[7/31] 2026 Goal Progress

This is my regular check-in on the goals I've declared. This week, I came up with a quit-smoking device — "a box that only opens once an hour" — and got way too excited about it all by myself.

Health

Gym still on hold. Rehab at the osteopath

Got the stitches out of my hand. But since I've been told not to put much weight or load on it this week, no gym for me.

I can still go to the osteopath though, so I've been doing rehab there and picking up some knowledge about how my hand is doing.

Also, the air conditioning is giving me a bit of a headache. The weather has been all over the place — suddenly cold, then thunderstorms — so I've been feeling slightly under the weather.

Staying up late. But not sleep-deprived

Lately, I've had more nights where I go to bed around 1 to 3 a.m. The main culprit: the drama "Silicon Valley," which I started watching last week. Plus the pattern where I think "I want to do a little something tonight" and suddenly it's the middle of the night.

Mornings never feel great no matter when I go to bed. Maybe it's AC sickness, or maybe the fan is making the room too cold — no idea. But a morning bath pretty much resets me.

I'm getting 6-7 hours every night, so I'm not sleep-deprived. Just a little off-rhythm, I'd say.

Quitting smoking: total failure. So I'm building "a box that only opens once an hour"

Quitting smoking is going terribly.

But here's the thing — I had an idea: use my tech skills and devices to make quitting fun.

Instead of stopping cold turkey, I want to cut down the number of cigarettes. And then I realized: the reason I can't cut down is that my IQOS is always within arm's reach.

If I just put it on top of the bookshelf next to me after smoking, then when I think "alright, time for a smoke," it's "oh wait, it's not here" — "too much hassle to go get it, never mind." A cigarette I would've smoked if it were nearby disappears just by making it feel slightly annoying.

Using this insight, I'm going to build a time-locked box that only opens once an hour, with a 3D printer, my own code, and IoT.

I actually own three IQOS devices, so I'll sell two of them on Mercari or wherever and keep one. Lock that one inside along with the cartridge pack. One-hour lock. An hour later, it pops open.

The pop-open will probably feel so rewarding that I'll smoke. But even so — even if I stayed awake 24 hours, I could only smoke a pack and change.

Right now, I'm awake and smoking across roughly 14-15 hours a day, and going through about two packs (40 cigarettes). 40 divided by 15 is 2-3 cigarettes every hour. And it's actually concentrated into 6-7 daytime hours, so that's 6 or 7 an hour. Pretty abnormal, honestly. With this box, I'd be forced down to about 8 cigarettes over 7-8 hours. I think it's kind of revolutionary.

On the 3D printer side, since there's electricity running through it, I need to make sure the filament structure doesn't melt — I'm designing all that right now, bouncing ideas off AI. It's really fun.

Work

AI is just the assistant. Responsibility stays with humans

This week's highlight: a moment came up where I had to face, once again, what development in the with-AI era really means.

Over the past few months, I've focused on building a situation where AI just does all kinds of things on its own — and it worked. But the "is this actually releasable?" part still very much needs humans.

AI output is still readable when it's code. But when it summarizes a code review in Japanese, the sheer volume of text kills any desire to read it. And since code quality is mostly covered now, we tend to slack on the things we did as a matter of course before AI: properly testing locally, properly checking staging, catching up on the domain knowledge we need. The stuff that used to be table stakes.

The result: the rate of small bugs in production goes up. I experienced this phenomenon firsthand.

So I shared this with the company — and with myself — as a warning: AI is just the assistant; responsibility stays with humans.

I also bought Ushio-san's recently released book, and it said pretty much the same thing. It pushed hard on "understand things properly — actually read the code and understand it." So it seems I'm heading in the right direction.

The smarter the model, the noisier the AI

This is probably everywhere on the internet already, but AI — Claude especially — tends to get noisier as the models get smarter. Opus5? So noisy.

The reason the text balloons is a classic Japanese-language thing: the moment it switches to polite speech, the word count explodes. "Shall I go ahead and do this for you?" — just say "doing this." Done.

Here's the funny part. If you tell the AI "drop the polite speech," it starts calling me "omae" (rude "you"). Tell it "stop with the omae," and it goes full drill-sergeant: "Do it now." "Here are the three things you will do." Like, who is this guy? Then when I say "please stop the commanding tone," it snaps right back to polite speech.

So I gave it several samples along the lines of: "Convert to phrasing that doesn't hurt people. Say 'let's do this' — that's neither polite speech nor a command, just soft communication." That stabilized it pretty well.

Designing around "I don't feel like reading this"

Code reviews and GitHub Actions comments spew out an absurd wall of text. Humans should read all of it — but "I don't feel like reading it" is a separate problem. How you present the stuff nobody feels like reading — that experience design is what this era demands for the engineer's developer experience.

Especially for junior folks: AI throws out words they don't know, and catching up on those matters. We all used to do that as a matter of course, but AI moves so fast that we get rushed, skip it, and go "well, I ran it and it worked, so minimally OK I guess." I suspect a lot of companies have members doing exactly this.

AI accuracy will keep going up thanks to the US and China competing. So the genuinely interesting part is how to efficiently design output that humans find easy to read and easy to understand. "This is readable because we're Japanese" is not a universal foundation — harnessing things specifically for Japan is interesting.

Concretely, imagine a code review like: "I've summarized why this code is safe in slides. Here's a link to a video where I explain it. Runtime: one minute." You play it, go "ah, got it," grasp the overview — and it explains "touching this would be a breaking change, so I created a duplicate method as a temporary refuge and left a comment to delete the old method after release," "tests cover these edge cases."

In mob programming or pair programming, when someone asks "why did you write it this way?" — being able to explain on the spot, writing explainable code, that's important, right? Having humans guarantee that the AI can properly explain itself too. It's a new timeline, and honestly, never boring.

Also — reading text hits differently depending on your condition that day. Some days you can read everything; some days you can't. Sometimes your eyes hurt. By using more senses — ears, eyes — you raise the odds that things actually get through. My gut says that's how you get stable performance. Just an idea, but it's there.

Money

US stocks finally back in the black

The US stocks I've been holding came up a bit and finally went green. Bought at the start of the year, in the red the whole time, and finally — green. ...Well, it's back in the red again now.

The newly added stocks have been climbing nicely, and I'm scouting for the right moment to take profit. Pretty smooth overall, I'd say.

The trading bot melted all of June's gains

The auto-trading bot has been almost comically bad. July was a disaster — it melted all of June's gains and then dug further into the negative. Which is, honestly, kind of fascinating.

But continuity is what matters here. Why was last month so good and this month so bad? Is it the character of the market? Can I engineer something that endures that character? I bounce these questions off AI and keep refining, every single day. If I can truly keep this up for a year, that alone would be worth it.

My knowledge of trading, finance, machine learning — it genuinely feels like it grows a millimeter every day. Good trend.

Zero progress on moving abroad. But Duolingo: day 211

Preparation for moving abroad: zero progress.

But Duolingo is going strong. 211 consecutive days of study now. Listening to English every day really does matter. Speaking too — though lately I've been slacking on the speaking side, the Video feedback part.

At minimum, I do one step a day because losing the streak would sting. I'm fairly sure that if I lost the streak, I'd cancel Duolingo on the spot — which is kind of funny to know about myself.

Everything Else — Paid off one of life's "debts"

The most striking thing this week. I can't get too specific about it.

At the start of the year, I set a goal: in 2026, steadily pay down the debts I've been carrying through life. And this month, after six months of work, one big debt finally reached its conclusion. Genuinely positive. Genuinely good.

The hand injury is part of the same story. Twenty years of baseball built up damage, and this surgery removed it. Face the debts you've left unattended, and clear them one by one. There are plenty more, but this year things have gotten remarkably clear, and the TODO list I wrote in January is getting checked off like crazy. It feels great.

What's amazing about AI: it just keeps going

Also — I get that AI makes everything more efficient, but what's truly amazing about AI is that it proceeds steadily, without drama. Seriously, this is incredible.

A human goes "alright, here's today's TODO list, top to bottom" — and burns out around item three, and suddenly you're lying on the bed. AI just works through the entire list until it's done.

Everything humans don't need to do, I already delegate to AI. But human work remains. So for that work, learn from AI: no complaints, whatever mood you're in, just steadily do it.

And humans are truly single-threaded, so crush tasks one at a time. Keep that in mind and you'll find you've made surprising progress. Once the TODO list is more than 50% checked, it's all downhill from there (past the summit — nothing left but the descent), so you can ride that "let's just crush the rest" feeling, motivate yourself, and finish the day's minimum tasks completely.

This kind of thing matters more than you'd think.

Closing

Next week, it's already August. It'll be hot, so health management continues. Honestly, some of it is beyond my control — but I still want to stay conscious about managing my condition and keeping up hygiene.

My hand is healing up pretty well, so instead of weights: core training and stretching. Plus, I promised my son we'd start parkour together, so next week I'm focusing on preparing for that.

As for work — I'm the type who talks way too much when I get fired up, and sometimes I put people off by conveying too much passion. As CTO, my position calls for proper word choice, proper delivery. High heat can turn a meeting pushy, so: watch myself from a bird's-eye view, stay calm. Doing it self-taught won't change anything, so I'll pull in outside knowledge — books and such — and keep reviewing myself.

That's the week I want to have.