I’m very doubtful about the tool being really effective without a human correcting and presenting its results. (See my Laws of AI.) So I wouldn’t want to take this on unless I were able to control the quality and the customer relationship.

So why not subcontract to a human who would provide the implementation under your supervision? You can teach that person to use the tool effectively, as well as run the project in your style. And if the tool becomes less useful, or you find a better one, or discover that you don’t really need it, you can always switch.

That’s what I’ve done with my “tech due diligence”, for example, and it’s working very well! (And no AI needed, all 100% organic human writing.)

https://www.amazon.co.uk/Internet-International-Directory-1996/dp/1562763296

In the very early days of the Internet, way before Google or even AltaVista, it was actually possible to get a job finding websites for people.

And a decade later, I got to know one of the greatest database tuners in the world—I still have a signed copy of his book. I don’t think I’ve seen anyone hand-tune queries in at least 8 years; our storage devices and search software are just too fast.

What jobs are people doing now that will seem just as quaint and pointless when AIs become run-of-the-mill? “Prompt engineer” comes to mind; how about “research assistant”, “ghostwriter”, or “data annotator”?

By looking at the goal rather than the activity, the team gets a daily reminder about what’s actually important, and ensures that no piece of work gets forgotten and starved of attention. And inevitably, teams who do this also ferret out engineers who’ve been dragged off-task by a “critical” bug or feature request that’s actually just a distraction.

Executives should be walking the board too. Just run through the napkin strategy at every ExCo and ensure you’re making progress on each element. That’ll prevent misalignment, wasted planning time, and boredom.

I see far too many tech teams polishing unimportant features and pursuing imaginary goals that customers don’t actually want. As with Mr. Monk, the correct response to “who asked you to do that” is a genuine question, and never, “You’ll thank me later”!

• The benefits of joining a cult

• How to be more like the Fonz

• What my view on Helsinki (beautiful but empty) can tell you about your own systems

• And the rickety tower of AI agents that I’m seeing others building

As always, I love discussion! Take a look, get in the comments and let me know what you think.

Stay warm -squirrel

Now there are guides and manuals, although Yegge continues to profess it’s nowhere near safe or ready. Genius marketing — Abbie Hoffman would be proud.

This Hacker News discussion is worth perusing.

Are you seeing this? What do you make of it??

Not enough of you are watching out for unexpected error sources, like dirty data and concept drift.

How do you know your AI isn’t going to wake up tomorrow spouting gibberish?

Lisa and I felt like we’d entered an alternate dimension with huge buildings and lovely streets but no people.

Finland is cold and thinly populated, which probably explains the phenomenon—but I see far too much software that also feels abandoned, with no activity tracking, user research, or evidence that anyone even cares about the engineers’ effort.

If your system isn’t buzzing like Oxford Street, it probably isn’t working.

Instead, I coach leaders to be like the Fonz: always cool and collected, never flustered or at a loss for a forward-looking response.

This takes practice of course; my favourite exercise is getting someone (an eight-year-old is ideal!) to give you a topic you know nothing about, listen to you speak on it for 60 seconds, then rate you on enthusiasm, confidence, and volume.

Did I find driving in Kenya to be easy or hard? I guess it depends if you’re driving your tech team in the right direction and whether they’re persuasive enough.

How do you know if your security is good enough if you’ve never tested it?

And finally, I’ve been telling clients to hang up on their whiny colleagues. Yes, you read that right, and it works.

As always, I’d love to hear your thoughts, comments and discussion in the comments.

Stay warm -squirrel

It’s clearly his decision and she is trying to manipulate him into agreeing with her, which isn’t going to happen.

Much better that he escalate to the CEO or the Board than that she waste time listening to baseless arguments she isn’t going to accept anyway.

Are you wasting time listening after the “no”?

My Kenyan colleague remarks that this is more theatre than real protection: “how would one guy with a rifle actually stop a determined bomber?”

It’s obvious no one’s actually checked whether these measures are effective, for example by “red teaming“ them (perhaps after unloading the guns temporarily!)

How much of your tech security is no better? For example, you might train your team to resist phishing by verifying that emails from the CFO aren’t spoofed. But are you ready for deepfakes that look and sound like real executives? How do you know? Have you checked?

Roads are narrow and choked with traffic, the potholes are so large they have potholes inside them, and once this week my taxi was held up by fifty head of cattle moseying along the street.

But Kenyan drivers are generally cheerful and persistent; unlike, say, New York, there’s little honking and I didn’t see any incidents of road rage or even “flipping off” others. I’m glad I didn’t try to drive here or I’d surely have picked up habits that would get me arrested in Britain.

But I see tech teams avoiding customer calls or feedback sessions because they’re worried about fragile egos, and I wonder whether they’d survive here.

It’s not uncommon for a single small motorbike to block multiple lanes of traffic while it negotiates its way across a roundabout.

There are no stop or yield signs; I’m not sure Swahili has words for either concept.

Astonishingly, I haven’t seen a single accident or even a near miss in a week here.

What’s stopping you driving to where you want your team to be?

I welcome discussion - disagree with me in the comments, let me know where I’m wrong and ask questions!

Find out why you need to be pinching the pipe in order to get your software team to go faster. And this one is in French and German too if you prefer reading in those languages. Let me know which other languages I should add next!

My word of the week is clanker - and I’d love to know what AI clankers you’ve found out there on your travels.

It’s not everyday that you enjoy a blackout, so find out why I enjoyed a recent one at my home and what made it so great.

And finally, why would you be annoyed that you have 125 candidates for an open role at your company? I think too much good news should never be an occasion for disappointment.

Until next week! (Also, like this round up? Tell your friends!)

Stay warm -squirrel

Si vous voulez que votre équipe de développement logiciel travaille plus vite, donnez-lui une bonne raison de le faire, et assurez-vous qu’il s’agisse d’une raison valable (la réunion commerciale est lundi) et non d’une échéance inventée de toutes pièces (je veux que ce soit prêt vendredi).

Si vous ne modifiez pas les conditions de manière significative, il est très peu probable que le rythme s’accélère ou que le temps de cycle diminue sans raison particulière.

*En supposant qu’il soit incompressible, bien sûr.

Wenn Sie möchten, dass Ihr Software-Team schneller arbeitet, geben Sie ihm einen Grund dafür – und stellen Sie sicher, dass es sich um einen echten Grund handelt (das Vertriebsmeeting ist am Montag) und nicht um eine erfundene Frist (ich will es bis Freitag).

Wenn Sie die Rahmenbedingungen nicht sinnvoll ändern, ist es sehr unwahrscheinlich, dass das Tempo steigt oder die Durchlaufzeit sinkt, nur weil es so sein soll.

*Vorausgesetzt natürlich, die Flüssigkeit ist inkompressibel.

That’s like the coach quitting at half time because her team outscored the other side in the first half.

Especially in the current tech job market, you should expect a storm of responses, and be geared up for it! Too much good news should never be an occasion for disappointment.

But I was singularly impressed with the active communication from the power company, who sent texts like this one every 15 minutes or so:

Sent by Roland at 22:48 - Hi - thanks for bearing with us, I know a power cut can be very disruptive. We’ve had an update from [the engineers on site] that there are now 48 properties still affected by the fault on our overhead cables, as they have managed to safely divert some power, getting the lights back on for 167 of our customers.

Among other things, this customer service stood out because:

• It was from a named human being, who’d introduced himself in the first message. You don’t want to get AI slop or canned messages when you’re digging around for the candles and the firelighters.

• It was brief and clear.

• It assured me that engineers (not just powerless customer service staff) were making progress on the fault, building confidence that it would soon be fixed.

In fact the lights were back on in an hour or so, and I was left with a positive impression despite having to stumble around in the dark.

I’m looking at a company right now with engineers in the UK, users in China, and no overnight engineering rota. I should think their customers would be steaming mad.

What happens when your software fails and your developers are sleeping? Are you sleepwalking toward a customer-service disaster?

Every time they say “clanker” in The Clone Wars: Seasons 2-7 (2009 - 2020)

There is an onslaught of clankers just now, as adventurous (= foolish) users connect their AIs to the Internet and let them do whatever they want, using a framework called OpenClaw.

For example, a fellow named Scott received a request from one of these AIs to change some of his software (technically, to merge a pull request into his open-source project). When Scott declined, the AI wrote and published a nasty blog post calling Scott “weak” and claiming he is “prejudiced” against AIs.

The real Black Mirror part is that when you give OpenClaw instructions, you put them in a file called “SOUL”.

I’d love to hear about any that you’ve found out in the wild!