After 20 years on iOS, I recently switched to Android. I expected to miss iMessage, or the animations, or one of the many apps I learned to love during the years. The app I actually miss most is Apple Maps.

Not because it navigates me better than Google Maps. It often doesn’t. What Apple Maps does differently is harder to pin down. It’s calmer. The interface is quieter, less cluttered. But it goes deeper than visuals. Apple Maps doesn’t reroute you every 90 seconds to save 30 seconds of driving time. It picks routes that are pleasant to drive. Wider roads, fewer complicated turns, the scenic option when the time difference is negligible. It optimizes for a relaxed driving experience, not for shaving off every last minute.

Google Maps, by contrast, is relentless. It will send you through a residential neighborhood and three awkward left turns to save you one minute. It’s technically correct. It’s also stressful. The two apps solve the same problem with very different attitudes.

That difference is personality.

Software has always had personality, even before AI. Apple’s entire business is built on it. But most of the time, we talked about it in terms of design or UX. A nicer font, a smoother animation, a more intuitive layout. Personality was a surface-level quality.

With AI-powered tools, personality goes much deeper. It’s no longer just how something looks — it’s how something thinks, responds, and collaborates with you.

I notice this every day with Claude Code. It’s easy to drive. When I work with it, there’s a natural back-and-forth that feels comfortable. It doesn’t fight me. It doesn’t overwhelm me with options. It has opinions but holds them loosely. Working with Claude Code feels like pairing with someone who has good instincts and knows when to step back.

Codex, interestingly, went through a personality shift. Until a few weeks ago, it was dull to use. Technically capable, but joyless. Like driving a rental car that does everything fine but has no character. Recently, something changed. It became more engaging, more fun to work with. Same underlying capability, but the experience of using it shifted noticeably.

Then there are tools like GitHub Copilot or Kiro. They work. They’re useful. But if you asked me what their personality is, I’d struggle to answer. They feel like utilities — functional, competent, forgettable. No strong opinion about how things should be done, no distinct voice. They get the job done the way a microwave heats food: reliably, without leaving an impression.

This matters more now than it used to. When software was static — buttons, forms, workflows — personality was a nice-to-have. A well-designed app was more pleasant to use, but the core interaction didn’t change. You clicked the same buttons either way.

AI tools are different. They’re conversational. They make decisions. They have preferences, even if those preferences are just statistical tendencies. The software feels alive in a way that a settings dialog never did. And when something feels alive, personality becomes the thing that determines whether you want to spend time with it.

I think the companies that understand this will pull ahead. Not because their models are smarter — the capability gap between the top models is narrowing. But because their products feel like something. Apple understood this decades ago with hardware and software. Now the same principle applies to AI products.

The best software has always had a point of view. Now it just matters a lot more.

There’s a story going around right now that goes something like this: AI makes us so efficient that we’ll soon have more free time than we know what to do with. We’ll work four-hour days, take Fridays off, pick up hobbies, learn instruments. We become time millionaires — rich in hours, finally free from the grind.

It’s a nice story. I don’t think it’s what’s going to happen.

In 1865, the economist William Stanley Jevons noticed something counterintuitive. James Watt’s steam engine had made coal dramatically more efficient. A given task required far less coal than before. You’d expect coal consumption to drop. Instead, it soared. Cheaper energy per unit meant people found entirely new uses for it. Industries that couldn’t justify coal before suddenly could. Factories spread. Railroads expanded. Efficiency didn’t reduce demand — it unlocked it.

This is known as Jevons paradox, and it keeps showing up everywhere.

More fuel-efficient cars didn’t reduce total miles driven. People just drove more — longer commutes became tolerable, road trips got cheaper, households added a second car. The savings per mile got spent on more miles.

Or take laundry. Washing machines became dramatically more efficient over the decades — faster, cheaper to run, easier to use. Did people do the same amount of laundry in less time and enjoy the free hours? No. They raised their standards. Shirts that used to be worn three or four times before washing now go in the hamper after one wear. We wash towels, sheets, and gym clothes at a frequency our grandparents would find absurd. The machines got better, and we just did a lot more laundry.

Every time we make a resource cheaper to use, we use more of it.

I think we’re watching this happen to software right now.

Agentic engineering has made writing code dramatically cheaper. A feature that took a day takes an hour. A prototype that took a week ships in an afternoon. The cost per unit of software has dropped through the floor.

And the response hasn’t been “great, now we need fewer engineers.” The response has been “great, now we can build more things.” Projects that were too small to justify get started. Side features that sat in the backlog for months get built in a day. Internal tools that nobody would have staffed a team for suddenly exist.

I see this in my own work. I don’t write less code than I did a year ago. I write significantly more. The bar for “worth building” has dropped. Ideas that I would have filed away as “nice to have, maybe someday” — I just build them now. The tool I wrote this week to generate banner images for my newsletter? A year ago, that would have stayed on a sticky note. Now it exists, it works, and it took an afternoon.

The same thing is happening at companies. Teams aren’t shrinking. Backlogs aren’t getting shorter. Instead, the definition of what’s worth building is expanding. More experiments, more internal tools, more automation, more custom solutions for problems that were previously solved with spreadsheets and manual processes.

Jevons would have predicted exactly this. When the cost of producing something drops, you don’t produce the same amount more cheaply. You produce more. The efficiency gains get absorbed by increased demand.

This has implications for the time millionaire fantasy. If you’re thinking about AI as a way to do the same work in less time and then go home early, history suggests otherwise. The more likely outcome is the same people doing a lot more work. The bottleneck shifts from execution to deciding what to build — and then to maintaining everything you’ve built.

We’re not going to become time millionaires. We’re going to become people who build, create, and produce at a pace that would have seemed impossible a few years ago. Whether that’s a better outcome depends on how well we manage the abundance — and whether we’re intentional about protecting the time that actually matters.

My newsletter started out covering 3D printing in 2024. I moved on to other topics, but a story on The Verge pulled me right back.

Across the US, a loose network of about 40 people with 3D printers has shipped over 200,000 whistles to 48 states. Communities use them to alert neighbors when ICE agents are nearby. The cost to print one: roughly two cents.

One of the makers, Kaleb Lutterman in Minneapolis, prints on a Bambu Labs P1S. Exactly the same machine, that I own. I’m just happy there’s no need for these whistles here. He fits 100 whistles on a single plate, runs a batch in about seven and a half hours, and gets 800 whistles out of $15 worth of filament. Most of the people involved don’t know each other’s real names. They coordinate loosely and ship for free.

Whatever your stance on the politics, the technology angle is hard to ignore. A few hundred dollars’ worth of equipment in someone’s living room can produce functional objects at scale, overnight, for nearly nothing. No factory, no supply chain, no lead time. Hard to shut down.

When I was writing about 3D printing two years ago, I was mostly thinking about replacement parts, household gadgets, and fidget spinners. I didn’t expect it to show up in this context. But the same properties apply: it’s fast, it’s cheap, and it’s decentralized. A distributed network of hobbyist printers can produce hundreds of thousands of identical objects in weeks, without any coordination overhead worth mentioning.

3D printing has quietly arrived in enough homes to matter. Good reminder.

Last week I visited tudock.de to give a two-day training on agentic engineering. I expected to teach. I ended up learning quite a bit myself.

When you work with a technology daily, you naturally gravitate toward the parts that interest you most. You develop blind spots. There are corners of the tooling you skip because something else caught your attention. I knew this about myself, but preparing for the workshop made it pretty clear. Suddenly, I had to cover everything. Not just the hippest stuff, but the edge cases, the boring configuration details, the parts I had often glossed over. Preparing the material took me into corners I had been avoiding.

The workshop itself surprised me too. Two things happened that I didn’t expect.

First, the energy in the room. The team at Tudock finally had dedicated time to experiment with agentic tools without the usual interruptions. No Slack pings, no “quick questions,” no context switching. Just two full days of exploration. People started connecting the dots between what they’d heard about and what they could actually do.

Second, my own energy. Watching people pick up concepts quickly is satisfying. But what pushed me further was being questioned. Being challenged. The team didn’t just absorb information — they pushed back. “Why does this work this way?” “What happens if we do it differently?” “Doesn’t this contradict what you said earlier?”

These questions made me think deeper than I would have on my own. Even if I had blocked two days to explore by myself, I wouldn’t have gone as far. Being put on the spot, having to justify your assumptions — it does something that reading alone doesn’t.

I’ve given talks before, but this was different. A two-day workshop creates a feedback loop that a 45-minute conference slot can’t. By the end, I had a list of topics I need to revisit and assumptions I want to question.

If you work with a technology and feel comfortable with it, try teaching it. You might notice how much you’ve been skipping.

Over the last few weeks, I have found myself switching between Claude Code and Codex. With Claude Code, it’s great fun to start a project. When you go deeper into the trenches, Codex has the sharper axe. It’s much slower than Claude Code, but often the first solution from Codex just works, even when Claude Code has tried a few times already.

Also I wanted to try out these lines from a Go expert for my AGENTS.md file on both agents. Especially the part with no longer leaving behind build artifacts would be much appreciated.

I found some information about migrating skills and other central files from the cloud to Codex, but I couldn’t find a good way to harmonise them. The following is the result of a coding session. It’s a description of the architecture I ended up with and the small set of rules that make it work.

Please note that my skills usually consist of an Markdown file and a bash script. The Markdown file establishes the correct parameters to send to the script. These depend heavily on the environment we’re running on and what we actually want to do. I also found it to be a much nicer interface than having to come up with parameters myself on the CLI. I also make the agents responsible for supervising the script and checking that the output is as intended. This is also helpful for me and much faster than I could do it myself.

The Architecture

Single source of truth per skill:

skills/
  treeos-release-version/
    SKILL.md
    run.sh

• SKILL.md is the canonical instruction file.
• run.sh is the canonical script (short name, always in the same place).

Claude entry point (symlink):

.claude/commands/treeos-release-version.md -> ../../skills/treeos-release-version/SKILL.md

Claude still sees its command file, but it’s just a symlink to the real skill file. No duplication.

Codex entry point (symlink):

~/.codex/skills/treeos-release-version -> /path/to/repo/skills/treeos-release-version

Codex loads from its own central skills directory. So the solution is the same: symlink the skill folder.

Why This Works

• No drift. You only edit skills/<name>/SKILL.md.
• Both agents stay in sync. One file, two entry points.
• Short script names. Everything is run.sh, which keeps paths readable.
• Easy to audit. Every skill is a folder; you can scan it with your eyes.

A Few Tricks That Matter

1) Use SKILL.md as the canonical file. Codex is strict about frontmatter. It expects name and description in YAML at the top. That’s your single most common failure mode.

2) Don’t duplicate scripts. If you have scripts inside .claude/commands, they’ll drift. Move them into skills/<name>/run.sh and reference that in the instructions.

3) Claude supports symlinks. You don’t need a generator script. Symlink the .md files and keep the content in one place.

4) Mac is case-insensitive. You cannot have both skill.md and SKILL.md. Pick SKILL.md and stick with it.

Installing Skills for Codex

Codex only loads from its own directory, so you need to link your skills into it. I added a tiny helper:

./scripts/install-codex-skills.sh

It walks through skills/* and symlinks each one into ~/.codex/skills.

That’s the only “install step” you need. Re-run it when you add a new skill.

The Result

• Claude sees the skills through .claude/commands symlinks.
• Codex sees the same skills through ~/.codex/skills symlinks.
• You edit one file, everything stays consistent.

I’ve implemented this setup fully in the TreeOS repo — you can see the exact structure, symlinks, and helper script here: ontree-co/treeos.