After playing Krampus last week and distributing coal to atrocious managers and irresponsible tech companies, I’m in a kinder mood and ready for a positive look back at the year. Here’s a selection of the best and brightest contributors and innovators I wrote about in 2025.

6. Chicken sexers. Determining the gender of a newborn chicken is surprisingly tough, and remarkably resistant to systematisation. But in news just released this month, you can now get an AI to duplicate the chick-recognition process! This bodes well for other instances where we want to download an expert’s brain.

5. Fathom and similar robot recorders. [No link on Non Exec Email] AIs on Zoom calls are getting so common these days that some people even forget to ask whether they can bring a robot (my answer: not without asking nicely first). But where I do have permission, I find having a searchable, intelligible transcript and summary of every meeting accelerates my analysis and understanding hugely.

4. Tycho Brahe. Not strictly a 2025 highlight, more like one from 1575—but my capsule bio of this pioneering data engineer was a reader favourite, and he still has valuable lessons to teach us about meticulously gathering and cleansing data.

3. “Intelligent disobedience”. Lisa is on the long post-pandemic waiting list for a new guide dog, and my CFO (Chief Furry Officer) Aurora is filling in as best she can. But she’s not trained to disobey unsafe commands like a seeing-eye dog is. Just as we need more well-trained “eyes” for blind people, we need more engineers resisting unethical demands and refusing to build harmful software.

2. My lawyer. Solicitors aren’t winning popularity contests, but Lisa and I have a gem in Mr. Clark, our property attorney. He eschews complex rituals, the legal equivalent of agile “refinement” and “grooming”, and instead simplifies what he’s doing so we can understand its value and redirect him when needed – why don’t engineers do that?? We’re gutted he’s retiring next year.

And the perennial winner is:

1. Runbooks and checklists. Pilots use them religiously to keep calm and carry on when disaster strikes, and hospital staff rely on them to prevent medical catastrophes. But for some reason we haven’t adopted these vital operational tools nearly widely enough in the software world. I just built a “playbook” for a client last month and it’s not only boosting safety, it’s creating new opportunities to automate and to charge more.

Wishing you a happy, safe, and singularity-free 2026!

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 29th December 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

As the 25th approaches inexorably, I thought I’d share the content of my naughty list, the software, systems, and people to whom I’d like to send a lump of coal this year. I may be in a grinchy mood right now, but never fear, I’ll be back to my usual sunny disposition next week with some examples of better IT behaviour. Enjoy the parade of disgrace, and happy Festivus!

6. The Public Works Department of Madhya Pradesh, for not only designing, but actually constructing this hilariously impractical flyover. Did no one on site actually look beneath their feet to notice that what they were building made no sense?

5. British Petroleum, for totally neglecting psychological safety and allowing a Texas refinery to become so unsafe it eventually exploded.

4. Crowdstrike, for not only wiping out millions of computers in airports and hospitals, and then publishing a root-cause analysis devoid of root causes, but crowning the achievement by totally failing to publish the reports of either of the two “third party software security vendors” hired to explain what actually went wrong.

3. Robert New and his ilk. Mr. New was having a bad day and insisted that the pilots of his private jet land at the dangerous Aspen airport despite terrible conditions, even plumping himself down in the cockpit to make sure they did. That didn’t end well for him and the passengers, and it’s no good for others who, like New, apply unreasonable delivery pressure to their staff (technical and non-technical alike).

2. Slack—not the rope, but the software designed to destroy productivity. Is nobody listening to Cal Newport and Jonathan Haidt railing about the rise of distractionware?

And the winner is…

1. Vibe Coding. My friend Gene Kim may be a big fan, but I think there’s a lot more involved in making AI profitable than “giving in to the vibes”.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 22nd December 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Zip files are ubiquitous, it seems. You probably used one this week, although you might have thought you were opening a Word document or reading an e-book or planning to attend a Jimmy Buffet concert. You might know that “zipping” is an eponym for combining and shrinking files on your computer, and you may have fired up WinZip or one of its competitors recently to send pictures or share a mixtape. But I bet you’ve never looked under the bonnet to see just how data gets smaller, and you almost certainly haven’t noticed that ChatGPT and its friends are nothing more than super-advanced compression systems.

The first law of shrinking anything, whether it’s your kids or a skyscraper or all of Wikipedia, is to strive for perfect fidelity. “Lossy” compression, which trades off quality for size, is relatively easy but not very good; if you’re old enough, you’ll remember the grainy, jerky videos of the early Internet, crushed massively to squeeze through the narrow audio channels of the day. By contrast, the gold standard is lossless compression, which preserves every byte and pixel perfectly while still managing to make the whole file smaller. You can only do this if your data has lots of redundancy and repetition, which thankfully there usually is: we can compress

10000000000000000000000000000000000000000000000000000000000000

000000000000000000000000000000000000000

as “1 followed by one hundred zeroes” or 10¹⁰⁰ or a googol (no, not that one), all of which are much shorter than writing out the number in decimal notation. (The decimal form is itself compressed – one googol tally marks would take up a whole lot more space!) Noisy or random data is much harder to shrink than these nice repetitive examples, which is why Twitch streamers hate snow.

And if you think about it, you’ll see why there has to be a hard limit on how small you can make anything without destroying it. It might seem possible to start with a zettabyte-size data lake and shrink it 1000-fold to an exabyte, but what if you kept going at the same rate, compressing 1000x again to a petabyte, then a pterobyte, then a terabyte, and so on (yes, I’m making one of those up)? Eventually you’d get the whole world’s data stored in eight 1s and 0s, and that’s absurd. To prove it to yourself, try zipping a zip file and then zipping the result; you’ll quickly find the file gets bigger as you squeeze harder, a Sisyphean task that’s rather like folding paper ten times.

Attempts to devise ever more efficient compression schemes helped inspire the study of information theory. Researchers have worked out theoretical limits on how much reduction is possible; the less shrinkable your prose is, the more “information content” it has. So politicians and influencers should be pretty easy to condense, Joyce or Hawking much less so. A clever Soviet mathematician named Kolmogorov invented the leading “information” metric, taking as his benchmark the length of the shortest computer program that produces the text you’re measuring. And once you look at it this way, you start to see that clever compression actually involves some degree of intelligence.

Notice in our “googol” example above, the representations get shorter as they get more sophisticated: “one hundred zeroes” just involves counting, whereas “10¹⁰⁰” requires recursion and exponentiation. That’s what happens in real zip compression too, with programmers squeezing out smaller files using slick methods like adaptive Huffman coding that determine how to adjust the compression method on the fly by analysing the data as it arrives. In our daily lives, we’re constantly using and inventing new, compressed representations of the world around us, like Maxwell’s equations or the Ladder of Inference or even “Make America Great Again”.

And if you squint at your favourite chatbot, she starts to look like a compression supercomputer, capturing much of the complexity of the gobs of data she absorbed during training with “just” a few trillion parameters. I won’t say ChatGPT and Claude are “thinking”, but they certainly seem to build sophisticated internal models of the world, some of which we’re starting to discover inside their “brains”.

Yes, it’s true that AIs do nothing fancier than predict the next word, and for that reason you can dismiss them as mere parrots if you like. But in fact, anticipating what’s coming based on the limited information that’s already arrived is exactly what you need to make your compressed data comprehensible, and there’s good evidence that this ability correlates well with “intelligent” performance. One researcher believes in the link so fervently that he’ll give you up to €500,000 if you can make a breakthrough in “smart” shrinkage. Personally, despite the inevitable market correction, I’m confident enough in the long-term utility of today’s cleverly compressing chatbots to start a new book about how to use them without chatting. Don’t hold your breath for The Singularity, but you’d be nuts not to be boosting your productivity and profits with an army of adequate assistants.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 15th December 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Genius sci-fi parodist Douglas Adams is most famous for the six-part trilogy The Hitchhiker’s Guide to the Galaxy, which chronicles the adventures of a ragtag assortment of ridiculous characters like Marvin the Paranoid Android, the Ameglian Major Cow (so docile it begs to be slaughtered and eaten), and the Ravenous Bugblatter Beast of Traal (so stupid that it thinks if you can’t see it, then it can’t see you). Surprisingly, he also gave us one of the best object lessons in sales and product design I’ve ever heard, in the form of the tale of the Golgafrinchans.

In the second volume, the two protagonists Arthur Dent and Ford Prefect (who is not a car) accidentally teleport onto a Golgafrinchan spacecraft and soon meet the cheerfully incompetent captain, who has been soaking in a bath for the entire three-year journey and isn’t exactly sure where they’re going. He does know, however, that his ship, the “B Ark”, carries a cargo of 15 million hibernating bureaucrats and middlemen—advertising executives, personnel officers, telephone sanitisers, and the like—and that they’re all escaping the imminent destruction of their home planet. Strangely, the rest of the population, the people doing all the actual leading and executing, were supposed to travel in the A and C Arks but haven’t been heard from since departure. The truth rapidly emerges: the “doers” never left, the planet wasn’t doomed, and the whole thing was a clever ruse to get rid of the most useless 33% of the population of Golgafrincham. (For more details, listen to the captain tell the story himself.)

Adams, of course, just wanted to skewer the many holders of bullshit jobs he saw around him. (And I should note he put consultants on the B Ark too, lest it seem like I’m claiming immunity from ridicule!) But he also inadvertently gave us a highly selective filter for deciding which B2B products to build and how to sell them: if your ideal buyer belongs on the B Ark, you’re as doomed as Golgafrincham. (And doomed it was, too, though the planet wasn’t eaten by a gigantic goat as the captain thought; instead, all its inhabitants died of an epidemic that started, of course, with a dirty telephone.)

Now I’m not simply telling you to qualify your leads; every salesperson learns to recite BANT along with her ABCs, so that would hardly be news to you. Instead, I’m telling you that you should be applying the Golgafrinchan Guarantee long before prospects land, when you’re designing your product experience and writing the press release; way ahead of cutting code, standing up a home page, or cold calling industry contacts. At that point (I hope!) you’re identifying your ideal customer profile through user research and product analytics.

But beyond these obvious checks, you also should be verifying that your ostensible purchaser is responsible not just for budget, but for revenue; not just spending, but ROI; not just meeting objectives, but setting strategy–and that your product makes a unique, visible contribution to each of these responsibilities. In other words, your invoices need to be approved by a motivated, A Ark leader, and what you’re selling needs to contribute directly to that person’s strategic objectives.

You can have all the enthusiastic champions you want, but to land and keep a customer, you need the Gold Owner to be thinking about your software in the shower, and the Chair to be flipping to the relevant section of the board pack first to see what your product has done for this quarter’s numbers. Without high relevance and visibility, you simply won’t get funding, approvals, and renewals. (By the way, this is why HR so commonly stands for “Hopelessly Removed”: training, values, and surveys are far too often on the outer fringes and impossible to quantify, hence the first up the ramp of the B Ark when belts tighten.)

Last month, one of my clients booked a strategy intensive with me to help them address just this challenge of benefit invisibility. In account after account, they were an anonymous nice-to-have, a perk that made people feel vaguely positive, but even the most passionate of their earlyvangelists struggled to measure their contributions and attribute value to them. In fact, their salespeople were doing too good a job: they convinced many organisations to buy based on little more than promises of motherhood and apple pie, but as a result they hadn’t managed to measure value, polarise the market, and differentiate themselves. Unsurprisingly, this led to wooly positioning, an aimless roadmap, and buyers straight off deck 7 of the B Ark.

To help these folks clarify their focus, I thold them the story of my first CTO role, at a startup in the City of London. Our clients were finance bros coming straight from central casting, constantly shouting out numbers and trades, throwing phones, and partying all night. They lived for their outsize bonuses, loudly boasting about the number of zeroes they got every January. So we had a simple rubric for deciding what to build for them: whenever we evaluated a possible improvement to our product, we asked ourselves which client was going to buy a Ferrari as a result of our work–and if there was no such person, we didn’t even start.

Because the banks rewarded aggression and initiative, the “Ferrari rule” was a shortcut to identifying the buyers with money, motivation, and most important, organisational muscle: the access to impose their will and the clout to make those decisions stick. For your industry, you might need another filter to find decisive buyers who never leave the driving seat, but for pity’s sake, please keep your product out of the minor leagues completely. (Unless you like rubber duckies, of course.)

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 8th December 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Quick, what’s this symbol mean?

If you said “Stop”, then your brain’s ventral stream has been successfully hijacked by generations of traffic engineers. The first of these was frustrated architect William Eno (no, not that guy), an early campaigner for the standardisation of road regulations who in 1916 advocated “vivid yellow letters or arrows on a black background” for warning signs like this. (Perhaps he hoped that the clashing visuals would cause shocked motorists to screech to a halt, or vomit, or both.)

Thankfully, our streets were saved from this garish colour combination by the Mississippi Valley Association of State Highway Departments. In 1923, the MVASHD sensibly suggested sticking to black letters on a white background, as well as using a “dictionary” of shapes that isn’t too different from what we see on the road today: round for rail crossings, octagonal for stops, diamond for warnings, rectangular for directions.

It only took the invention of red paint that doesn’t fade in the sun and a Vienna Convention (no, not that one) to make global the ubiquitous symbol that’s burned into your visual cortex.

In Britain we prefer to traverse junctions with another of Eno’s inventions, the roundabout, so we don’t put up a lot of stop signs. By contrast, when driving in Alabama recently, I found my progress interrupted repeatedly by the silent commands of scarlet polygons. And it wasn’t just me; using the Mississippi shape code, I could tell from looking at the backs of the signs facing other drivers when they, too, were being told to stop – and at most crossings, everybody had to. It was bizarre, and also strangely comforting, to see three brodozers standing at attention while my little Squirrelmobile zipped between them.

For a car-obsessed nation, it’s odd how much the Americans love deceleration. Perhaps they have a point, though, since I’ve noticed that the act of hitting the brakes tends to jolt me a bit as my brain does a context switch from going to not-going – and junctions are certainly good places for a little adrenalin. In fact, you may have noticed that in almost any endeavour, one of the best ways to grab your audience’s attention is by arresting your progress. Jokes are funnier when you stop dead after an unexpected punchline. Marketers use pattern interrupts to break the rhythm and wake you up. And musicians have a “tram lines” symbol that tells them when to pause the whole performance to make a point.

You can use exactly the same method with your staff, your Board, or your suppliers: say something outrageous and then shut up. For instance, just this week I told clients bluntly:

• You don’t need a tech strategy, you need a press release.

• Don’t bother making customer service more efficient; they don’t cost that much anyway.

• If you have to think this hard about a candidate, she should be a “no”.

You better believe each of these was followed by awkward silence, productive conflict, and a sharp right turn for my client.

Whether you’re disappointing helpfully, delivering a one-minute reprimand, or practising radical candour, a sudden plunge into silence underlines your point and makes people listen. Why not try a “four-way stop” in your next meeting?

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 1st December 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

A venture-capital client asked me this week about why “lite” versions of gargantuan “enterprise” products crash and burn so often. Inside and outside software, why can’t we shrink our elephants?

I thought instantly of the day two tough, burly gentlemen from Bloomberg burst into the youDevise office, ripped out the terminal they’d sold us, whose lease still had years to run, and hauled it out the door without a word. That was the end of our partnership and the declaration of war; the first shots came weeks later when they announced their own product as a direct competitor to our Trade Ideas service and started taking away our customers. Given Bloomberg’s size and ruthlessness, and the fact that they weren’t even charging extra for their offering, it seemed obvious that we’d soon be crushed by the juggernaut.

But only a few years later, our 80-person startup had been acquired at a handsome price with a healthy network of clients, and Bloomberg’s TMSG was sent out to pasture, only to be hauled out occasionally for student competitions. What went wrong, and why did we win so spectacularly?

There’s one obvious reason: we’d invented the niche “alpha capture” market, but Bloomberg were trapped in the wrong paradigm. In our little world, the unit of value was the “trade idea”, basically a stock tip sent by a bank to a hedge fund. The best and brightest at all the bulge-bracket banks sent thousands of these through our system every day, but for us they weren’t just fire-and-forget recommendations; each one had attributes like duration, alpha, and Sharpe ratio, and we kept them forever for ranking and backtesting.

By contrast, Bloomberg had started as a one-way source of news and market data, and later added bidirectional chat between traders–but throughout their wide-ranging empire, the value lay in instant delivery of up-to-the-minute information, not slicing and dicing the stream of messages as a whole. So although they tried to track and analyse each idea, they never managed to figure out how to calculate and present vital aggregate information like “my bank’s sentiment on Latin America” or “market opinion on the Zappos acquisition”.

And even worse, they couldn’t fathom why financial behemoths placed such value on ownership and control of their trade ideas. The banks hated Bloomberg’s high-handed belief that it had the right to track every message and keystroke on its terminals, which meant, for example, that the news division could and did snoop on market activity willy-nilly. So they preferred to pay us healthy license fees instead of giving away their intellectual property to a powerful and untrusted player.

Beyond being stuck with these crippling delusions about the market, Bloomberg was prevented by its size from using techniques that kept us well in the lead:

• We regularly launched Wizard of Oz features to learn quickly what users wanted without having to actually write the code to make them work. In fact, we started the whole company with a barely-working weekend hack I threw together that wowed a couple of the banks into hiring us in the first place.

• Our salespeople were guerillas, tailgating onto sales floors and selling to one individual trader after another until we’d penetrated so deeply that only a global license made any sense. To cut out approval delays, we carefully priced the product just below the amount a salesperson could spend on his or her own company credit card.

• We stole staff from banks whenever we could, tempting technical geniuses to join us with promises like “you don’t have to wear a tie” and “we don’t use COBOL”. These “refugees” were our most expensive investments–and they proved to be some of the most profitable, too.

Respectable suit-wearing Bloombergers, with a constant need to assure the Street that they was still printing money, could never be as skulduggerous and audacious as we were–everyone would have seen them coming a mile off.

But let me be clear: there was absolutely nothing techincally preventing Bloomberg from employing these market strategies, creative pricing, and sophisticated analysis tools to beat us into a pulp. For proof, consider IBM’s Project Chess, which used similarly unorthodox methods to take over the microcomputer market with the PC despite IBM’s famously cautious culture. Bloomberg could easily have launched a similar skunkworks, but the idea never even occurred to them, thanks to corporate cultural myopia. Bonuses at every level were based on one thing: terminal sales. That was the sole metric anyone used to measure the success of an initiative–so surely absorbing our technological distinctiveness into the terminal, for free, had to make number go up!

I can only imagine how surprised the strategists must have been to see their predictions go sideways–almost like firing a 15-inch shell at a piece of tissue paper and seeing it bounce back at you. In the end, what felled Bloomberg, as so many other Goliaths, were stones that didn’t even seem to them like weapons.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 17th November 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

In an era of blink tags, hit counters, and malware monkeys, the stark Google home page arrested attention with its simplicity.

The blank box was a perfect capability, allowing users to type whatever they wanted and see what came out: “inch same length Europe” or “avocado pit huge why” or “do dead people pay taxes”. At worst, you would get “no matches” and an ice-fishing yeti; at best, you might stumble on dinosaur exhibits or early memes or dancing hampsters. The intentionally contrarian design choice positioned Google as a jinn in a bottle, ready to answer any question and satisfy any desire. And just in case you couldn’t tell that this was a land of wondrous wish-fulfillment, the “I’m Feeling Lucky” button was right there for instant gratification, taking you straight to the first hit without a confusing choice of results.

The alert among you may have noticed that OpenAI and its competitors are playing just the same game. Here’s the ChatGPT home page – look familiar?

It’s the same seductive tabula rasa, inviting you to fill it up with whatever you want by showing you that coy little “Ask anything” prompt. Inquiring about literally any subject under the sun sounds like science fiction, and in fact it is: on the Starship Enterprise, crew members just blurted out whatever question came to mind and the magical, disembodied “computer” replied with the right answer in a dramatically robotic voice. I’ve written a lot about this “Star Trek interface” and its pros and cons for users – but today I’m thinking about how valuable it is for the programmers, and how you and your tech team could get the same amazing insights it offers.

As you may have noticed, both Google and OpenAI use their open-ended text box as a probe inserted directly into your prefrontal cortex, asking you occasionally whether you found a response useful or giving you two choices and requesting that you name your favourite. And behind the scenes, they’re constantly watching the stream of user inputs rushing past, fishing for insights and needs and potential new products – responding to inputs like “sore throat tea” and “feed or starve a fever” with Flu Trends, and answering “647/988” and “sin pi/7” with the Google calculator.

Your technologists probably spend lots of time (and your money!) on user research and focus groups and requirements analysis, trying to uncover elusive needs and surprising gaps that they can fill with nifty new features. (If they don’t, phone the Squirrel-Phone right away – you have a bad case of walled gardenitis!) But what would happen if you augmented these practices with a blank-slate capability like the Google and ChatGPT text boxes, giving real users the chance to demonstrate directly what they’d like you to do for them?

You’d have to prepare them for disappointment of course (yetis and fail whales help!), and you’d have to show them that you’re acting on their expressed wishes, rather than just engaging in UX theatre. But the examples above demonstrate just how valuable a “suggestion box on steroids” can be.

If you squint at your software, can you see a seam you could open up for users to do anything they want? If so, try filling it with an open-ended capability and see what weird and wonderful things you never knew your customers were wishing for.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 10th November 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Who doesn’t enjoy a good slasher flick at Halloween? The relentless killer dons a hockey mask or inhabits a doll’s body or possesses a 500-year-old turkey and lumbers through a spooky forest or abandoned summer camp, chasing after hapless teenagers who never seem to run fast enough to find help. It’s hard to take this hackneyed genre seriously when everyone knows the tropes and the parodies keep multiplying. But if you lighten up and enjoy the adrenaline, you’ll find a frightening film doesn’t do you any harm–and horror movies might even teach you something about optimising your team’s work.

Non-technical executives who manage or depend on tech teams share a problem with adolescents trapped by maniacs. Both have an urgent need to produce a result in an unfamiliar environment, but it’s hard to figure out how to measure progress accurately.

Should the kids split up and look for the exit? Or sneak around to find the bad guys to discover their plans? Perhaps they should search for a light source, or look around for a secluded spot where they can wait for rescue (and meanwhile get stoned and make out). As you have surely been shouting at the screen, all of these are catastrophically bad ideas that always end gruesomely. Instead of running around frantically doing, the group should stick together and do some thinking so they can find a safe route out or a way to signal for help. The solution isn’t obvious, and the knee-jerk actions don’t work, because the situation is completely novel: calling the cops isn’t going to work when the killer is on the phone upstairs. Even the earliest examples of the genre feature characters who wander away from safety and “never leave well enough alone until they are all in a mess”, demonstrating that action for action’s sake is counterproductive at best.

Unfortunately, I meet leaders all the time who haven’t seen enough horror films. Within their tech teams, they endorse and reward not learning and results, but actions and activities: hours in the office, story points completed, features released. And the outcomes are just as ghastly as in the movies: demotivated zombie engineers running around frantically trying to look busy, carrying out seemingly obvious tasks that don’t actually produce anything valuable, because no one’s realised that the friendly security guard is actually the murderer – er, I mean, no one’s paused long enough to notice that we haven’t counted the users or run the load tests.

Now to be fair, measuring pure action works some of the time. If you want to make a ton of scripted sales calls, say, or issue a blizzard of social media posts, go right ahead and count the routine operations to figure out who’s most productive. There are even programming tasks that we understand well enough for this kind of reductionism to work: designing a home page or building a basic e-commerce site both have well-worn playbooks that make it easy to gauge progress (so easy, in fact, that even chatbots can do it).

The difference is that while these well-trodden problems have complicated characteristics – you need expertise, analysis, and a library of recognised tactics to deal with them – they are not at all complex like the construction of pioneering software (and almost all software is de novo, even the CRUD-diest). In Snowden’s complex domain, there are no experts, because nobody has yet figured out the right answers. Whether you’re seeking to identify the masked killer or the killer feature, you need to stay by the fire and send out speedy, small probes into the darkness to find out what’s out there, before you make a run for the door.

In technology, as in the factory or the haunted mansion, keeping everyone active all the time is the way to a messy death. The key to survival is to put throughput ahead of utilisation, and accept that sometimes, it’s best to stare out the window and think.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 27th October 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

The arrival of rapid, motorised transit in 1820s Britain may have thrilled and liberated many, but zipping around on rails at incredible speed certainly wasn’t welcomed by everyone as an untrammelled good. One opponent, Thomas Creevey, found that when racing along at the unheard-of velocity of 20 miles per hour,

… it is impossible to divest yourself of the notion of instant death to all upon the least accident happening. It gave me a headache which has not left me yet.

And Creevey was hardly alone: Queen Victoria insisted that she never be transported at any rate faster than 40 mph, there were Parliamentary debates about fires started by sparks from speeding trains, and an influential magazine solemnly recommended that the government limit the top speed to “eight or nine miles an hour” on every railroad in Britain. There was even a form of “railway neurosis” attributed to the jarring, hurtling journey and the (unfortunately frequent) crashes it led to. It’s safe to say that going faster meant only confusion and disorientation to a large proportion of the population, and not only in Britain, but in Asia and North America too.

Just about two centuries later, it seems we’re in a similar period of quickening, with social acceleration dislocating and disrupting us–not affecting our physical motion this time, but speeding up the movement of ideas and information. Like steam trains for the mind, our pocket computers accelerate our access to the thoughts and opinions of others, delivering a stream of novel concepts to us at dizzying speeds. But the very rapidity and ease of our communications can lead to unexpected consequences, especially when we start introducing feedback loops into the network we’ve created.

Some of these loops are silicon-only, with machines replying to each other faster and faster in a broadcast storm that produces bizarre results. For example, a book about the ontogeny of fruit files became the unintended object of a bidding war between two automated Amazon sellers and wound up priced at $23 million. Less humorously, the Flash Crash of 2010 was triggered by about 15 seconds of irrational trading among hedge fund bots playing “Hot Potato”: one sold shares in response to high trading volume, another boosted its buying because it saw all those sell orders, which in turn increased volume still more and triggered additional selling. Eventually the whole market went crazy and started filling orders for wacky prices like ACN at 1 cent or AAPL at $100,000, and the exchanges had to hit the undo button.

And things get really interesting when you put humans in the loop. On the positive side, information liquidity and our ability to change digital assets instantly means we can adjust to market demand incredibly quickly using methods like Customer Development and Elephant Carpaccio. The most extreme example I know of this OODA-loop optimisation is YouTuber Mr. Beast obsessively trying new thumbnails and tag lines sixteen hours a day as he rapidly iterated toward the recipe for the perfect viral video. You don’t have to operate at Jimmy’s “lunatic” level to benefit from this acceleration, either; just this year I’ve helped several clients move to once-a-day software releases and get immediate feedback on what engineers are doing–and whether their work is profitable. (Hit reply if you’d like my booklet on Daily Feedback explaining exactly how to do it.)

Just as in the railway mania days, moving faster comes with plenty of negatives too. It turns out that locking humans and computers in a feverish, self-reinforcing dance can be truly dangerous. You’ve probably heard about the damage done by attention-farming apps and the rise of AI psychosis. But you might not realise that there’s a long history of exploitation here. Back in the Stone Age, I strode into my local arcade with a couple of dollars and the belief that I’d soon master Donkey Kong, not realising that by design, the average game lasts under a minute. My allowance was being sucked out by addiction merchants who were, even back then, primitively hijacking my hedonic treadmill.

Taking “hook them while they’re young” to a logical extreme, in the 2010s some very odd characters created a series of increasingly disturbing “unboxing” and “superhero” videos aimed at getting toddlers to watch, entranced, for hours on end. Naturally this generated untold wealth from monetisation, but I shudder to think what it did to those developing brains–brains whose clicks were driving the very feedback loop that told the bad guys how to get even more views from their bad dreams.

Since then YouTube, of course, has taken some steps to stem the tide of nightmare-inducing “kid vids” (though I’m not sure the kidfluencers are much better). But just this month we got Sora 2, an impressive breakthrough in AI video production linked to a TikTok clone. Unsurprisingly, the results are both wonderfully odd (cheerful TV host Mr. Rogers wandering through battle scenes from the Second Punic War) and deeply disturbing. My gloomy prediction is that this new reward loop–prompt to video to Cameo to views and back to prompt–will drive us even further toward the uncanny valley of “content” our subconscious wants but our selves don’t recognise. At least it’s Halloween, so unwrapping candy and wearing Spiderman costumes will fit right in.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 20th October 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

You’ve probably never heard of Danish astronomer Tycho Brahe, but this noseless nobleman was arguably the greatest data engineer of the 16th century.

Most people who study the skies are inspired by a poetic experience, awed by the “black silk studded with the sequin-glint of stars”. Not so Brahe; when he saw Jupiter and Saturn align in his sixteenth year, he didn’t wax lyrical or predict floods and tornadoes, but instead noted with horror that the standard astronomical tables of the day made spectacularly wrong predictions about the event. When he looked further, he found the published calculations were rife with mistakes, in many cases due to sloppy calibration and poorly made instruments—and he thought, “I can do better.”

To correct these egregious errors, he immediately began keeping meticulous records of his own daily observations using the best tools available (the telescope, of course, wouldn’t appear for another half a century). He convinced the king to appoint him to the position of court astrologer, and in return for casting horoscopes and making dubious annual predictions about politics, he had virtually unlimited funds to build a cutting-edge observatory on an island between Denmark and Sweden.

https://permalink.kb.dk/permalink/2006/manus/742/eng/

With financing secure, he and his assistants proceeded to spend every day of the next thirty years writing down all the astronomical observations they could think of, including altitudes and azimuths, ascensions, magnitudes, temperatures, anomalies, and intervals. They “cleaned” what they gathered just as data scientists do today, by culling abnormal readings and correcting for confounding factors like refraction.

And when comets and “new stars” appeared, the team made sure to record the attributes of these unexpected guests with the same care:

[H]aving confirmed that my vision was not deceiving me, but in fact that an unusual star existed there…immediately I got ready my instrument. I began to measure [the supernova’s] situation and distance from the neighbouring stars of Cassiopeia, and to note extremely diligently those things which were visible to the eye concerning its apparent size, form, colour, and other aspects.

(As you might be aware, modern scientists are doing something similar right now, frantically gathering every scrap of information about our latest interstellar visitor before it disappears again into the void.)

I’m telling you all about Tycho not only to hold him up as a paragon of exactitude in data collection, but also because his story teaches us a highly relevant lesson about data usage. You see, he didn’t create these thorough, exhaustive records just to promote his own ideas about a stationary Earth orbited by nearby stars and planets. He realised that future generations would benefit from a long-term study of every observable aspect of the heavens, not just the elements that seemed relevant to him. And he was right!

If you’ve heard of Brahe at all, you probably know that Johannes Kepler used his planetary-motion observations shortly after his death to prove that solar orbits are elliptical, with variable speeds. His mentor, convinced by the Bible and his own eyes that heavenly objects moved at constant velocity in stately circles around the Earth, would surely have been shocked to see his data used to establish such a heretical fact.

But his measurements were valuable in many other cases too. For example, Tycho searched fruitlessly for years to find any sign of the “stellar parallax” that would have to be observed from an orbiting Earth. In fact, there is such an effect, but it wasn’t noticed until the 1830s for the simple reason that the stars are much, much farther away than Tycho or his contemporaries could have imagined. Ironically, the Danes were so careful and thorough that their failure to detect parallax was strong supporting evidence for this result that “totally revised the scale of the Universe” centuries later.

And amazingly, Tycho’s massive database remained relevant even longer than this. Seventy years after funding dried up and the observatory closed, the French astronomer Jean Picard (no, not that one) travelled to the site of the now-demolished Uraniborg Castle to determine just where and how the observations had been made. This wasn’t a pilgrimage or an archaelogical expedition; Picard wanted to quantify just how much better his instruments were than those of a hundred years before and how far off Brahe’s figures had been, and to do that he first had to locate the exact spot where the measurements had taken place.

In the event, the Frenchman found that his Danish counterpart had been remarkably accurate, with only tiny discrepancies in the longitude of certain stars that he attributed to instrument error. But it turns out that some of those inaccuracies were actually due, in part, to a hitherto-unknown phenomenon called the “aberration of light”, first discovered in 1727 and not completely explained until Einstein cooked up E = mc² at the dawn of the 20th century. This means Tycho was still giving clues to new discoveries nearly 350 years after he first started writing down what he saw!

Now, I readily admit there’s little chance that your business data will be paying dividends centuries from now, whether you’re observing customers’ burger preferences or the efficiency of certain workflow choices or something totally different. But Tycho’s lesson is that it’s impossible to predict which data points will prove to be unique and where you’ll be able to extract surprising value from them, so it’s worth recording whatever you can in case it’s useful.

Consider the sabermetricians and the flotsamologists, discovering neglected sources and milking them for all they’re worth. Plus I can introduce you to plenty of my clients, too, in industries from transportation to biotech to manufacturing, who are digging out nuggets every day from the data their users willingly share.

And since accurate information about your customers is uniquely yours, it forms an ideal moat against competitors–one that’s potentially highly valuable as we develop more and more ways to let loose our robotic researchers on data stores.

Since it’s relatively cheap to clean and save the information your business is producing anyway, why aren’t you doing it? All you need is a data engineer, a pipeline, and a dream. The gold nose is optional.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 13th October 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

In early 2001, Robert New was having a bad winter. The equipment leasing company he’d founded in Florida a few years earlier had gone public after only eight months, then used the funds it raised to gobble up a bunch of similar firms. But as CEO he’d failed to integrate the acquisitions effectively and the company developed a bad case of indigestion, filing for bankruptcy at the end of 2000 just as the bottom fell out of the stock market and triggering a lawsuit from angry shareholders. The skiing trip in Aspen, Colorado he’d planned for a coworker’s birthday must have seemed like a long-overdue distraction.

New and 14 of his friends and colleagues turned up to LAX well after their planned departure time in the afternoon of the 29th of March, but no one was worried about missing the flight because their host had borrowed a private Gulfstream jet for the occasion, just like the ones his firm had once owned and leased out. The pilots waited around for the leisurely boarding of the passengers, then eventually took off forty minutes behind schedule.

Because planes have to approach over treacherous mountains, landing at the second highest airport in the US is prohibited after nightfall. So the delay, plus the low clouds and snow showers reported at their destination, meant there was a significant risk that the crew would have to divert to another city. But Robert wouldn’t hear of it, angrily telling his assistant and the charter company multiple times that the plane was going to Aspen no matter what. The company’s scheduler duly relayed his demands to the pilots with a phone call to the cockpit en route, and the captain acknowledged that arriving as planned was important because Mr. New had “spent a substantial amount of money on dinner.”

As the aircraft entered the landing queue and the sun touched the horizon, clouds hid the airport and snowflakes began to obscure the view through the windscreen. Further along the conga line, two planes did go-arounds because they couldn’t see clearly enough to land. But just at this crucial moment, through the cockpit door strode a passenger–probably Robert New, though we don’t know for sure–and buckled himself into the jump seat. He didn’t say much after he sat down, but I’m sure the pilots got the message: land now, or else.

The events from here on seem tragically inevitable: skipped callouts and altitude checks, a fib to the air traffic controller (“Yes, now we do [see the runway]”, when there was no possibility they actually could), a misguided attempt to get below the clouds by accelerating the descent, and finally a slow pinwheel into a hillside and the loss of all 18 souls on board.

There’s no shortage of people and systems to blame in this sad history, but I want to focus on the abusive pressure Robert New exerted on the flight crew. There’s no indication in the record that he overtly threatened the captain or the first officer. It’s not clear whether he even broached the subject of a diversion directly. But his actions made his demands unmistakable, and I have no doubt whatsoever that his idée fixe influenced the pilots to take the dangerous shortcuts that led to the catastrophic end of aircraft N303GA.

I’m sorry to say that I’ve been on the receiving end of bullying demands from those in power many times, and I’m sure many of you have seen or experienced this type of behaviour as well–though I hope never with such a tragic result. Thankfully, the Aspen crash is a rare exception in aviation, because we drill pilots over and over in using tools with names like CRM and TEM that are designed to insulate them from exactly this kind of manipulation and influence. And other professions have similar guidelines, like the ethical frameworks we teach doctors and lawyers to apply without fear or favour.

My question is, why don’t software engineers have a standard code of conduct to help them resist unethical demands and uphold professional standards? It’s not like we lack examples of programmers led astray, from Nishad Singh at FTX to Gareth Jenkins at Fujitsu (and see this excellent Pragmatic Engineer article for even more).

Emboldening coders to say no to unscrupulous requests would be a tremendous social good, it seems to me. After all, my wife’s guide dogs were trained in “intelligent disobedience”, which means they would refuse to take her into a dangerous situation even when ordered to. If we can do it for service animals, it certainly seems like we can, and should, do the same in software.

Let’s tell engineers unequivocally not to store passwords in plain text or build manipulative addictionware, even if the boss is pounding the table and the client is screaming. We should never let a bully like Robert New intimidate us into landing our software in the dark, creating deceptive designs and fatal defects, just because he holds the purse strings.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 6th October 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

I recently chaired a high-spirited discussion about “Tech in 2026” with a group of super-sharp executives from some of the wealthiest companies in the world, and the only thing they wanted to talk about was AI.

As I listened to them assess the risks, describe their usage policies, and argue about the best way to boost productivity with bots, I could see they had given the technology an honest try and were thoughtfully grappling with its plusses and minuses. But it gradually dawned on me that a strange sort of Sapir-Whorf phenomenon was warping their brains: most of them literally couldn’t see beyond the chat window. Every use case and example they mentioned involved linguistic tennis with a bot, assuming implicitly that we have to ask and prompt and demand to get results.

It’s easy to see why they’re stuck in the chat trap. From the start, OpenAI offered us too-cheap-to-meter conversation, a bare box on the screen reminiscent of early Google in its seductive invitation to type anything that strikes your fancy and receive a coherent, often useful response. And now that it has a voice, the GPT genie grants your wishes with astonishing ease. Like Captains Kirk and Picard, you just say what you want and it appears, whether it’s tea, Earl Grey, hot or a picture of a pelican on a bicycle or a detailed report on the decline of German science after the Great War.

But it’s the very user-friendliness of this Star Trek interface that’s hiding the real value of ChatGPT and its friends. The dirty little secret of LLMs is that despite their friendly faces and air of glib confidence, they are actually atrocious writers and speakers. The hallucinations we’ve all seen are just their most obvious flaw, relatively easy to detect and edit out. Much worse, their writing is boring and often subtly inaccurate, and you can smell their sycophantic style of delving and bullet-pointing a mile off. Publishing the slop is a quick route to embarrassment, and blindly following its advice could land you in court.

As I’ve been screaming into the void at every opportunity for quite some time, your bots belong offstage, summarising and researching and coding under human supervision. These “centaurs” consistently outperform both computers and people acting alone, and no one has to eat bacon ice cream. For example:

• My client RealEyes use AI to detect unspoken listener reactions on the spot, allowing a lecturer or speaker to adjust content and delivery in the moment.

• I worked with a company a few years ago who used a bot to absorb millions of pages of documentation for industrial machines and then to produce a non-AI, deterministic software program that followed a decision tree to debug problems with the equipment.

• Several of the asset managers in the Squadron have told me they’re using AI to filter potential investments, from startup firms to acquisition targets, eliminating unqualified opportunities and feeding the good ones to human analysts.

Notice that in each of these cases, nobody’s typing or chatting or prompting. What’s happened instead is that with a little forethought and clever programming, these organisations have managed to produce research results of decent quality that augment, rather than replace, human decsion-making.

The future of AI isn’t in making emails a little better or having navel-gazing conversations or assembling a group of “AI agents” that hold meetings instead of doing their jobs. It’s in recognising language models for what they are: enthusiastic, clueless, and super-speedy research assistants.

My most successful clients are building software and processes around these silicon sidekicks and feeding their outputs to skilled, carbon-based life forms. If you and your team are still chattering into your phone or screen, hoping to guess the magic prompt that does what you want, you’re way behind. Instead, ask your engineers (or me!) about building your own eminence of centaurs.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 29th September 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

The ancient Greeks were troubled by moles in their gardens just as we are, and agricultural historians say they apparently asked these unwelcome visitors to depart via polite letters inserted into their tunnels. I’m sure the farmer who built my house 600 years ago was equally perturbed by the moldwarps undermining his crops, and probably tried to smoke them out. But despite centuries of human effort to relocate the talpids, I know that at least twice a year I’ll have to report the latest row of hills to my local molecatcher. He diligently sets traps to protect our plants, which usually halts the invasion of the moment–but he doesn’t seem to be any more successful than the Greeks at convincing the mostly-blind army to dig somewhere else.

Many of the executives I coach have a similar problem with their emails, meetings, and to-do lists: no sooner do they clear out one batch of requests than another load of tasks hits the inbox. Mr. Sisyphus never got anything done, and neither do these leaders, trapped in a cycle of perpetual busy-ness and reply addiction.

The radical solution, of course, is to stop using email altogether, like Donald Knuth or Christopher Nolan, but that takes clout and chutzpah to pull off. Are the rest of us doomed to perpetual battle with the invasive species spawned by Jira, Calendly, and Salesforce? Do we have to resign ourselves to periodic email bankruptcy? (And don’t even get me started on Slack or Teams, the spawn of Satan.)

I’m optimistic though, because the more I look at these issues with my clients, the more I’m convinced that we ourselves are feeding this messaging whac-a-mole–which means it’s in our power to stop the cycle.

To see why, check your most recent outgoing emails, and count the number of requests and questions and demands in each one. Are you setting up a meeting, asking for more details, bringing in more participants to the thread? Probably all of these and more! In fact, I bet you’ll find in most cases that you’re triggering more than one response per incoming message–and if everyone else does the same thing, we’re suddenly fighting exponential growth, which never ends well.

Accordingly, I counsel leaders to close off the message explosion at the source: their keyboard. I have them process each email in their overflowing inbox, one by one, just like my neighbour does with his flock of sheep every June. Then I ask them to “shear” each one by asking themselves, “How can I make sure I never see this again?” Like Bacon’s books, some message topics can be delegated (“Amy will lead this project and handle the planning”), others volleyed back at the sender (“Take care of this and report progress at our usual Friday session”), and still more can be converted to processes or automated (“Count these alerts and put them on the weekly dashboard”). Only a very few deserve to be “chewed and digested”, and even those can be scheduled or converted to reminders. I find the shearing needs a few repetitions to become a habit, but once it does, I find the flood slows to a trickle.

On hearing about this method, one client told me he’d worked in the White House and saw the President apply a similar technique, working through a pile of memos at lightning speed and marking each one with “Agree”, “Disagree”, or “Let’s discuss” (and not many of the last one!) Ensuring even the most consequential decisions don’t lead to communications cascades was what allowed the Commander-in-Chief to concentrate on strategic thinking, not the operational minutiae of one of the world’s largest employers.

That same client now has whole days clear in his calendar for reflection and planning, and he’s much more relaxed about his “deliverables”. Are you and your executive team similarly and systemically eliminating message avalanches, or are you buried under the weight of incoming demands? If “shearing” doesn’t work, you could always try pushing a Post-it note into your USB port:

“I adjure you, ye Jira tasks here present, that ye neither injure me nor suffer another email to do so. If ever I catch you here again, by the Mother of the Gods I will uninstall you in seven keystrokes.”

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 22nd September 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

In the 1990s, Bill and Hillary Clinton got themselves in all kinds of trouble with failed land deals, nepotism, and spying on opponents, but for me, the most memorable blunder of all was Bill’s reckless, puerile affair with an intern. As usual, the coverup was worse than the transgression, and when it all came out, the President tied himself into verbal knots trying to explain away the previous vigorous denials from the White House and his attorneys. For example, when confronted with his lawyer’s statement that “there is no sex of any kind in any manner, shape or form, with President Clinton”, he pettifogged:

It depends on what the meaning of the word “is” is…. [I]f someone had asked me on that day, are you having any kind of sexual relations with Ms. Lewinsky, that is, asked me a question in the present tense, I would have said no. And it would have been completely true…. [A[nyone generally speaking in the present tense, saying there is not an improper relationship, would be telling the truth if that person said there was not, in the present tense; the present tense encompassing many months. That’s what I meant by that.

[https://www.washingtonpost.com/wp-srv/politics/special/clinton/stories/bctest092198_4.htm]

Sifting through the word salad, we find a surprising kernel of truth: the verb “to be” does indeed have multiple meanings! Consider these examples:

• We can walk across the road now – the traffic light is green.

• It is raining. Let’s play board games today and take a walk tomorrow.

• The street is closed. We’ll have to use a detour until the roadwork is done.

• My puppy Aurora is a golden retriever.

These sentences are all in the present tense, but the duration being described varies widely. The signal will turn red in a minute or so; rain lasts a few hours unless it’s a monsoon; construction might take weeks or months; dogs never change their breed. A linguist would mumble something about the stative copular verb admitting instantaneous, state, and habitual meanings, but we can just observe that, conveniently for politicians who want to Humpty Dumpty their previous statements, even a seemingly simple word like “is” has unexpected complexity when you get up close.

These definitional surprises crop up everywhere once you start looking for them:

• A day is not always 24 hours long – it doesn’t even have a single well-defined length.

• The “boiling point” of water isn’t 100 degrees in most locations.

• Shakespeare wasn’t the only author of “his” plays.

• Not all of France is in Europe.

• Nobody agrees on how many Ten Commandments there are.

We get ourselves into tangles like these for a very good reason: oversimplifying intricate concepts is the only way to fit the modern world into a single human head. If we tried to stuff schoolchildren’s brains with the true complexities of the world they live in, telling them that the Hundred Years’ War lasted more than a century and that money mostly doesn’t come in dollars, pounds, or rubles, they’d never pass a single test, nevermind graduate.

I’ve carefully attuned myself to pick up these distortions and meaning mistakes when I’m coaching – and not only because technology has its own confusing cryptolect, where daemons aren’t evil and cookies aren’t nutritious. In fact, my clients can usually translate the technobabble, but most often trip over their own reductionist tendencies, forgetting to agree on what important words mean and fooling themselves into accepting “best practices” that aren’t. So I teach them to recognise and correct their unproductive assumptions and to test their beliefs about others’ reasons for acting as they do.

Like Socrates, I’m using elenchus to produce a state of productive confusion. I hope that most of the time, after talking to me for a little while, you’ll realise that what you know for sure just ain’t so and become curious rather than defensive. It can be tough to balance advocacy and inquiry, but if you find yourself slipping into Model I, just remember that just like Bill, you can’t even be sure about the meaning of “is”.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 15th September 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Canadian classical pianist Glenn Gould was an eccentric genius who alternately delighted and infuriated audiences with the exaggerated precision and wild expressiveness of his playing. To hear how radical he was, just compare the first few notes of Mozart’s Alla Turca as performed by Gould to the same piece interpreted “normally” by Barenboim. He committed many heresies, including regularly ignoring explicit markings in the score by playing, say, fast and staccato where the composer expressly called for adagio legato. In 1962, before conducting Brahms’ First Piano Concerto at the outrageously slow tempo Gould had chosen, Leonard Bernstein felt the need to give an extraordinary disclaimer to the live audience and worried privately that everyone would leave after the first movement (though nobody actually did).

Gould’s idiosyncrasies extended far beyond the keyboard; he demanded the heat be cranked up whenever he played, refused to perform music by composers he didn’t like, and, notoriously and astonishingly, insisted on using a very particular chair. He turned up to every concert with an oddly shaped seat specially modified by his father and carefully adjusted the legs so he could sit nearly on the floor, crouching like a very uncomfortable gargoyle attacking the keys:

But the result of all this outré activity was transcendent music that influenced every pianist and conductor who came after him. His 1955 version of Bach’s Goldberg Variations is still hailed as one of the most influential keyboard recordings of its century, “revolutionizing the art of interpretation”. Gould’s music even went to space onto the Voyager Golden Record.

The inside sales director my company hired in 2010 was much like Gould in his demands–and his results. I’d never seen anyone ask IT for so many hardware and software conveniences. We joked that his desk looked like the bridge of the Starship Enterprise, with a phone that seemed to be made of buttons, a customised, super-ergonomic shoulder rest, and expensive plugins to our CRM that identified prospects, dialled their numbers, and told you what they had for supper last night. But once he started relentlessly dialling for dollars all day, our rapidly expanding prospect pipeline showed us clearly why he needed more tech than a Borg drone.

I could multiply examples all day, from Van Halen’s brown M&Ms to Ronaldo’s 6 meals a day to Churchill’s customised breakfast tray (with a cutout shaped to fit his belly). The point is that to get great results, you sometimes need to accommodate esoteric requirements. Far too often I meet highly skilled teams, tech and non-tech alike, forced to limp along without key resources like data pipelines or design skills or sales ops tools. They cite budget limitations or management inertia or similar excuses they’ve heard from those in charge, but the real reason they’re hobbled is that nobody’s bothered to learn the lesson from Gould’s chair: that starving your staff is as mad as asking the Beatles to play a capella.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 8th September 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

All over the UK, you find signs like these, which seem at first glance to be very courteous reminders about socially acceptable behaviour:

I’ve heard tourists cooing over these “polite notices” as charming examples of the cool reserve us Brits are known for, completely unaware of the ironic subtext: those posting the signs are hoping you’ll misread “polite” as “police”, with the not-so-subtle hint that the local bobby might haul you off to gaol if you don’t do as you’re told! I suspect this subterfuge was more effective back when actual “police notices” like this one were more common:

Non-inhabitants: You needn’t look up “cwt”, you’ll just find out its “eight stone” and be even more confused.

But the deliberate trickery lives on, with horse riders and motorcyclists sporting bright yellow jackets to fool passing motorists into leaving them more room on the road:

When I unmask these not-very-polite obfuscators for you, I’m not just providing a glimpse of a less salubrious side of the British character; I’m also pointing out that, like everyone else you know, you’re inattentive and slipshod by nature, which is why these sneaky sign-makers can fool you into seeing words that aren’t there. Indeed, you don’t even look at most of what’s directly in front of you, and right now your eyes are sliding right past most of the letters in the words on your screen. If you don’t believe me, ask yourself why you didn’t catch the deliberate typo I snuck by you earlier in this essay.

And it’s not just in noticing visual detail where Homo sapiens score poorly, either. For example, people are terrible at counting: manual vote tallies are notorious for their inaccuracy, and even without the pressure of a live audience, it’s impossible to enumerate for any length of time without loads of flubs. In general, if you need a rigorous result adhering to exacting standards, don’t give the task to a human. Our obliviousness is why eyewitnesses are the worst witnesses, and don’t even get me started on gorillas.

As usual, our limitations drive innovation: we’ve developed a host of adaptations to make up for our allergy to accuracy, such as checklists, punched cards, and blink comparators. These tools all work because they shift the burden of correctness from fallible humans to dependable machines. And since computers are the latest and best in this long line of defect-detection devices, you’d think we’d be smart enough to use them constantly to refine our products and find bugs.

Nevertheless, I see clients all the time who employ armies of humans mindlessly tapping at keyboards day and night, manually verifying the same product features over and over. Engineers know that skipping the unit tests makes coding slower, but still manage to fool themselves and their bosses into believing that automated verification is optional. Instead, put your people to work exploring the weird and remote corners of your product’s state space, and let the hardware do the hard work of checking what the software says.

I know of no better or faster return on your investment in expensive programmers than asking them to create and maintain a speedy, high-quality suite of tests; the payoff is immediate and huge. Do you know your software’s test coverage? Have you read through your verification suite lately? Do you have correctness checks running continuously? If you’re not sure, find out today, so you can stop paying humans to be incompetently fastidious.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 1st September 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

An old chess adage warns players against excessive caution in the face of apparent danger: “The threat is stronger than the execution.” It’s often incorrectly ascribed to the Latvian-Danish grandmaster Aron Nimzowitsch; he may not have said it first, but he did become famous for tying risk-averse players into knots as they tried to avoid the “ghosts” he created on the board. The mere possibility of a pawn being captured or a knight moving onto a fortified square drove his fearful opponents to take excessive defensive measures, and ducking the punch often harmed their positions more than simply accepting the blow.

Now most of you know I’d be the last person to suggest intimidation as an effective management tool, not least because I’m always trying to get people out of their fear-driven defensive routines and into curious, Model II thinking. But I’ve found over and over that revealing and highlighting real, external threats creates highly effective motivation without unduly scaring anyone or creating Nimzowitschian defensive errors.

For example, I’ve never seen Robbins-style “one more inch” speeches do anything to boost performance and engagement in tech teams; it’s far too obvious that the goal is artificial and the consequences inconsequential. That goes double for made-up deadlines and bogus “objectives” invented in the boardroom just to “give everyone something to aim at”, with no realistic demand to back them up. But when you have a public demo or a conference presentation or a regulatory requirement barrelling down the track and your team has to stop the train, then the drive is internal and powerful for those responsible. There’s both carrot and stick: winning will create joy and boost careers, but no one wants the humiliation of public failure.

This embarrassment engine is very powerful, but has to be harnessed carefully. First, the threat has to be specific and unchangeable; it’s no good if everyone thinks the delivery date can be finessed or the objectives met halfheartedly, rewarded with a pat on the head and a participation trophy. That’s why external deadlines work so well; when I was in e-commerce, every year in December I’d write to St. Nicholas asking for a few days’ grace, just until the 27th or 28th, but he never granted my wish–and that forced me to disappoint people helpfully to achieve the immovable target. (Check out my free event on real and fake deadlines to hear more such stories and solutions!)

Second, you need a small group that’s directly responsible for handling the issue, with a fully engaged leader solely focussed on the outcome. That’s why feature teams work so well and divide and conquer doesn’t: the data team can blame the back-end programmers, who throw up their hands and say the mobile developers delayed them, and so on ad nauseam. With all the needed resources and skills at hand in a single team, there’s nowhere to hide–and results, good or bad, go on the scorecards of those doing the work.

Finally, those scorecards need to be highly visible and frequently marked. Internal demos, product reviews, and beta tests are all good ways to be sure the sun is constantly shining on your technology. Conversely, if a milestone is reached in a forest and no one is around to notice, it definitely makes no sound–and your staff will likely conclude that the threat is unimportant and its consequences minor. This is one of the less obvious benefits of elephant carpaccio: delivering every 24 hours gives leaders the opportunity to reinforce the value of the work with great frequency and to steer the team with rapid, on-the-spot corrections.

It took work with dozens of clients to appreciate the value of well-harnessed embarrassment and recognise its absence, but now I see both everywhere: the established firm that dominates its market and can’t muster the gumption to enter new business areas; the keen startup in a crowded space whose staff are rewarded for safety rather than innovating to beat competitors; the tariff-hit company bouncing back by using AI creatively to meet revenue targets.

Open your risk register to find the actual threats you face, assign responsibility for tackling them to small, capable teams, and provide frequent, psychologically safe opportunities to be publicly accountable for mitigation. Put to use this way, the threat of an embarrassing result is not only stronger than the embarrassment itself, but positively fuels a culture of accountability and delivery. Where could you use it in your organisation?

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 25th August 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Three-year-olds have it right: keep asking “why” until you get to the bottom of the issue (or the adults start begging for mercy).

For instance, when my first technology organisation got too large for me to stay close to everyone, I started playing the “Why Game” with software developers. The rules for this two-player game are simple:

1. On the first turn, the Exec asks “What are you doing today?” and the Engineer answers.

2. On the second, third, and subsequent turns, the Exec asks “Why” and the Engineer answers. Typical “whys” include “Why are you doing that?”, “Why is that our goal?”, and “Why did she make that decision?”

3. The game ends when one of two things happens:

   1. The Engineer’s answer is about money: profits, costs, cash flow, company valuation, new markets, anything like that. Everybody wins.

   2. The Engineer answers, “I don’t know”. The Exec loses.

A typical game might go like something like this:

“What are you doing?” “Migrating data to a new database.”

“Why are you doing that?” “It was the next ticket to work on.”

“Why is that ticket high priority?” “It will help us speed up reports.”

“Why do we want faster reports?” “Users complain about slow responses.”

“Why do complaints matter?” “We don’t want to churn customers.”

“Why is churn important?” “We’ll lose recurring revenue.”

“Congratulations, we won!”

As you can see, a few rounds of the Why Game will get you a quick read on whether your tech team is working toward profitable outcomes–and more importantly, whether they actually know what those outcomes are. You can manage by objectives and shuffle your OKRs all day, but unless you’re constantly drawing explicit connections between your napkin strategy and what’s on the kanban board, there’s likely a widening gulf separating the profit you want from the features the engineers are giving you.

Relentless repetition of “Why?” is also the secret to making disasters useful, converting them into double-loop learning that can prevent a host of future issues. For example, the regulator investigating a Texas refinery explosion looked deeply into its triggers and found that the true root causes were cultural, affecting multiple BP facilities:

The Texas City disaster was caused by organizational and safety deficiencies at all levels of the BP Corporation. Warning signs of a possible disaster were present for several years, but company officials did not intervene effectively to prevent it.

And Nobel laureate Richard Feynman investigated the Challenger crash armed with nothing more than a willingness to keep digging for reasons, discovering systemic self-delusion among NASA leaders about the robustness of numerous systems, not just the booster seals. [For lots more on this, check out our recent Squadron event on root-cause analysis and the newsletter that led up to it.]

Speaking of Feynman, check out this video of him explaining why “Why” is both an excellent and an impossible question to answer, with detours into the nature of marital relations and the properties of ice, all in seven minutes:

What’s most interesting here for me is the way asking “Why” takes you “up” at least one level of abstraction, elevating the topic in a few steps from a mundane household accident to orbital mechanics or the identity of electricity and magnetism–much like repeatedly clicking on the first link of a Wikipedia page constantly moves you closer to Philosophy. Perhaps it’s this lifting to more general, strategic topics that makes “Why?” seem threatening in some situations. I certainly agree with Chris Voss that I wouldn’t want to create defensiveness by asking “Why?” if I were on the phone trying to talk down a terrorist holding hostages.

But for most organisations, I know of no faster way to get to useful truths than a well-placed and well-explained series of whys. Why aren’t you and your team asking them? And why don’t you do something about it?

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 18th August 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Two of my heroes, Jack Reacher and Richard Mitchell, agree on the importance of details. The placement of a single comma or apostrophe can totally alter the meaning of a sentence and change the direction of a company or a country.

In my own practice, I’m always coaching clients to look for subtle, highly significant signals in their own speech; my unsubtle method for doing this is usually shouting “STOP!” in the middle of one of their sentences to point out an error. And it’s paid tremendous dividends, particularly when the person I’m coaching becomes attuned to common thinking mistakes and begins automatically making corrections, as instinctively as when the doctor taps your knee.

Here’s a short list of reflexes to develop that will help you and your team tune your own radar to detect and fix dangerous cognitive distortions:

The error you make: Uncritically accepting an unverified “attribution”, that is, your belief about someone else’s reasoning or motivation.

Example: “The CTO is drowning in customer requests and must be frustrated by not having enough staff to fix them all.”

Reflexive response you can develop: Be sure to ask the person involved about what he or she is really thinking. If you confirm your attribution, you can use it with confidence (“I’ll approve hiring some more engineers to address the feedback”); depressingly often, your System 1 assumption turns out to be wrong, and you need to do something else (“Ah, you have enough capacity but you’re missing designers to come up with solutions; let’s get you some specialist help”).

Error: Prematurely ruling out options. Often signalled by language like “shouldn’t”, “can’t”, “not allowed”, or “impossible”.

Example: “We don’t have budget for another contractor, even though the project would have a clear 5x return.”

Reflexive response: Question the dichotomy. Rarely are there only one or two options to choose from, and even less often is one choice categorically forbidden (assuming you don’t have a gun to your head or aliens threatening the planet). Once you discover the actual consequences of the “unworkable” choice, you may very well decide you can live with them!

Error: Automatically appealing to authority, using terms like “best practice”, “right answer”, or “doesn’t meet the spec”.

Example: “The Spotify model is the standard way to organise teams, no matter how complicated it sounds.”

Reflexive response: Abandon betterism and the “wisdom” of crowds (“everyone else is doing it” doesn’t work any better here than it did with your mum). Instead, figure out specific reasons why the proposed action matches your strategy and fits your organisation.

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 11th August 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/

Photo: “Danger of Death“ by Teflon, licensed under CC BY-NC-SA 2.0 via Flickr.

Lisa and I bought a new house recently, and by “new” I mean that it’s only 500 years old, not 600 like our old house. One charming feature it came with is a sign above the basement stairs reading “DANGER OF DEATH” in huge letters, with an illustration that seems to show a person being attacked by a pointy, backwards “N”. The sign wasn’t too specific on which typographic monsters were lurking beneath us with intent to kill, but out of an abundance of caution we decided never to go downstairs.

Among the hundreds of engineering teams I’ve worked with are some whose products might merit similarly dire warnings. For example, one company’s software helps companies assemble rockets (the kind Mr. Musk loves to blow up), another ensures hospitals have clean equipment, and a third warns operators when heavy machinery is about to bash into squishy humans. In each case, a misplaced semicolon or untested edge case could have catastrophic results, and in fact that’s really happened to others in astronautics and medicine and automotive safety. Thankfully, these clients have all learnt the lessons of past disasters, and multiple layers of automated and human checks are in place to ensure bugs get swatted early. You can bet I don’t advise these folks to release daily or test in production!

Unfortunately, I encounter far too many tech teams with mostly harmless systems who nevertheless take “better safe than sorry” to ridiculous extremes. If, say, your software manages a warehouse full of fashionable handbags or tells bankers how the market’s moving, your engineers have no excuse for applying safety-critical standards, checking every feature six ways from Sunday, and taking months to get changes out to your customers. I worked with one team recently whose results are orders of magnitude more accurate than competitors’ right now, yet they’re spending months and months on polishing and perfecting—the very definition of diminishing returns.

When they trooped in for a first look at our new home, the specialist restorers in charge of making it habitable took one look at the scary sign and laughed. The only peril they could find was one loose board on the steps, which took all of 30 seconds to nail back in place. As they ripped down the warning, they remarked that very few houses, even the most ancient and neglected, have exploding wires or bottomless pits waiting to do in the unwary homeowner. Odds are the same is true of your technology–unless you’re giving out earthquake warnings or directing drones, stop protecting yourself against imaginary mortal peril and focus your engineers on discovering what users need and will pay for – that’s a hard enough task as it is!

This first appeared in my weekly Insanely Profitable Tech Newsletter which is received as part of the Squirrel Squadron every Monday, and was originally posted on 4th August 2025. To get my provocative thoughts and tips direct to your inbox first, sign up here: https://squirrelsquadron.com/