Blog

I Handed My 3D Printer to an AI for One Evening. It Missed Three Times - and That's the Best Part

The camera inside the printer showed an empty bed, the nozzle parked in the corner, and the first layer got going. I look at my phone and realize: no plastic is coming out. The nozzle is hot, the head is moving, and nothing is following it. The printer was honestly printing air.

That was neither the first nor the last time things went sideways that evening. Which is exactly why I want to tell you about it - not the glossy result, but how it actually came together.

Microsoft Has Nine Products Called Copilot. I Nearly Built the Wrong One Twice.

Count the products Microsoft currently sells with “Copilot” in the name: the free consumer chatbot, the paid enterprise add-on, Microsoft 365 Copilot Chat, a separate Business Chat that is not the same product, Copilot for Sales, for Service, for Finance, for Security, Copilot in Windows, and GitHub Copilot, which shares nothing with the rest except the name. Nine by the narrowest count. Closer to 80 once you count every variant.

The Instruction That Deleted the Wrong Twenty-Six Languages

I told my own site’s automation to shrink the blog down to four languages. Two days later I found out it had quietly wiped twenty-six languages off the homepage too, and neither I nor a single visitor had noticed.

The Trips I Never Took

The map said I had been to 39 countries. One of them was a 2014 safari in Madagascar. I have never set foot in Madagascar, and I was fairly sure I would remember a safari.

I had pointed a small AI pipeline at my photo library, roughly 68,000 images going back years, and asked it to draw where my life had actually happened. About 53,000 photos already carried GPS coordinates, so this looked like the easy part: plot the dots, connect the trips, colour by year. The result was beautiful, completely convincing, and wrong in a way that took me embarrassingly long to see, because nothing about it looked wrong.

I'm finally getting the most out of my Apple Watch data

I wore an Apple Watch for six years before I understood what the data was for.

It tracked my heart, my sleep, my steps — four thousand a day, most of them to the kitchen — and wrote all of it into a database I never opened. Then I started cycling to skip an always-late train, the gym stopped being a twice-a-year place, and one question began showing up before coffee every morning: go hard today, go easy, or rest? For years I answered by feel. Feel, it turns out, is a coin flip in a lab coat. So I sat down with Claude and built the thing that answers it properly.

The notes that were never quite true

A few months ago, a note in my vault almost talked me into a real decision. It was articulate. It was specific. It was beautifully formatted, with a confident little table and a number sitting right in the middle of it. And the number was wrong.

It wasn’t wrong because anyone lied. It was wrong because it was a summary of a summary of a summary, and somewhere four generations up that chain a figure had drifted, the way a story drifts when it’s retold. By the time it reached me it had lost all memory of where it came from. It sat in the same folder, in the same font, with the same calm authority as the bank statement two notes over. Nothing on its face said I am a guess about a guess. That was the whole problem.

The Atlas Hidden in Our Photos: What 67,608 Family Pictures Knew About Where We'd Been

I had 67,608 photos. Twenty years of them, the oldest dating to 2006, scattered across 73 different cameras and phones the family has owned since. They were all backed up, all “safe,” and completely unreachable. Nobody scrolls back through twenty years of pictures to find the afternoon in a particular town, because the act of finding is harder than the act of remembering. The archive wasn’t worthless. It was a second brain with no front door. So I built one, and pointed AI at the whole pile to see what was actually in there. The thing that came out the other end wasn’t a better search box. It was a set of maps I didn’t know I was carrying.

The Strategy That Could Never Say Sell

For about two months, a trading bot I run never once told me to sell. Buy ideas arrived all the time, dozens on some days. Sell ideas: none. I assumed the market was simply rising and the bot was reading it correctly. It turned out the bot was mathematically incapable of recommending a sale, and I had built it that way myself without noticing.

The Edit That Invented Its Own Sources

I was finishing a long master’s thesis, and the bibliography needed a tidy-up before the defense: stale entries, missing edition numbers, drifting formatting. So I did the modern thing and let an automated pass clean it. It came back faster and neater than I could have done by hand, every entry polished, every field filled. It had also quietly invented book editions that were never printed and stamped real sources with identifier numbers that point to nothing. It looked so clean that I almost shipped it.

The RAG capability you were about to buy just shipped free on every Mac

My Mac rebooted itself overnight. No crash report on screen, no warning the evening before — just a fresh login window and that faint sense that something happened while I was asleep. So I did the thing I now do reflexively, because I run a small AI agent directly on this machine: I asked my Mac what killed it.

The Second Time Is the Signal to Automate, Not the Fifth

I shipped a diagram that was wrong and didn’t notice for two days. It looked completely fine. The boxes sat in the right places, the labels were correct, and the lines connecting them pointed at the wrong boxes. On one too many hand edits, a connector had quietly slipped one slot over, and nobody caught it, because at a glance there was nothing to catch. That little mistake gave me a rule I now use far beyond diagrams: the moment to stop doing something by hand is the second time you do it, not the fifth.

The Report That Lied for Five Days and Never Threw an Error

Every morning an assistant I built sends me a short brief: what’s on today, what to watch out for, what needs a reply. For five mornings in a row it opened with some version of “nothing scheduled, free day at home.” I was on the road the entire time with a packed schedule. The brief was cheerful, confident, and completely wrong, and not a single alarm went off anywhere in the system.

The Junk Drawer on Your Phone: What 7,915 Screenshots Actually Contained

I had 7,915 screenshots on my phone. Five years of them, back to 2021: recipes, half-read articles, places someone recommended, apps I meant to try, boarding passes long expired. A pile that big stops being a memory aid and becomes landfill, because you never scroll back far enough to find the one thing you saved it for. So I pointed AI at the heap. The lesson that came out is the one I now use at work: split the job by difficulty, spend the cheap tool first and the expensive one only where the cheap one fails.

The Dashboard Said +158%. The Ledger Said Minus $2,149.

For about a month, the stats page of a trading side project I run showed a total P&L of +158.75%. This week an overnight audit of the same database produced a different figure: minus $2,148.88 across 140 completed paper trades. Both numbers were technically true at the same time. And I had built both of them myself.

When AI Is Confidently Wrong - Why You Need a System, Not a Gut Feeling

I’ve written about specific times AI handed me something confident and wrong: an invented citation, a report that lied for days. This piece is about the thing underneath them. Confidence and correctness are two separate dials inside these tools, and the machine turns the first one to maximum regardless of where the second one sits. Being careful in the moment doesn’t solve that. The real question: what do you build into how you work so the bad answer gets caught before it costs you?

I Copied Every Viral Web-Design Trick. I Got an Ugly Website.

You’ve seen the videos. “I built this stunning website in five minutes.” Slick result, three quick steps, a tool or two, done. I believed them. So I followed along: I collected the tricks, the templates, the “just do this” shortcuts everyone swears by, and stitched them together. The result was embarrassing. A page that looked like it was built twenty years ago, mismatched pieces, no coherence, the digital equivalent of an outfit assembled from whatever was on the floor. Each individual trick had looked great in its own video. Together they were a mess.

From Five Days to a Year With a Daily AI Assistant: What Actually Changed

Around Wednesday of my first week, the assistant lied to my face. A clean, specific, confident fact, dropped into a paragraph of things that were all true, stated with exactly the same calm certainty as everything correct around it. No hedge, no asterisk, no tremor in the voice. I almost let it through. That was day five. I’ve now used an AI assistant every single day for more than a year, and that Wednesday turned out to be the most useful thing that happened in the whole stretch, because it taught me early what most people learn late: fluency is not accuracy, and this tool will never tell you which one you’re looking at.

Why Most AI Demos Lie — and the 5 Questions to Ask a Vendor Instead

Every AI demo you’ve ever seen worked perfectly. Think about how strange that is. It’s not because the products are all flawless. It’s because a demo is built, on purpose, from the cases where the product shines: clean inputs, friendly examples, the happy path with the rough edges sanded off. A demo isn’t a test. It’s a performance. And if you buy based on the performance, you find out about the rough edges after the contract is signed.

The Site Was Perfect on My Phone. On a Three-Year-Old One, It Looked Broken.

I once opened something I’d built on an old phone — a tired, three-year-old handset on a weak connection, the kind I’d never normally test on. The smooth, animated, rather pretty thing I was proud of stuttered. It lagged on every tap, the animations dropped frames, and a page I knew was fine felt, unmistakably, broken. For a second I was annoyed at the phone. Then it landed: this isn’t my spare phone, this is my customer’s main one. On the device in their pocket, my polished work looked like a bug.

I Cut My AI Bill by ~75% Without Losing Quality. Here's the Decision That Did It.

The fastest way to overspend on AI is to use the most powerful, most expensive option for everything, including the boring 80% of work a far cheaper option handles just as well. That’s exactly what I was doing. Then I cut my monthly AI bill by roughly three quarters and, crucially, checked that quality held. No magic, no clever hack, just a handful of unglamorous decisions and one test that kept me honest. If you’re running AI across a team and the invoice keeps creeping up, this is the part worth copying.

Why I Took the CAPTCHA Off My Own Contact Form

For a while my contact form made every visitor tick a “prove you’re human” box before they could reach me. The spam still came. Not as much, but enough that I’d open my inbox to the same gibberish, now wearing a little badge saying a robot had certified it as not-a-robot. One morning, deleting yet another, I caught the absurdity: I was charging a toll to the exact people I most wanted to hear from, and barely inconveniencing the ones the toll was meant for.

My Night Shift That Never Sleeps, and the Guardrails That Let Me Trust It

Some of my most useful work happens while I’m asleep. Routine jobs run overnight, and I wake up to results: things tidied, summarised, prepared, waiting for me. It sounds like the dream of automation, and it is, right up until the moment something breaks at three in the morning with nobody watching. That’s the part the “automate everything” crowd never mentions: unattended work is only valuable if you can trust it unattended. Otherwise you’ve just built a machine that quietly makes a mess while you sleep. Here’s how I made overnight automation safe enough to actually rely on. It’s almost entirely the boring part.

The Assistant That Remembers Everything I've Ever Written Down

One thing quietly changed how I work, and it isn’t the part everyone obsesses over. Most AI tools are brilliant strangers: every conversation, they meet you for the first time. You explain your context, they help, and then they forget you completely. Next time, you start from zero again. It’s like having a genius consultant with no memory, impressive in the moment but exhausting over time.

The upgrade that actually mattered for me wasn’t a smarter assistant. It was one that remembers. It knows what I’ve already told it, what I decided last month, how I like things done. Memory, not raw intelligence, turned out to be the real win. And almost nobody talks about it.

How to Tell If an AI Tool Actually Works Before You Trust It

I almost rolled out a tool that was, in plain terms, guessing. It passed my test perfectly: every example, right. Then I pointed it at actual, messy, day-to-day data, and it fell apart so badly I had to read the number twice. The tool wasn’t broken. My test was. That gap is the most expensive mistake I see people make with AI right now: trust the demo, roll it out, find out weeks later it only ever worked on the easy stuff. Here is how to catch it before it costs you.

Running a Website in 30 Languages Without Embarrassing Myself

My website is available in thirty languages. Here’s the uncomfortable truth. I can’t read most of them.

Machine translation has made “go global” almost effortless. A few clicks and your site speaks Japanese, Arabic, Finnish, Greek. Then comes the part nobody warns you about. The translation is fluent enough to look completely fine, and wrong often enough to embarrass you badly, in a language you have no way to check. You’re publishing your professional reputation in thirty languages and personally vouching for the quality of maybe two. That gap is a real trap, and here’s how I try not to fall into it.

My AI Wrote a Confident Lie. The Fix Was Management, Not Prompting.

The post was ready to go. My agent had written that week’s update for one of my channels — fluent, confident, properly formatted, the kind of thing I’d normally skim and send. My thumb was over the button. Then one line snagged me: it called a strategy “profitable,” cheerfully, with a number attached. The strategy was in alpha. The number was invented. Three more seconds of not-reading and I’d have published a clean, polite, completely false claim to a channel of real people.

Building an AI Agent That Actually Works in Production

The demo worked perfectly. Of course it did: I’d run it five times before showing anyone. Two weeks later, at some unreasonable hour, the same agent got stuck calling the same broken tool over and over, burned through its budget, and produced nothing at all. That gap, between the version that wows a room and the version that quietly falls over when no one is watching, is the entire story of putting AI agents into production. I’ve now run mine, Tree AI, every single day for over a year. Almost everything I actually know about agents I learned in that gap, not in the demo.

Everything I Depend On Runs on One €6 Box

One afternoon I finally added up the small monthly charges I’d stopped noticing: a workflow tool, an analytics subscription, an AI bill, a notes app that had quietly started charging per seat. The total had crept past €80 a month. Then my eye landed on a line I’d been paying without thinking about it at all: a small server, €6 a month, mostly idle. The punchline was uncomfortable: almost everything on that €80 list could just live on the €6 box.