exact evidence shown to the grader, and its exact answers. Anyone can replay a run offline with no API key and get the same scorecard byte for byte. We checked that too. Argue with our bars, which are public. The numbers follow from them.

The four subjects

We wanted a spread across fields, with nobody singled out.

Every is an AI media and software company predicting the thing it builds. Its list is the vaguest of the four, which is a feature, because it tests whether the method can handle “agents go mainstream.” Mike Ippolito’s 27 crypto calls are the most numeric (venture dollars, volume multiples, AUM ranges). Bob Doll of Crossmark Global has published ten numbered predictions every December for more than thirty years and, unusually, scores himself every quarter, which gives us a self-grade to put next to ours. Scott Galloway’s twelve headed predictions are the most widely read of the four and the most like the genre: a thesis per heading, a number here and there, and a World Cup wish at the end.

None of them asked for this. All four wrote things down in public, which is braver than most.

What the scorecards say

Bob Doll

Ten predictions, which split into 14 checkable claims (three of them are the midterms, which aren’t until November 3). As of October 24: 21 percent met at the strict bar, 29 percent at standard, 36 percent at lenient. Generosity gap: 14 points. Five claims are unresolved at the strict bar because the year isn’t over or the data doesn’t exist.

The hits are the macro ones. Inflation stayed sticky (core PCE 3.0 percent in August, exactly where it was in December). Credit spreads widened (high yield 59 basis points wider, investment grade 10, which is precisely the strict bar’s floor). The forward P/E fell from 22 to 19. All three met every bar.

The misses are the market ones, and they aren’t close. He said earnings would fall short of a 14 percent consensus; FactSet has 2026 earnings growth at 32.8 percent. He said the 10-year would trade “primarily between the high 3s and the mid 4s”; it closed above 5 percent on 22 days and topped out at 5.31. He said stocks would fail to advance by double digits; the S&P 500 is up 11.4 percent on price with ten weeks left. His own October scorecard calls the yield prediction “half correct” and counts it as too soon to call. Our strict bar held him to his own band and calls it missed. So does our lenient bar, because 5 percent is “mid 4s” on no reading I can find.

The sectors call is the one where the bars disagree most: met at lenient (his favored three beat his disfavoured three on average by nine points), missed at strict (communication services and financials both trail materials), and unresolved at standard because two data sources disagree about October’s rotation. That cell is the method working as intended. A single grade would have hidden all of it.

His self-grade, as of October 5: five heading in the right direction, zero wrong, five too soon. Ours, at the standard bar: four met, five missed, five unresolved. The difference is almost entirely the word “wrong.”

Every

Every’s predictions live in a December newsletter, a holiday podcast with CEO Dan Shipper and COO Brandon Gell, and a January episode with Reid Hoffman. 60 claims came out, the classifier ranked them, and 22 were graded (20 of Every’s own plus two of Hoffman’s, pinned as guest claims for a reason that will be obvious in a moment). As of October 24: 4 percent met at the strict bar, 32 percent at standard, 55 percent at lenient. Generosity gap: 50 points. 16 of 22 are unresolved at the strict bar.

The one claim met at every bar is Shipper’s: “There’s a whole class of highly creative, highly visual people who have been held back from making full experiences because they can’t code. Now they’re going to be able to code.” The AI in Design Report surveyed more than 900 designers in October and half of them had pushed AI-generated code to production. Figma’s report has designers participating in development up from 21 to 41 percent. Lovable passed 60 million projects. He called that one.

The pair we pinned: Shipper said the coding-agent leader most likely to stumble was Cursor, “caught between the IDE paradigm and the Claude Code paradigm.” His guest Reid Hoffman predicted the opposite: “I don’t think any of the horses that are in the race will particularly stumble.” In June, SpaceX acquired Cursor for $60 billion in stock, at about a $4 billion run rate. Nobody would call that a stumble, so Shipper’s strict and standard bars aren’t met. It’s also no longer a horse in the race, so Hoffman’s strict bar isn’t met either. Shipper’s lenient bar IS met, because the reporting around the deal describes exactly the competitive squeeze he named. Neither of them had a bar for “gets bought by a rocket company,” and neither did we. That’s what the unresolved column is for.

“2026 is when agents go mainstream” misses at strict and standard. Microsoft shipped agent features by default across Word, Excel and PowerPoint in April, which is a lot of seats. But no survey shows a majority of adults or workers using an agent, Gemini’s Android agent is still a limited beta, and Apple’s Siri features slipped to iOS 26.5 or 27. Mainstream inside Office, maybe. Mainstream for everyone, no.

Two of Every’s claims were about Every. “I think we’ll hire next year, but I don’t think we’re going to hire a ton” misses at strict: the company went from about 20 people to about 30, and the podcast episode that says so is titled “We automated everything with AI and tripled our headcount.” The unified memory layer across its apps is unresolved. Spiral has memory through its API (“not inside the app yet,” in Every’s words), which is progress toward the thing, and the bar asked for the thing.

The rest of the ladder reads like a map of what can and can’t be measured in October. Agents in daily operations at leading firms: met at standard (BNY runs 134 “digital employees,” Goldman rolled Devin out to its engineers, McKinsey has large firms scaling agents from 27 to 40 percent), unresolved at strict because the finance-specific clause needs more than one bank. Layoffs blamed on AI: met at standard (Challenger counts 101,743 AI-attributed cuts through June; Block, Snap and Atlassian said so in their own words). Shipper’s follow-up, that most of those layoffs aren’t really about AI, is met at lenient (the Yale Budget Lab called AI “a convenient scapegoat,” and Sam Altman used the phrase “AI washing”) but unresolved at standard, because nobody has published an analysis concluding that most of them were something else. Hoffman’s “much more negative” discourse: met at standard on Gallup’s “more harm than good” share jumping from 31 to 39 percent, unresolved at strict because Pew’s headline number moved only two points against the five we asked for. Gell’s midterm deepfakes: met at standard on three documented deepfake ads from national Republican committees, unresolved at strict because nobody publishes the reach figures our bar wanted. The mass-market robotic pet: unresolved everywhere. Casio’s Moflin has sold about 20,000 units since late 2024 and Sony stopped selling Aibo in Japan.

Every hasn’t revisited its list. Shipper made a separate set of predictions on Lenny’s Podcast in May, and those haven’t been revisited either.

Mike Ippolito

27 predictions in a New Year’s Day thread, which split into 46 checkable claims. The classifier ranked them by specificity and boldness and the top 20 were graded. As of October 24: 10 percent met at the strict bar, 35 percent at standard, 45 percent at lenient. Generosity gap: 35 points. Seven claims are unresolved at the strict bar.

Two calls were met at every bar, and both had a number in them. “Prediction markets will continue to succeed in 2026. Expect 2X from this year’s levels”: Kalshi and Polymarket did about $44 billion in all of 2025, and June through August of 2026 alone beat that. “REV will bottom around mid 2026, and start recovering into year end”: Solana’s real economic value bottomed at $14.3 million in June and was $30.1 million by September, per Blockworks’ own research arm, which is a nice thing to get right about your own company’s data.

Two calls missed at every bar, and both were about sentiment. “Ethereum will completely reverse sentiment and have its best year since 2021”: ETH is down 21 percent for the year, worse than 2025 and worse than bitcoin. “ZEC was a one off and will have a bad 2026”: Zcash is up 129 percent and crossed $1,000 in September after Grayscale launched a product on it. (Full disclosure: Rick is a Zcash enthusiast and have written about it. The bars were written before anyone looked at a price chart, and you can read them.)

The interesting cells are the ones where the bars split, and his list has more of them than Doll’s because more of his claims are made of adjectives. “Base will begin to come under some amount of scrutiny… the content coin push will fail”: the content coins did fail (Zora volume down 99.8 percent from its peak, and Jesse Pollak said “I was wrong”), so standard and lenient are met. But Base is still the largest L2 by value secured, so the strict reading, which required it to lose a category, isn’t. “Vault AUM will grow… from ~$5B today to $15-20B”: the trackers disagree with each other ($7 billion on Blockworks’ dashboard, $11 billion on vaults.fyi), none of them reach $15 billion, so strict misses and lenient is met. “Saylor’s reputation will take a hit”: Strategy’s premium to its bitcoin fell below 1.0 for the first time in June and Fortune ran “probably sell some Bitcoin to fund a dividend,” so standard is met. The stock has outperformed bitcoin this year, though, so the strict bar, which asked for that too, isn’t.

Three of the unresolved cells are instructive. “VC investment will be anemic. Expect a slightly diminished figure from 2025’s $25B, $12-$18B” is running at about $20 billion annualised, outside his range. But his $25 billion baseline for 2025 isn’t what Galaxy Research reports ($20.3 billion), and whether 2026 is “diminished” depends entirely on which baseline you accept. The grader called it a conflict and so do we. “Some sports betting apps will [succeed]” is met in spirit (DraftKings is doing $1.3 billion annualised) but DraftKings launched in December 2025, so a bar that says “launched in 2026” can’t be met by it. That’s a lesson about writing bars, and it’s on us. And “equity perps… <5% of perp volume by EOY” is probably right (about 3 percent, by our arithmetic across two sources) but nobody publishes the ratio directly, and the grader, correctly, wouldn’t take a number we computed as a number someone reported.

He hasn’t graded his own 2026 list yet. On New Year’s Eve he scored 2025 at 21 of 27 on a podcast.

Scott Galloway

Twelve headed predictions plus the Scotland line, which split into 15 claims. As of October 24: 0 percent met at the strict bar, 27 percent at standard, 47 percent at lenient. Generosity gap: 47 points. Eight of 15 are unresolved at the strict bar, and the strict row is empty because almost every heading on his list is a thesis rather than a number.

The standard-bar hits are the big-picture calls. “Space: the next big thing”: SpaceX priced its IPO at $1.75 trillion in June, raised $75 billion and closed its first day up 19 percent. “Vice of the year: prediction markets”: June 2026 alone did as much volume as all of 2025, New York sued Kalshi in July and Polymarket in September using problem-gambling language, and DraftKings and Flutter are down 44 and 65 percent for the year. “Synthetic relationships take center stage”: twelve states enacted companion-chatbot laws in the first half of the year and Character.AI and Google settled. Standard met, strict missed, because there’s still no federal law. “The ‘college is dead’ narrative collapses”: spring undergraduate enrollment rose 1.3 percent, per the Clearinghouse.

The misses are specific, which is why they’re misses. “Big Tech pick: Amazon” is up 7.4 percent against the S&P 500’s 12.3. It beats the equal-weighted Magnificent Seven, which have had a rough year, but our bars asked for the index too. “AI stocks correct”: the Nasdaq-100 fell 11 percent from its June high in July and recovered, so lenient is met and the 20 percent correction he implied isn’t. The recession he said would follow didn’t arrive: two positive quarters, unemployment at 4.2 percent, the Sahm rule at zero. “Best investment you don’t have access to: TikTok U.S.” carries a number that was wrong before the year began. The deal closed in January at about $14 billion, not the $28 billion he wrote, and Oracle carries its 15 percent stake at about $2 billion. Our bars were anchored to his figure, so the cells read missed, and the honest footnote is that no later valuation has tested either number. And Scotland went out in the group stage, which he saw coming (“Hey, I can dream”).

The unresolved cells are where the year hasn’t finished. OpenAI lost the enterprise lead to Anthropic by Menlo Ventures’ count (40 to 27 percent, though that report is from December, so it predates the year it’s being used to grade), and nobody publishes Nvidia’s accelerator share in a form a bar can use. Waymo is doing more than 500,000 paid rides a week across 14 markets, so “dominates” is met at lenient, but the doubling the standard bar asked for is unproven. Tesla’s official Optimus count for external customers is zero, which meets the standard bar for “humanoid robots are the self-driving cars of 2015,” while Unitree’s roughly 5,900 shipments break the strict one. Colbert’s show ended in May and he’s co-writing a film; no podcast yet. YouTube is the largest single distributor of TV time at 14.2 percent but still smaller than the subscription streamers combined, so Hollywood’s meteor is a lenient-bar story for now.

His mid-year check-in is behind the Prof G+ paywall. Nothing outside it grades the list.

What the gap says

The pattern is the one you’d guess, with a size you might not. Doll’s ten predictions are mostly numbers (a GDP rate, a yield band, a consensus figure, a seat count) and his generosity gap is 14 points. Ippolito’s 27 mix numbers with stories, and his gap is 35. Galloway’s twelve are theses with the occasional number, and his gap is 47. Every’s are almost all stories (agents go mainstream, a new kind of engineer, the year’s addiction), and its gap is 50.

Pool all 71 claims and split them by whether the classifier found a number, date or threshold in them before any evidence was seen. The 22 with one have a gap of 23 points. The 49 without have a gap of 45.

So far, so expected. What we didn’t expect is that the numeric claims also score LOWER at the standard and lenient bars (14 and 32 percent against 39 and 53). A number is a bar written down in advance. It makes the author easier to catch, and it protects the reader, which is the whole ask of this piece.

Look at the resolved-only rates to see what a single-bar grade hides. Every: at the lenient bar, every claim that could be resolved was met, 12 of 12. At the strict bar, one of six. Galloway: 70 percent at lenient, zero at strict. A generous grader and a hostile grader would both be telling the truth about the same list, and they’d publish 100 percent and 17 percent.

The second pattern: “unresolved” is a large category in October. Five of Doll’s 14, seven of Ippolito’s 20, eight of Galloway’s 15 and 16 of Every’s 22 can’t be called at the strict bar yet, because the year isn’t over, because the data is quarterly, or because nobody publishes the number the prediction implied. Year-end scorecards tend to round those to hits. We’d rather show them open, and we’d rather the next round of predictions came with the number that would settle them.

The third is about self-grading. Doll’s own October scorecard has zero wrong; ours has five missed at the standard bar. Ippolito scored his 2025 list at 78 percent; an outside reviewer scored it at 48. Galloway’s mid-year check-in is paywalled, and Every hasn’t graded itself. I don’t think any of that is dishonest. It’s what you get when the bar was never written down. The only fix I know is writing the bar down first.

What we learned building it

The critic check earned its keep. Before any evidence is gathered, the classifier is asked, for each bar, whether a skeptical critic would accept it as a fair reading of the claim. It flagged a strict bar that demanded a two-point margin the author never promised. It flagged a bar that asked whether Ethereum was the largest chain for tokenized assets, which it already was when the prediction was made, so the bar tested nothing. It flagged a lenient bar that had quietly dropped half of a two-part claim. All three were our mistakes and all three got fixed.

It also taught us not to chase it. The check’s acceptance moves with wording in ways that aren’t monotonic (a bar that named an index scored higher than the same bar in plain words), so we treat a flag as a prompt to look, explain the ones we keep, and never iterate a bar to raise the number.

One rule we learned the hard way: never write a lenient bar as a universal negative. “No 2026 inflation reading was at or below 2.3 percent” came back at a 0.65 probability of being met while the strict bar on the same claim was met at 0.73, because proving a negative needs complete data and the grader, correctly, wouldn’t take our word for it. That one is now handled by a rule rather than a rewrite. The bars are nested by construction, so a stricter bar met implies the looser ones met, and a looser bar missed implies the stricter ones missed. Three cells across the first two runs moved that way, and each one is labelled as implied in the report.

And a rule about the grader’s independence: it answers each bar on its own and doesn’t see the other two. That’s why the nesting rule has to exist, and it’s also why the probabilities are worth reading. A cell that says 0.68 met and 0.68 contradicted, which is what the sector call got at the standard bar, is two data sources disagreeing about October, reported honestly.

One more for anyone grading predictions from a web page. Every’s prediction post carries the line “Updated Jul 27, 2026.” Before trusting it, the tool fetched the Wayback Machine snapshot from January 5 and diffed it against the live page. Every difference was site furniture: a new sign-in banner, a like count that went from 14 to 33, a capitalised “Christmas Eve.” The predictions were word for word the same. Good news for Every, and a check that now runs on every source, so a list is graded as it was published, whatever got edited later.

Caveats

This run happened on October 24 with ten weeks left in the year. Every claim with a December horizon is marked provisional, and the midterm claims are simply open. Both the reasoning model and the classifier are text-only; a few claims hinge on things a screenshot would settle. Guest predictions made on a subject’s podcast are graded apart from the subject’s own. And our bars are ours. The review sheets are in the repo, every number traces to a bar and an evidence row, and if you think a bar is unfair, the right response is a pull request that changes it and regrades.

Before you publish your 2027 list

The season opens in about a week. Two suggestions.

If you’re about to publish predictions: write your three bars now, while you still mean them, and publish them with the list. A prediction with its bars attached is a different kind of promise.

If you published a 2026 list: run it on yourself and post the receipts. The tool is a skill that runs in Claude Code or Codex with no API keys (the classifier and the X API are optional upgrades). Give it a name, a year and the post where the predictions live. It does the rest and leaves a folder anyone can replay. It even drafts the thread.

In Claude Code:

/plugin marketplace add rickmanelius/called-it
/plugin install called-it@called-it
/called-it “Your Name” 2026 https://your.site/predictions-2026

In Codex and other Agent Skills hosts:

npx skills add rickmanelius/called-it -g -y.

Where everything lives

• The repo: github.com/rickmanelius/called-it. The four runs in this piece are in it, folder by folder, with the review sheets, the evidence and the grader’s answers.

• The gallery: rickmanelius.github.io/called-it has every scorecard as a page, plus the spreadsheets, the CSVs and the self-review thread drafts. [Placeholder link until the repo is public.]

• The 90-second explainer is embedded in this post.

If you grade someone (including yourself), open a pull request with the run folder and it goes in the gallery. If you think one of our bars is wrong, open a pull request that changes the bar. The regrade is one command… and I’d rather lose an argument about a bar in public than win one nobody can check.

Cybersecurity had one of its worst weeks ever.

You know things are rough when the NSA and Japanese government officials are telling people to take their websites offline. At times, the Internet Archive (yes, the literal backup of the internet) was down for almost half a day.

AI models are finding dozens of major security flaws daily, and humans are struggling to patch them in time. But before you panic… yes, it will get worse before it gets better. But it will get MUCH better. Once these critical bugs are fixed, we will have a much more secure and robust internet.

Smart TV Liposuction

Sorry for the terrible visual, but your Smart TV is likely full of spam apps (to serve you ads for your enjoyment, of course).

Well, the AI community has discovered you can hook up Claude and suck all that fat right out.

You’re welcome.

“I Can’t Find Talent”

The big (and justified) concern about AI is that it’s going to take everyone’s jobs. And there is plenty of evidence of people in tech who have been searching for over a year and just can’t get hired (see Mo’s recent thread).

But the flipside is happening as well.

Multiple colleagues are having trouble sifting through 1,000s of AI-generated and AI-enhanced resumes. Everyone seeking jobs is burying hiring managers in so much AI slop that it’s hard to find the right candidates.

Spoiler alert: human networking is working great, and more important than ever.

Explore Your Backyard: Part 2

In my last Grab Bad Edition, I shared why exploring your backyard matters. Since then, I took my own advice. Results:

• Reptile convention (this was shockingly awesome for the kids)

• College Tour (2-hour tour of the local campus).

• Weekend visit to Arkansas (first SEC football game).

• Makerspace (my kids are going to LOVE the 3D printer samples).

Now I have my Grok Bot scanning local news for any upcoming family-friendly events.

Friendly reminder: there’s so much happening in your backyard, you gotta look! And take the plunge.

Sales Is The Next Bottleneck

Coding used to be the hard part. Now it’s trivial, and many developers find themselves stumped by the following question: “What should I build next?”

When one thing becomes easy and abundant, something else becomes hard and scarce.

I believe that is sales, marketing, and BD. Now that the world is flooded with vibe-coded products and tools, actually finding and maintaining customers becomes the critical need.

Prediction: If you want to be invaluable in 2027, get better at sales.

Kids Are Best AI Skeptics

Parents fear kids will get sucked into using AI to replace thinking.

I bet this is happening, but I’m also observing the reverse.

My oldest daughter uses AI but rejects it as a primary creator. She uses it only as a sparring partner and refuses to let it write for her.

I’ve seen some of her peers outright reject or mock it. “Oh, that’s an AI song, bleh.”

In my experience, most people live like John Locke. We hate feeling constrained or restricted, especially when someone else imposes it. But what if constraints were not only not bad… but sometimes… good?

This is where the Playground Study illustration comes in. Let me try to explain it verbally, then share a visual to help it land stronger.

• Imagine two identical parks.

• They are both FULL of young children.

• The children are all running around playing games.

• Both parks are right next to a busy highway.

• The first park has NO fence between the grass and the highway.

• The second park has an impenetrable wall.

Socratic Questions

Now take an educated guess on these three questions:

1. Which park will have kids playing in the grass next to the highway?

2. Why?

3. What are the broader implications of this?

Unless the kids have a death wish OR negligent parents, the kids in park 1 will stay far from the highway, lest a car accidentally veer off OR they run too far into the road.

But this is never a problem in park #2, because the cars AND kids can never cross. A car hitting the wall would bounce off. A kid running to catch a fly ball would bump into the fence. In either situation, the harm to the kid is either eliminated or significantly reduced.

So What Are The Implications

A hard boundary (the fence) provides the safety kids need to explore the entire park.

So what’s the implication for your business?

Where are there situations where adding a hard rule actually resulted in MORE exploration, innovation, and testing of efficiency? Where did constraints lead to MORE overall freedom, play, and flow… and not less?

Juxtaposed, Visualized

Here’s the visual I created.

I hope it helps.

And if you’d like to know more about this study, there’s an excellent LinkedIn article by Andrew Horn that goes into more detail. I originally heard about this in a book (I wish I could remember which one), and if you know it, please let me know so I can cite it properly.

Two days ago, the same process took under 2 hours.

I know, I know. We’ve become numb to all these AI success story facts and figures. And in the grand scheme of things, 18 hours saved is a rounding error.

To be the bigger headline is…

I ACTUALLY ENJOYED DOING TAXES.

I know, it’s… crazy.

My personality type is not well suited for long, repetitive tasks. Categorizing 3,000 credit card and 1,000 bank entry statements is about as soul-sucking as it gets. Then there is wading through long questions from the CPA, sorting and searching through a bunch of files and systems, etc. It’s a lot of needle-in-the-haystack work across 5-8 systems.

So of course I file extensions every year to delay the suffering, but there is still suffering.

And Then There’s Grok Bot

Last year, AI still wasn’t up to the challenge for taxes IMHO. We were still pre-Opus 4.5, and the error rates and hallucinations were still too high to trust with something so important. So while AI might help with one piece of the puzzle, it was still a lot of work.

We are now well past that point.

It took about 30 minutes of setup to get all the connections and context in place. After that, it just cranked for 30-45 minutes, reported along the way with check-ins, and created a full packet with detailed answers and reports.

Here was my setup:

• New Grok Bot Agent: Company Taxes

• Connector 1: Company Google Drive (contains previous tax year docs)

• Connector 2: Company Email (contains all communications, PDF receipts)

• Connector 3: Xero (accounting software)

• Connector 4: Slack (for internal conversations)

• Context 1: Export of all Chase credit card and bank statements

• Context 2: CPA questions

And that was it.

Best of all, it was a smooth, unified experience. I just fired off the initial request and responded to a few categorization clarification questions. When the CPA had two follow-up questions, Grok Bot was able to pull from the 2021 and 2022 tax returns to find the answer for me.

Why Procrastination Worked

Even in March of this year, the AI models were powerful enough to get the job done, but would have required a lot more time and cajoling. And it’s the cajoling that I hate and resist.

By procrastinating with an extension, 6 months of AI hardness and model advancement addressed all those issues! I reaped the saved time and stress.

I’m expecting next year (April 2027) to be even easier. However, I plan on filing an extension to get the added benefit of an additional 6 months of AI advancement. Who knows? Maybe next October it’ll be two minutes and zero questions. That would be magic!

Basically, it summarized a book’s core concepts down to a 10-15 minute podcast that I could squeeze in during my morning walks. Life was super busy, and I was jealous of the “thought leaders” that somehow managed to read 100 new books a year. So I gave it a go for six months.

It felt… productive?

And yet, I can’t recall a single book from that list. NOT ONE!

Obviously, I needed a different approach. It hit me to try and follow these words of wisdom from Tim Ferriss (paraphrasing):

What if you tried the 180-degree opposite strategy for the next 48 hours?

Bingo. Here’s how that played out.

Two Decades Later: Strong Recollection

I was an atheist in college, but I was always willing to spar and debate on any topic.

So when Tony Cappaert organized and invited all the fraternity brothers at Phi Delta Theta to a bible study (I was the GRA), I said, why not?

Spoiler alert. It didn’t convert me to Christianity (I remained an atheist), but I became a HUGE fan of Pastor Don McClure’s sermon series. We worked through a 67-part series on The Sermon on the Mount. It was so detailed that Don would spend an entire sermon, sometimes 45 minutes, dissecting one sentence.

I listened to the entire thing end to end… twice.

I can still remember and cite not only the passages, but the stories he told, the examples he used, and the historical context behind it all.

It was the 180-degree opposite of Blinkist. It was full immersion and internalization of the message.

Back to Books

What does this have to do with business books?

Well, it’s the same thing you’re seeing with all the AI buzz on social media. People are running around trying tools and screaming about how great they are on LinkedIn… but when you press them to share details, they admit they usually stop after a superficial demo and don’t have much depth behind it.

So I decided to spend an entire year with ONE book. One of my 3 Words of 2026 was 10X, and so I have been reading Benjamin Hardy’s 10x Is Easier Than 2x. And I’ve been going 1-2 paragraphs at a time for 5-10 minutes a day.

But that’s just the reading. I’m also forcing myself to slow down in three important ways.

1. Writing: After reading, I force myself to write down 10 ideas that I could apply.

2. Committing: I select one thing that I can and will get done by 5 PM.

3. Action: I actually follow through!

See, that’s the hard part everyone skims and skips: the doing.

Results

It’s October 2nd, and I’ve only missed two days this entire year. And… I’m only midway through chapter 3 of the book! 😅

That’s ok, because unlike my time with Blinkist, I am actively internalizing this book every. single. day. And best of all, I’m seeing the changes happening in my life.

So maybe the better way to read a book is like the Bible. No, I’m not trying to convert you to Christianity! What I am suggesting is that you adopt a Focus and Followthrough strategy for books you really want to read, instead of just getting the dopamine hit of zipping through a hundred titles on Blinkist.

“Sent from my iPhone” at the bottom of every email was a genius move by Apple.

Not only was it free marketing (especially in the early days, when iPhones were expensive and novel), but it was an excuse in advance. Basically, “I’m on the go, so expect a SHORT reply. Please ignore spelling/grammar errors. Thanks!”

Now we need “Written with AI”, but at the top of ALL content written by AI.

It’s the same heads-up, but it implies something very different. “I’m in a hurry, so expect a LONG reply. Please ignore the em-dashes. Thanks!”

We can visualize this on a 2x2 grid of human FOCUS and output VOLUME.

Summarizing:

• Short and distracted: “Sent from my iPhone”

• Long and distracted: “Written with AI” (e.g., pasted with no/minimal editing)

• Long and focused: a memo someone actually thought through.

• Short and focused: the rarest box: essence only (e.g., a Naval Tweet).

AI made volume free (just add tokens!), but human attention is a finite resource.

So if you want your communication to make an impact, aim for short and focused.

Where I Stand

I’ve been a fan of this idea for a while.

• I wrote AI;DR to highlight my “You didn’t bother to edit it, so I won’t bother to read it” posture. Although I love Tobi Lütke’s slop grenade term even better.

• Humanizer is table stakes. Anything AI touches gets a humanizer pass, which catches the patterns that make text read like a model and fixes them.

• Substack made me write an official “How I Make This” statement. Here’s mine, word for word: While I use AI lightly at the end of my process (Grammarly for spelling, grammar, and style suggestions), I type the entire first draft by hand.

• At work, I lead with this whenever there is ambiguity. Example: “AI-generated first draft. Human edits: light.” This way my colleagues know whether it’s an early draft OR a final copy.

My Prediction

By Q4 2027, AI writing policies will become a standard at all companies.

This idea already exploded today.

Dylan Field posted Figma’s rule: if you use AI in your writing process, add “written with AI” at the top.

Varun Anand shared Clay’s company-wide AI writing policy, which has a FANTASTIC heuristic.

“More time should be spent writing a document than consuming it.”

And Ishan Goswami went furthest. Quoting Clay, he announced that Exa, an AI company, bans AI-written external comms outright: “If you don’t take the time to write it, why should anyone take the time to read it?”

That’s close to my AI;DR line with one word swapped. Exa says write, but I say edit. I’m fine with AI in the process, as long as a human thought about it BEFORE hitting send.

Finally…

A label won’t rescue slop (if you lie or try to pass off pure AI writing as your own). It’s just a respectful way to tell the reader which box you’re sending from, so they can decide how much of their time to give it.

• FDX (Forward Deployed Executives)

• AI as guardrails for humans.

• Banana Water (organic, of course!)

Lots of fun laughing and riffing on all things tech/AI. YouTube link below.

Dan Shipper reposted one of my favorite rant clips from Gary V from ~2008.

I remember watching this video on repeat shortly after I made the big leap to abandon my PhD. Why? Because I realized that writing scientific papers and working in a lab doing someone else’s research was making me miserable.

2. Get a YubiKey

Security hacks and breaches are going to get nuts with AI (it’s already happening). And if you keep your passwords in a software vault, you’ll eventually get hacked too.

Hardware passkeys like YubiKey are cheap, secure, and effective. Add them to any bank accounts (traditional and crypto), primary emails (e.g., Gmail), and social media accounts (e.g., X/Twitter).

Do it now. Thank me later.

3. Big Ass Calendar

You know the parenting mantra. “The Days Are Long, But the Years Are Short.”

My friend Bryan is an amazing dad whom I DM all the time. We constantly push ourselves to make “Golden Memories” with our kids… because it’s the highlight reel that they’ll remember long after our short time is gone.

The Big Ass Calendar is a great way to keep this idea front and center.

We got ours this year, and it’s tacked up in our garage right next to where I park. I look at it every time I leave for work and again when I get out of my car back home.

Highly recommend.

4. Tokenomics is Warping Our Economics

Make this make sense.

• Engineer: Salary $10K-$20K/month.

• APPROVED: AI token budget of $1-5K/month.

• DENIED: SaaS app costing $20/month.

LOL. Not every AI $ spent on AI tokens is a guaranteed ROI.

5. Book: Simplify

I’m only one chapter in Simplify, and it’s an OUTSTANDING read… potentially better than Dan Sullivan’s best seller 10x Is Easier Than 2x.

It’s one of those “stop right now and start applying” reads. And it mirrors what I wrote in The New AI Superpowers: Focus and Followthrough (although the Simplify authors did a MUCH better job).

6. The New EQ: Are You a Slop Grenade?

I wrote about AI;DR, but Slop Grenade is even better.

Why is this important for emotional intelligence? A lot of people think they are being helpful by prompting Claude and shipping the raw output straight into company communication channels.

But in an age of overwhelm, the most EQ skill is LIMITING how much mental load you’re imposing on others.

So don’t hurl slop grenades for others to deal with. Get specific and concise. If they ask for more information, provide it.

7. Explore Your Backyard

A colleague posted that they are trying to be more active by walking outside, and they were thrilled to be reminded how beautiful the scenery is around their house.

We all have little treasures all around us (parks, walks, sunsets, etc).

Put down the screens. Go analog. Search.

Today I’m going to get a library card (FINALLY). It’s literally two blocks away from my office (and I drive past it every day). Not today… Today I explore some old books in my backyard.

Next Up: Jev

The AI world is on fire with this new primitive by Typesafe (named “Jev” after Jevon’s Paradox). I have been testing a lot, and I plan on sharing a few articles in the next week.

A few years later, I met another Techstars Mentor who was repeating this playbook in Houston and Oklahoma. He described how Houston had almost no startup scene, but eventually followed the Boulder/Denver trajectory. They were just 10 years behind that curve. And Oklahoma was another 10 years behind Houston. Each city would get there, but it would just be later.

I arrived in OKC in 2020, and I’m sad to report that this timeline will take much longer than I expected… if OKC is ever going to get there at all.

What Have You Done For Me Lately?

Truth be told, I had aspirations of being part of the growth of the startup scene in OKC, but my attention was always with my clients on the coasts. So I stood by, hoping someone else would lead the charge.

While local VCs and accelerators have made valiant efforts, they still haven’t taken root here as much as I had hoped by now.

So I started taking matters into my own hands, at least in one aspect.

After my mom died in April, I was reminded of the importance of prioritizing community. It was both soul-filling and her superpower. Her best friend of nearly 70 years (they met in preschool at age 4) made me appreciate my two best friends of 44 years (we all met when we were 1).

So I started a local Meetup here in Edmond. It’s nothing earth-shattering, but it already feels like it’s having a profound impact on the 8-9 people who have shown up.

Why?

People Are Starved for IRL Connection

Since the world went full-remote during COVID, I think we haven’t come to terms with how critical local, in-real-life connection is for most people.

At last Wednesday’s meetup, two new folks shared how isolated they have felt in their profession. One gentleman noted he’s willing to travel long distances from out of town for one local meetup of about 5-6 people, to have colleagues to interact with.

In the interim, they are all isolated in their home offices.

My Commitment

Brad said people should make an earnest 20-year commitment if they’re serious about building community.

I’ve lived here for 6 years, and my wife and I plan on being in Oklahoma until both our kids go to college (at least 11 years from now). So while it’s not a full 20, and I gave it a half-hearted start, I’m determined to take it more seriously from here.

I need it.

And others need it.

Otherwise, we’ll all keep feeling like strangers in a crowd.

And if all it takes to get started is a Meetup invite and a couple of pizzas, it’s a no-brainer.

You’ve probably experienced this when a site like Facebook or GitHub goes down, and the world panics while everyone watches the status page… frantically refreshing so they can use it again.

A short MTTR is ideal. A long MTTR can be a disaster.

The Other MTTR

Let’s face it. AI is going to cause a lot of crashes. And not just servers in the cloud, but career paths and entire org charts.

People accustomed to fitting into nice, neat roles with a clear 5-10 year plan are going to feel uneasy as their roles and responsibilities change quarterly (instead of alongside a promotion every 2-4 years).

The big question (to paraphrase 90s band Chumbawamba)… when these people get knocked down, will they get up again?

And how fast?

What will their mean time to reinvention be? What’s their MTTR?

A few years?

A few quarters?

Or can they join a new team and be valuable in under a month? A week? Immediately?

From PhD to a No-Nothing Developer

After spending 10 years getting a BS and PhD at MIT, many were shocked that I “threw it all away” and became a backend Drupal developer. “Developer” is too strong a word, because I could barely understand the complex UI, let alone write useful PHP code or database queries.

I went from the world expert on van der Waals forces of carbon nanotube systems to… the bottom of the ladder.

It was not easy. And it took me many, many years to claw my way back up to a normal income level and a position among my peers.

But then? I kept on rising the ranks, eventually becoming one of the leaders in the space.

And Another One

So did I stay in Drupal forever? Hell no.

I threw that all away and started again. And then again. And then again.

Depending on how you count, I’m into my 6th startup or business. And every time I jump into a new space… yes, I bring a lot of transferable skills and experience. However, I also have to spend A LOT of time drinking from the firehose trying to catch up to everyone else.

Now I can join a company and start figuring shit out on day 1. Will I be perfect? Hell no. But I’ve survived through many near-death experiences in business, so I’m largely unfazed by my “normal” day-to-day disasters.

Continuous Education to Continuous Reinvention

With AI, we no longer get to choose when to leap into reinvention.

The world is changing so fast that many of us will be forced into continuous adaptation.

Some will succeed in this new paradigm. Many won’t.

Those with slow MTTR (requiring years to reinvent themselves) will learn things just in time for them to be out of date again. How awful! They will forever be at the back of the line.

Those with a fast MTTR? They have the highest probability of adapting to and benefiting from every new wave of change. On net, they will be the leaders at the forefront.

So… what do you do?

Change is coming, so embrace it.

If you haven’t built up a high tolerance for change, start building this muscle NOW.

Start introducing more change into your life now. New hobbies. New experiences. Read books that challenge your thinking. Visit new places. Question all of your existing strategies. Try new AI tools.

When Netflix wanted to improve their MTTR, they would actively introduce “Chaos Monkeys” into their infrastructure to test how resilient it could be. Start shutting down servers. Disconnecting network lines. How fast would the system detect and come back online? If it were slow and fragile, you’d design the system to be stronger next time.

Add a dash of chaos monkey into your life. Or as Scott Adams would say, if you’re bored, you’re not embarrassing yourself enough. Embarrassment can ONLY happen when you put yourself way out of your comfort zone. No risk of embarrassment means you’re not taking risks.

Yeah, I Yolo’d This Article

I wrote this article in a single-word vomit while waiting in the hot parking lot outside my daughter’s school dance.

I could (and should) have spent more time refining it. But whatever.

I had been sitting on this idea for a week, and it ALMOST died on the vine.

But I know it’ll connect with at least a handful of readers. And it would be a disservice not to share it.

This is true for you too.

Get out there more. Do hard things. Do weird things. Do half-baked things.

It’s ok. The world will judge you anyway. At least make it an interesting life to judge!

PS. Thank you, Grammarly, for the rapid spell check. Only 40 this time! Giddyup 🤠

AI made building cheap. Starting around November 2025 w/Opus 4.5:

• The things that used to take a developer two weeks started taking an afternoon.

• And the thing that took an afternoon started taking ten minutes.

Building was cheap, but managing all the output got expensive (at least without a plan).

Many AI users are sitting on piles of prompts they wrote once, can’t find, and never ran again. No habit formed. No system built.

The AI “prompt libraries” were supposed to be a part of a well-oiled machine, but they now look like an episode of Hoarders.

Hundreds of work-in-progress prompts all stuck in purgatory. This is a travesty, because the fix is a little more time in the right places. We need to follow the patterns and playbooks by those who have already solved this.

Kieran Klaassen at Every wanted a codebase that got better with every feature shipped. I want the same thing for workflows: every lap around the loop should make the next workflow cheaper, easier, and better.

This post describes these loops. One version for a soloist, one for a team, and the honest reason most of the effort occurs before and after the build steps.

The Inspiration

Orie Steele put together a deck he calls vibeslop. It’s an entire software development lifecycle for the AI era: Pitch, Sketch, spec-kit, Ship, Score, with the named frameworks at every stage (e.g., Jobs to Be Done, the Hook Model, Double Diamond, etc.).

It’s serious work, and I mean that as a compliment for an article title, as a joke. It’s also built for engineers, not the rest of the organization.

I wanted the version that fits on a whiteboard and works for someone who has never opened a repo, because it’s the work I do day in and day out with my clients.

Why Me

I’m an engineer and repeat startup founder who became an executive, so I live on both sides of this loop. I wrote about that split as the Forward Deployed Executive, so I won’t re-explain it here.

What I’m about to share is a combination of solutions I’ve applied across a dozen clients across as many industries. I’ve already published articles on various facets of this, but I wanted to bring it all together into one Sesame Street Simple map.

And while there are already multiple takes on AI-native software development lifecycles (SDLC), the reality is that there hasn’t been as much focus and success in mapping this out for the non-engineering portions of an organization (which can represent a large majority). Also, AI is still evolving rapidly, so best practices will likely continue to evolve across all types of organizations.

My bet: learning to run a loop like this is the skill that survives the next model release.

The Value

Here’s what you get for running a loop instead of a one-off. Four things, and each one depends on the one before it.

Quality. You stop one-shotting. The second and third passes are where the good stuff shows up, because follow-through is the actual superpower.

Consistency. One-off, then prompt, then skill, then automation. Every step down that ladder narrows the variance in what comes out.

Automation. You can’t automate what you don’t trust (hence the need to master quality and consistency first).

Compounding. Time saved and improved outcomes free up resources to go around again. That’s the whole game.

Again, builds are cheap. Given that, expect to spend 90-95% of your time into the workflow design and the structure around the build. That’s the part that pays back on every future lap.

The One-Off Strategy

You know this one. Someone had an idea, typed it into a chat window, and it worked. The demo impressed everyone at the meeting.

Then nobody ran it again.

The prompt lives in a Slack thread, or a Notes file, or in the head of the one person who wrote it. Next time the job comes up, it starts from zero.

The paradox is that it’s the packaging and automation where the real, long-term value comes from in any of this work… and most people stop 10 yards from the finish line.

The Solo Loop

Five stages for each lap around the loop.

• Prioritize. Rank what’s worth doing against a rubric (I score Value, Effort, Reach, and Recurring). Ignore the rest.

• Organize. Gather what the job needs before you open a tool: the knowledge, context, APIs, and systems it touches.

• Build. Proof of concept first. Once it’s approved, polish it into the version someone will actually use this week.

• Document. Write down what good looks like, in cases you trust, plus the SOP for running it. Plan for organization and recall. Your future self will thank you.

• Automate. The goal: you can one-click trigger an entire process: bonus, the work fires on a schedule or a trigger.

The loop closes at the bottom, and that’s the part I care about. Time saved and skills upleveled go straight into the next thing on the ranked list. You never start from zero twice.

Expectation control: this is where the hours go. Build is the short bar. Organize and Document are the tall ones because planning the job and writing down what good looks like make up most of the work. Skip those two to “just build it” and you get a one-off with extra steps.

The dopamine trap of moving fast and furious ends with no long-term durable workflow improvements. Resist it. Finish the loop!

The Team Loop

You’ve heard the proverb: if you want to go fast, go alone. If you want to go far, go together.

There’s a big difference when you go from solo mode to teams. Now you gotta prioritize the right things, ensure proper change management and training, etc. If you don’t, your amazing workflows stay on an island with you instead of leveraging the entire team or organization. Resist the temptation to skip this!

You can slice and dice this further, but I like to think of this as 8 steps:

• Idea! Get the ideas out of people’s heads, write them down, and keep a shared list you can draw from as needed.

• Selection. Rank the list using an effort-and-value rubric, and name the outcome before anyone opens a tool. This is the kill box.

• Mapping. Sketch the UX, who’s responsible for what (RACI), and paint what done looks like. If you uncover many problems, update the priority scores and send it back.

• Build. PoC, then MVP, then production. Don’t skip the ugly middle.

• Benchmark. Does it work beyond a “Hello World” demo? Test against the end-users’ golden set until it’s delivering value.

• Automation. Reduce as much friction as possible to make remembering, finding, and triggering easier again and again.

• Sharing. Ensure everyone knows where to look and how to use it, and that they can successfully achieve the outcome.

• Actualization. Leadership, culture, collaboration. This is where feedback and new ideas come from, and then you go around again.

Two arrows point backward, and they will save you tons of headaches if you stay disciplined and use them.

• Mapping sends work back to Selection when the map doesn’t survive contact with the real job.

• Benchmark sends work back to Build when the real work fails.

Past Benchmark, nothing comes back. That’s on purpose. Benchmark sits between Build and Automation, so passing it is what earns a trigger. Every other framework I looked at puts evals inside Build (to gate the merge) or after ship (to measure the outcome). This one puts them in front of autopilot. You don’t get to automate something that can’t pass real work.

Ideally, you keep forward momentum wherever possible. And of course, don’t be too rigid. This set of constraints is meant to help you achieve full actualization of your efforts! Not to straitjacket you into sitting still. But lots of frenetic motion without a plan can be even worse. Find the balance point in your organization and apply the parts that work for you.

The Future Phase

I want to go much deeper on this, but only if this topic resonates with enough people. No matter what, I plan on using it in my AI consulting practice for the foreseeable future.

The longer follow-up would include:

• The frameworks at each stage (the way Orie did it in his article)

• Skill files per step so an agent can run the loop with you.

If this post gets at least 10 comments from people saying they’d be interested, I’ll proceed.

Follow-up

• Give this to your agent and have it draft your own version of the loop. Then argue with it (and make it yours).

• Subscribe to my newsletter so you get a notification.

• Skills are coming, one per step. Some are already on github.com/rickmanelius/skills.

• Share it if it helped. If this gets 10 comments, I’ll know it matters enough to build the long one.

• Comment “follow-up” and I’ll DM you when the long version is published.

PS. Here’s a HyperFrames explainer for those that prefer another angle at this.

Me: “No problem! I’ll do it myself.”

But my 5-year-old had other plans. She’s known for asking a question every 15 seconds. And the few times she was silent, the AC kicked on and started humming in the background.

I needed 90 seconds of clean audio. But after 30 minutes of interruptions, I rage quit and sent the AI voiceover instead.

The HyperFrames Background

I LOVE HYPERFRAMES.

Distilling a complex topic down to a minute-long explainer video has been a game-changer. They are even better with voiceovers because they can add storytelling, context, and emotion.

However, the two ways of adding this to a HyperFrames video are both… not ideal.

1. Stock AI. They are precise, but people hate them because they sound fake and generic.

2. Record your voice. Great if you have a good mic, a good voice, and a moment of silence.

Clients despise #1. I despise #2. There had to be a better way…

The Podcast Room Tax

Here is the frame I keep coming back to.

Every voiceover charges you the same three things:

1. A quiet room

2. A good mic

3. A clean take

Call it the podcast room tax. And it’s getting more expensive every day given that I’m generating more and more video content.

The solution? What if you could pay the tax once and benefit forever? This is what I explored this morning. I wanted to record my own voice and then feed it into AI to reproduce my tone, cadence, and voice.

The initial result? Not bad.

Yes, it needs some tweaking… but it’s getting there.

Getting Started with xAI Custom Voice

Here’s how I did it.

1. Go to console.x.ai

2. Open Custom Voices

3. Record your voice live in the browser by reading the passages.

4. Copy the ID for your voice template and hand it to your agent.

Now I know what people are thinking. But can’t everyone use this? The constraint is the consent model. IMHO it is the right one. There is no “paste in a podcast episode and become that person” path. You must show up, talk, and use your own voice.

What surprised me

• My clone talks faster than the stock narrator, 3.13 words per second against 2.33, so a script written for the old voice finishes early.

• TTS also reads your typos aloud (so be sure to spell-check!).

• It still feels too “annoucer-y”, especially in the beginning of the recordings. But as they keep playing, I feel it sounds more and more like me.

• It’s cheap! Something like $5 for an hour of audio is way cheaper than I expected.

• It’s fast! From request to done takes barely longer than the audio itself.

• It’s addictive! I’ve already generated 5-6 different audio takes now that I have the skill baked and ready.

My first use of HyperFrames wasn’t perfect, but I’m sharing it as is just to timestamp my first experience. I expect that this will get better with each iteration to the point where you won’t be able to tell at all.

15 minutes could save you 15% or more on car insurance.

Well, what if 15 minutes could net you $ 1,500 in lost cash?

That’s what happened with Will Tjernlund (ok, it was really $1400).

So… how? It’s a website called “Missing Money.”

I know, I know. Gotta be a scam, right? I can feel people’s BS meters already being triggered. I was skeptical too, but the referral came from my uncle, who uses the site every year to find unclaimed property he’s owed.

I also spent about 30 minutes searching around to see if the website had ever been flagged. However, the real proof is that I used it myself and collected checks. Results below:

My Results: $450

Since I had never used the website, I had about 10 entries, dating back to 2010 in Colorado and 2005 in Boston. All told, we found $250 for me and $200 for my wife.

I was owed money from companies ranging from Epson printers to old Chase bank accounts. Hell, apparently I never collected a refund from my old Massachusetts E-ZPass.

The Meta Lesson: Hidden Resources

Some people may search for “Missing Money” and find a few hundred dollars. Some will find $0.

However, I bet there are several situations in your life where you can find these hidden cash-out opportunities hiding in plain sight.

One of my relatives has a bike he bought in 2010. It’s barely used and still worth at least $2,000. It’s currently hanging in his garage, and my guess is there is ZERO chance he’ll ever use it again (at his age and physical condition).

Why not sell it and use the funds for something else?

My wife and I do a yearly Facebook Marketplace purge of stuff our kids can no longer use. Some years are $50, and others we make selling $500 or more of stuff we no longer use.

Or does someone still owe you money?

The meta lesson is that if you look hard enough, there are missing money opportunities.

My recent experience with Missing Money was another good nudge for me to keep my eyes open.

PS

You know it’s a government operated website when the design is this old-school. LOL

1. Clearing my desk

2. Writing a small TODO item on a bright 3x5 index card

3. Placing this card on my keyboard so I can’t miss it.

Why?

To overcome the resistance of getting the workday started.

Steven Kotler says most people waste 2-10 minutes “dithering” between tasks. They seem to be gathering up the willpower to jump in and get the next thing done.

I used to be that way when I showed up at my office.

I’d plunk down, open my laptop, and immediately waste 5-10 minutes scrolling through Twitter before deciding what to work on first and getting to it.

But now, I leave myself an index card; a momentum matchstick.

I sit down, see the TODO I can knock out in 1 minute, and jump right in.

Why The Matchstick Metaphor?

A matchstick is a single strike that can ignite a fire.

Momentum and flow work the same way.

Attention drives focus.

Focus drives flow.

Just finishing one task can lead to the next, and then the next.

I love stretching a good metaphor. Let’s go nuts here.

1. You strike a momentum matchstick.

2. To generate a fire of flow.

3. To burn through your TODO list.

You may groan, but you will remember this (you’re welcome). And enjoy an AI visual:

Personal Results

Lately, I’ve found myself immediately dropping into a 1-2 hour session, knocking out 5-20 things off my TODO list.

I encourage you to give it a try.

• Leave yourself the smallest possible action.

• Start it as soon as you see it.

• Finish and repeat.

Don’t forget to leave yourself another card for the next day.

I get asked this question a lot. For the past 15 months, I’ve been leading AI training and team transformations, and building, building, building (my current interest is in open-weight models on local AI hardware). As a result, I tend to find and share as many resources as I can with my colleagues. And while there are great resources everywhere… the answer is simple.

Me: “You need to be on X.”

This is where I pause and wait for revulsion (from at least 50% of my colleagues).

X: “Ugh. But I hate Elon so much. I’ll never use X.”

And that’s the crux of the issue… one’s choice to harbor hatred towards a person/organization is not without career consequences. But before we dig in…

I Love a Principled Stand

When my mom died 4 months ago, there was a flurry of conversations about her on Facebook.

And I refused to create a new account to participate.

It wasn’t an easy decision to delete my account in the first place. MIT was one of the first campuses to get access, and I had photos and connections that span back to 2005. However, I never trusted Mark Zuckerberg. Over the years, there have been dozens of ethical violations involving user data. It was clear to me these weren’t accidents, but it was just who he was. So in 2021, I deleted my account.

This decision inadvertently severed many friendships.

The thing about a platform with 2 billion active users is that it’s the center of the social conversation. And if you’re not there, well… sorry. Most of these people are no longer easily reachable. Enjoy the FOMO.

It sucked not being able to connect with people sharing memories about my mom. That had to wait until the funeral when I could see them all IRL.

Still, while I’m not at peace with her loss, I am at peace with my decision to never use a Meta platform (Facebook, Instagram, and WhatsApp).

I love a principled stand, and I accept the (social) consequences of that decision.

Even Zuckerberg Capitulates to Elon

First, let’s separate objective facts from subjective opinions and feelings.

I’m sure you’re familiar with the moral alignment chart (hat tip to D&D). It’s a 3x3 grid that ranks a person on two scales: good vs. evil and lawful vs. chaotic.

We like to think of a person as living in a single fixed square. However, Elon is one of those eccentric individuals who can shapeshift based on context.

There are many sides of Elon. I think I know where he stands most of the time (e.g., chaotic neutral with a dash of chaotic good), but many disagree (e.g., various forms of evil). That’s fine; no matter where you stand, know that many others are locked into another perspective. No need to get upset over it, but…

The fact remains, X is THE definitive place to keep up with the AI conversation.

This is the reason Boris Cherny became active on X just as his Claude Code project was taking off.

Since he joined and started replying all day, every day, both his personal profile AND the project have rocketed in popularity and adoption.

9 months later, it’s become common for other developers and researchers to be pushed onto X. It makes sense, because this is where the customers live and where the conversations are happening.

Hell… EVEN MARK ZUCKERBERG, the CEO of a 2-billion-person social media platform, makes announcements of his new AI models on X instead of Threads for Facebook.

There’s Alpha in What Others Hate

When GitHub had an outage this week (again), Cursor used it as an opportunity to launch their new code hosting platform.

I was excited to hear this because a few months ago GitHub had one of the worst months in SaaS history, with uptime dipping below 90%. This is unheard in enterprise cloud software, where SLAs of can range from 99.9% to 99.99% uptime (with contractual penalties).

I expected my technical colleagues to respond with something akin to “cool” or “FINALLY” given they were experiencing the brunt of these outages. Competition leads to innovation, which should lead to a better ecosystem.

Instead, the responses ranged from (at best) apathy to (at worst) total rejection of the idea. How could an Elon-owned company build anything close to GitHub?!?

Their choice to hate Elon as a person/founder/entrepreneur prevents them from even evaluating what could be a superior product. This is putting them (and their team/company) at a potential disadvantage. After all, this is the guy who has built companies that can catch rockets, implant chips into people’s brains, and self-drive cars at scale. Hosting software should be a cakewalk compared to the other 3.

And Again, With Grok Bot

Lots of AI people on X are sharing positive takes about Grok Bot.

Ryan Carson (super early on AI agents, loops, etc)

Claire Vo (founder of ChatPRD and an overall exceptional voice in the community)

Or Investor Gavin Baker, who feels this is the next “Claude Code moment.”

Even non-technical folks are raving about how Grok Bot meets the “just works” bar.

But here’s the thing: many technical colleagues have:

1. Ruled out Elon (because hate, evil, politics, etc.).

2. Therefore, rule out Grok (because it can’t be as good as Anthropic/OpenAI)

3. Therefore, rule out Cursor (because Elon purchased it, defaults to Grok)

4. Therefore, rule out Grok Bot (because they can’t even get here because points 1-3).

And so it goes.

Maybe they are happy with OpenClaw (X sentiment is that it’s now 3rd, behind Hermes and Grok Bot) and GitHub (even though the outages persist), but many willing to engage on X think otherwise.

They were willing to try, experience firsthand, and make an informed decision.

I think that’s the smarter, wiser approach.

This is Not About Elon

Love him. Hate him. I don’t care.

I don’t care whether you love or hate Zuckerberg, Bezos, or X (where X is a famous figure).

What I do care about is that people may be making a bad career bet by NOT considering superior, cheaper tools, technologies, and platforms, all because of a choice.

And just as I respect my decision to never use a Meta product because of Mark, I respect people who draw a bright line on Elon. Fair is fair.

But I would love for you to consider this idea.

You Can Learn from Someone You Dislike

I’m not the biggest fan of certain VC thought leaders on X. They can say some really dumb, tone-deaf things on the regular (e.g., Chamath’s “this is below my line” quote).

However, now and then they’ll say something brilliant that stops me in my tracks.

There’s a saying in Al-Anon (and other 12-step programs). Take what you like, and leave the rest.

I can filter out Elon’s politics and stupid memes, and eventually he says something so insightful that I journal about it and apply it within my professional career.

I don’t have to like (or even respect) him to benefit from his hard-earned wisdom.

It Doesn’t Have to Be All or None

I’m not the biggest Dana White or UFC fan. It’s just not my jam. But damn, he’s an A++ operator in his business and his brand. I watch all his interviews because his perspective is so different from what I’m usually exposed to in my circles. And many of his insights are dead on (and some of them aren’t, and that’s ok too).

If you hate and block him, you miss out. That’s fine. No judgment. Just know that you don’t get access to the occasional nuggets of gold because you hate when he talks shit.

I’m not for everyone either. But I hope that, despite the occasional cuss word or dumb political view, you find enough value in coming back here and engaging in a spirited discussion with me.

Agree? Disagree? I’m ok with either.

I would love to hear from you in the comments, especially if this helped you recognize you might be blocking a “good” opportunity because of a (real or perceived) “bad” person.

I mean, what if you’ve pegged person X into the wrong position Y?

Until next time. Thanks for listening ✌️

Oh, and the best place to keep up with AI is X (I’ll die on this hill).

Alas! Hat tip to seclilc for tweeting this gem out two days ago.

I’ve been thinking about it ever since. Why? Because there is growing grumbling among everyone about AI writing. And it’s not just others; it’s me! I am getting to the point where I physically flinch (sometimes dropping my shoulders and hunching, or having a slight eye twitch) when someone I respect sends me unfiltered and unedited AI output.

Look, I get it. It’s Q3 2026, and we should expect that everyone is utilizing AI at SOME point in their process (sourcing ideas, creating outlines, refining prose, etc.).

However, I have a new policy.

If you’re not bothered enough to review and edit it...

...then I’m not going to bother reading it.

Yes, there are certain situations in which we should expect 100% AI-generated copy. Customer support would be a perfect example. We’re not looking for artisanal “did you make sure to reset your phone” style dialogue.

But if you’re my colleague and we’re in a Slack discussion and you post a wall of Claude output, then I’m afraid I received a different message than you intended.

The same is true for people’s newsletters and social content. It’s your name on it; are you proud of the prose and weird AI-isms sprinkled throughout it? If so, great. But I can ask Claude directly if I wanted to.

TL;DR (too long; didn’t read) was the solution for social media.

AI;DR (AI; didn’t read) is the solution for AI slop.

May you embrace this policy yourself and seek out those willing to care enough to prioritize a human touch when they talk to you.

Updates

Update 1 2026-08-17: Daniel in the comments shared a wonderful website dontpastetheai.com. Amazing. I especially loved the Take me to the Angry Version version. The intro header says it all.

Update 2 2026-08-18: This made it to #1 on Hacker News on 2026-08-18 with >600 comments. It apparently struck a cord.

Update 3 2026-08-27: Given the popularity of AI;DR, I created graphics that you can use to spread the word!

Graphics

Dictionary Versions

Poster Version

Be sure to reply/share the best thing you read this week (or something that lit you up). Any topic goes!

1. Heart Health

In the last 3 months, 3 people I know had heart attacks (two didn’t make it).

Given I’ve been told I have a big heart (that’s a joke), and given my genetic predisposition (that’s not a joke), I’m now going through a battery of tests and scans to make sure I’m on top of this.

I don’t want to become the Tin Man and lose my heart. My two young kids don’t deserve to go through live without a father.

2. Going Loco for Local AI

Frontier AUI models are powerful, but the people running those companies do crazy shit.

In the last few months, we’ve seen models getting temporarily banned (Fable), degraded when you ask sensitive questions (again, Fable), and now Anthropic is watermarking ALL content from all its models (again, Anthropic).

Many are taking matters into their own hands, literally. I just invested $5K in an NVIDIA GDX Spark to test and run local models. You can follow my journey on X.

3. Peptides Ain’t Weird

For years in men’s circles, it was shameful to talk about things like testosterone replacement therapy (TRT). Well, the Overton window on that and other health/longevity “protocols” (gotta love that marketing reframe) has dramatically opened.

I’m somewhat late to the peptide game, but it seems this is already crossing over into mainstream. What used to be weird is now what everyone is exploring.

4. The Analog-Digital Barbell

Work colleagues see me going all-in on AI and ask… aren’t you worried about becoming less human?

Not at all, because my life has moved to both extremes of the barbell.

Work has become all-in AI. I’m embracing it as hard and as much as possible.

Life has become all-in analog. I’m outside with my kids. We’re swimming, climbing trees, playing Uno.

It doesn’t have to be either-or. You can and-both your way into a great life.

5. First, Care

Merlin Mann wrote a phenomenal article that I read 1-2 times a year (First, Care).

Unfortunately, it doesn’t exist on the web anymore. You have to know where and how to get it on the Wayback Machine, which itself may not be around forever.

It’s a reminder to me that if something matters enough, you have to make it last. I’ve been thinking of ways to permanently record the things that matter, so that my kids could have access to them long after I’m gone.

So I’m probably going to go back to writing some handwritten letters and add them to a vault. They’ll have all sorts of other memories of me, but I want to make sure they never lose some of the important ones.

And You?

What ideas sparked joy? Created a change in you? Would love to hear it.

DDEV was a startup I co-founded back in 2015. And for the first few years, we sucked at marketing. We had an amazing tech platform, but we lacked the time and creativity to explain it in a Sesame Street Simple way.

Eventually, our CTO came up with a brilliant idea. Explainer videos were all the rage because they helped create a compelling story in 1-2 minutes. This is exactly what we needed, so we committed to getting one ASAP.

Problem. They were VERY expensive. Good agencies would charge anywhere from $1,000 to $10,000, and the process was very time- and graphic-intensive. Our CEO ended up spending over 100 hours of his own time going back and forth with iterations on one he built from scratch.

Ain’t nobody got that much time and money these days!

Luckily, with AI, you can build these for a few dollars and in less than 20 minutes.

Here’s my go-to playbook

HyperFrames

Confession: I’m a HyperFrames Fanboy (shoutout to Matt Van Horn for getting me hooked).

Everyone uses AI to create transcripts and summarize meetings, but few read them.

A picture is worth a thousand words, but even the best infographics can only explain so much.

Enter the 2nd coming of explainer videos! They are beautiful, quick, and high signal-to-noise.

I even had a CEO once say, “Super appreciated the video by the way. I was able to get fully caught up over wedding dessert last night”

When was the last time your boss viewed your PowerPoint deck during their sister’s wedding? But videos are quick to consume and respond to. And in a world in which AI verbose-as-fuck, explainers are the best TL;DR solution we got.

Howto

The fastest way to get started? Give your AI agent this article and tell it to repeat these instructions for your website (or any website).

I tested this against a colleague’s website (Matt at Masterpoint.io), and I think it turned out well.

Here’s how to prompt this yourself with your website.

1. Install the HyperFrames plugin (takes 1-5 minutes the first time)

2. Install humanizer (optional)

3. OpenClaude or Codex

4. Use the following prompt.

I want to create a professional, compelling 60-second explainer video for Masterpoint.

1. Visit https://masterpoint.io with a focus on the homepage and product pages.

2. Use the company branding (fonts, colors, styles).

3. Ensure all copy is passed through /humanizer to not sound like AI.

4. Build the video using /hyperframes.

Save the video w/voiceover on my desktop, please!

5-20 minutes later, your video will pop out on your desktop!

Don’t like it? Hyperframes can make modifications quickly. Ask to modify:

• Storyboard

• Visuals

• Copy

• Branding

• Music (change or remove)

• Voice (change or remove)

• Duration (I prefer 30-90 seconds)

I’m usually happy with the first output, but sometimes I’ll iterate 1-3 more times, especially when the audience is a customer or a leadership/board meeting.

But 1-3 iterations is fast and cheap compared to 10 years ago.

Other Ideas

HyperFrames is one of my favorite AI skills that I’ve learned in the last 90 days. I use it daily for a variety of use cases.

1. Summarizing proposals and decks for execs

2. Product demos (polishes raw screen recordings)

3. Pitching product concepts

4. Company week-at-a-glance overviews

5. Meeting summaries

6. Tailored sales outreach messages

Dog-fooding

I’m starting an experiment today.

For all my long-form articles moving forward, I’ll include a 60-90 second explainer video as the TL;DR. If it works, I’ll continue. If it sucks, I’ll kill the experiment.

But it’s an attempt to live my values and prove the point. To that end, here’s this article as a HyperFrames overview. Enjoy!