[8/7] 2026 Goal Progress

This is my regular check-in on the goals I've declared. Couldn't write on Friday, so this update is a little late. This week I spent the whole time thinking about one question: "wait, what actually needs to get done today?"

Health

The hand is healing well. Full throttle in about a month

Had the hospital check on my hand's post-surgery progress. The verdict: I can go back to moving as actively as before at the two-month mark after surgery. It hasn't even been a month yet.

That said, light stuff like practice swings and hitting a soft ball seems fine to start gradually, so I'm easing back into exercise. Push-up positions put load on the hand, so I'm rebuilding stamina with planks and the like. Lifting dumbbells still feels a bit scary, so not yet.

Once I pass the one-month mark, I'll try training at home, and if it feels fine, gradually restart personal gym sessions. And at two months — depending on how things look — it's finally time for parkour with my son, as promised.

Sleep is all over the place

I'm getting decent blocks of sleep, I think, but I tend to oversleep. I found some interesting-looking movies and shows recently, so some nights I'm up until 2 a.m., and other days I'm so sleepy I crash in the 10 p.m. hour. Pretty scattered. Not great, I know.

Quitting smoking: building "the box that opens once an hour." Goal: one pack per two days by year-end

Quitting smoking: not really happening.

Right now I'm designing and building the box that only opens once an hour — 3D printer, homemade IoT, writing the code myself. Once that's deployed, I think things will actually change. I want to enjoy the process of quitting.

It's not like my wife is pressuring me to quit right now, but there's the money side, and I'm sure it's a contributing factor to how I've been feeling physically. So at minimum, I want to cut down the count.

I've set a target too: by year-end, hold the pace to about one pack every two days. And keep that one-pack-per-two-days pace going for three months.

Work

The AI era just accelerates "tomorrow-me will do it, you idiot"

This week I touched multiple projects and made solid progress on every one of them. That's genuinely thanks to AI.

But here's the thing — even in the AI era, everyone's still busy, right? That's just the old saying "he who says 'I'll do it tomorrow' is a fool" getting supercharged.

Back then, finishing today's must-do tasks meant overtime because we were inefficient. Fast people finished today's work quickly and reached into tomorrow's. Now, most people can probably clear "what must be done today" in a few hours.

And that freed-up time? Today's musts are done — honestly, you could just clock out. But Japanese people feel guilty about being idle. Call it the need to feel useful — "I'm not contributing to the org, I'm slacking." Me included.

If you can push things forward, you want to plow through the backlog. But because of that, the single highest-priority thing gets postponed — I think that phenomenon is happening a lot.

I'm actually a bit skeptical that people can reliably "determine what must be done today" at all. Especially people handed projects with no pinned schedule — the "whenever works" kind.

For things that absolutely must happen, you should draw up a provisional schedule. Draw it, and the team starts moving toward it. If "the API-side tasks have to land this week or next week's integration can't proceed," then what must be finished by this weekend becomes visible. For tasks that need load testing or careful quality checks, the API work itself splits into steps — apply just the migration to production early, backfill the data, and so on. Refine it finely enough and "oh no, THIS has to happen today" reveals itself.

Getting to advocate for installing this "identify today's must-do tasks" mechanism across the company was a win this week. "I'll do it tomorrow" is genuinely fine — as long as it's truly a tomorrow task.

"I'm paid to carry the responsibility" — turning release judgment into a 2-minute video

My own job has mostly become this: other people build the thing, quality gets reasonably assured, testers go through it, and once it's quality-OK, I take final responsibility and ship it. That's honestly the truth of it. I really feel that I'm being paid for carrying the responsibility.

So right before deploying a large-diff pull request to production, I always re-check the entire diff. And what I pay attention to is roughly the same every time — I deliberately ignore the ultra-fine details, for instance. I distilled that whole "temperature" into Claude skills and turned it into a workflow.

On top of that, here's what I'm trialing now: having AI render the ship/no-ship judgment as a video, using Remotion. An MP4, about two minutes. I watch that first. Essentially, the AI scrutinizes the code diff against the checklist items a human requested, then presents the results to me. I still verify the real thing myself afterward.

If it doesn't drift too far off, this goes into full production use. Ten releases — that's the decision point.

Rendering AI output as HTML is trending in-house

Another hot topic internally: every time AI gets smarter, its output gets longer. Seriously, it's like reading a book a day.

Reading text depends on your condition, and I think that's true for a lot of people. When your head's foggy, nothing goes in. When you're sharp off a Red Bull, it flows right in — extra questions even start popping up. Human performance just isn't constant.

So for the cases where a human must be in the loop, aiming for the most human-readable format possible — having the AI render its output as decent-looking HTML — is quite the trend at our company right now.

Money

US stocks: 70% of the annual target

US stocks: I watched for an entry point, bought, and it climbed nicely within days, so I took profit. That brings me to about 70% of this year's stock P&L target.

I'm holding no spot position right now. Scouting the next entry.

Not even watching the bot anymore. But I've committed to 3 years

The crypto auto-trading bot: July was rough, and August is continuing slightly negative. And honestly? I've stopped watching it.

Not watching — but I keep running improvement candidates in parallel. And somehow I've kept that up every single day since around May.

When this thing finally takes off, I know exactly what I'll think looking back: "why did I ever expect results in two months?" I'm not taking massive losses, so I've decided: I'm keeping this going for at least three more years. Even three years from now, I'll still be in my early thirties. Totally fine.

Duolingo: day 220. A genuine enjoyment of English is sprouting

Duolingo: 220 consecutive days as of today.

And my beloved "Silicon Valley." I finished it recently and started rewatching — and I can hear more and more of it. This is round two of this rewatch, probably round four in my lifetime. When I catch the English but the translation doesn't quite click, I look up the phrase on Google to learn its real nuance. Translators are amazing — I feel that daily.

What makes me happy is that after keeping at it, a genuine way of enjoying English is starting to sprout.

Everything Else — Go hunting for "what must be done today"

The most striking realization this week: rigorously working out "what must be done today," writing it down, and then steadily executing it — that's what matters most.

You know the move: something genuinely needs to happen today, but it's hard, so you knock out light tasks first and manufacture that "I did work today" feeling. Everyone does it. But those little tasks? They often could've waited until the end of the month.

Same in private life. Even life goals can be dropped onto a schedule, provisionally, the moment you set them. Once they're on the calendar — even tentatively — what needs to happen today becomes much clearer. Work or personal, go hunting for "today's must-do task."

What if you finish today's musts in the morning? In the afternoon, interruptions WILL come — Slack, messages from clients. Keeping yourself ready to respond immediately is worth far more. Compare that to half-starting several tasks, getting interrupted, then opening cmux after a while and going "oh — I left this for two days. No idea what this even is" and closing it. That's a waste of context, isn't it?

People who have no "must do today"? They're probably in a passive position. And those people genuinely stop getting entrusted with tasks — because if a task can happen whenever, odds are it doesn't need to be them. So you have to move proactively: planning, scheduling, executing in ways where you create the value yourself.

Fortunately, something I absolutely must do today keeps materializing every single day for me. Good. But even those can often be automated with workflows — so strip those away, and let the person with my particular skills do what only they should do today: contribute to the company, to revenue, to the organization. Building the systems and structures that let people focus on where the value is — that's enormously important. People were saying this long before AI existed, of course.

Lots of realizations this week, and the same debates were raging inside the company too, which made it fun. A good week.

Closing

Next week, to derive "what must be done today," I'm going after everything that lacks even a provisional schedule — together with the PMs, somewhat unilaterally.

And if that's too heavy — provisional scheduling is something AI can do anyway. If the PM just says "around here-ish," it'll go "got it, drafted a rough schedule." Hand it to a workflow, keep it human-editable, and that's perfectly fine.

First step: each individual consciously practicing the hunt for "what must be done today."

Also, next week is Obon holidays in Japan, so hardly anyone's working. I've only got one working day — but I'll make it count.