[4/18-4/24] 2026 Goal Progress

Health

Personal Gym Session #14 — This Time, Pushing the Rep Count

Went to my 14th personal gym session.

Up until last time, I'd been chasing heavier weights. But this week, I switched to a plan focused on rep count instead.

• Bench press: 50kg × 12 reps × 3 sets, plus 40kg for 4 more sets
• Deadlift: My back was a bit tight, so I went light — 70kg × 10 reps × 4 sets, focusing on volume
• Squat: 50kg × 10 reps × 4 sets, staying conscious of form and load

Full body is sore. Might've pushed a little too hard on deadlift, because my lower back is kinda hurting. It's not bone-level pain though, so I want to fix my form to make sure the tension goes to the hamstrings.

A Week of Fighting Off Illness

My health was off the whole week.

I can't tell if it's hay fever or a cold, but I've got a runny nose, my head's foggy, and I just haven't felt up to anything.

Even so, I kept going with the habits I'd committed to. That's worth giving myself credit for.

Sleep Was All Over the Place, but No All-Nighters

Sleep was a bit inconsistent, but I didn't stay up late.

Today I was up at 6-something, and yesterday before 8. Slowly but surely, it's getting better.

Smoking — A Ray of Hope Called "VELO"

Smoking is still a total failure, but this week had a promising development.

A friend introduced me to "VELO" — a nicotine pouch you can buy at convenience stores. I tried the strongest one, and the nicotine hit was so strong it actually made me feel sick, to the point where I didn't want a cigarette at all.

This might be my breakthrough. Once my current carton is done, I'll switch to VELO. Starting a week from today, I'll give it a real shot. If it works, I'll probably list all my IQOS stuff on Mercari.

First good sign I've had in a while. Fingers crossed.

Work

This Week's Highlight: Built a System Where AI Implements Stuff On Its Own

To put it mildly, this week was one breakthrough after another.

One-shot PRs coming up from AI had become common, but when there were conflicts or I wanted to add feedback, having to pull things locally to handle it was pure overhead.

So I built it so that when you drop a comment on a GitHub PR, the AI commits the fix.

• Feedback from Claude Code reviews gets handled automatically (with breaking changes explicitly disallowed)
• Review comments and release-blocking items are submitted as auto-fix commits
• Additional requests are handled via PR comments too

As a result, writing code locally has basically disappeared from my workflow.

The Bottleneck Is Now Verification

PR quality is pretty good, so the human bottleneck has shifted to verification and review.

For verification, the fastest path is confirming it actually works before diving into the code, but we only have so many staging environments.

So now I'm building a system that auto-generates environments per PR (the thing that's been trending on X).

Native apps already have a setup where slapping a label on a GitHub PR pushes that build to TestFlight or AppTester. I want to extend this to API and frontend PRs too, so all environments get a preview spun up, and we can hand it straight to testers.

Next Up: "Auto-Implementation from Notion"

Once all this is done, there's basically no reason to code locally anymore.

The next step is putting a planning file in a task management tool like Notion, and having a GitHub Action pick it up and implement it automatically.

Keeping everything inside GitHub isn't great from a PM perspective, so starting from Notion is the key point.

Honestly in the "What Do Humans Even Do?" Phase

Once this is in place, it starts tying into broader questions about company survival strategy. We're genuinely entering a "so what do humans actually do?" phase, and it's a fascinating time.

Since basically everyone's on Claude Code now, I'm carefully setting up the guardrails — what not to do, security, skill rules, CLAUDE.md, that kind of thing — to prevent accidents.

I feel like I'm keeping up with AI itself pretty steadily.

The Real Problem: "It Doesn't Feel Like We're Moving"

There's one thing bugging me.

Throughput is definitely higher than it was before AI, or even when we were just using it locally — but the human bottleneck has become so prominent that it's starting to feel like "nothing's getting done," "we're not moving."

From the outside, it probably looks crazy fast. But from inside, there's this weird unfamiliar feeling of "hmm, I can't quite ship stuff as fast as I'd like". I'll probably get used to it eventually.

I've also been hesitant to run Four Keys-style quantitative evaluation on the project — feels a bit dated now, so I haven't really bothered. Either way, speed and volume are clearly up, headcount is down. First-world problems, I guess.

Money

Portfolio Unchanged, US Stocks Recovering

Portfolio hasn't moved.

But US stocks have been creeping up, and the unrealized losses are almost gone. I'm thinking about slowly selling off, so action is coming soon.

A Power-Generating, Battery-Storing Exercise Bike Might Finally Come to Japan

Not much happening on the savings side, but there's this exercise bike from a UK company that generates and stores electricity I've wanted for ages.

They kept telling me "we can't ship to Japan." Then I randomly contacted them recently, and they came back with "we can ship to Japan now, actually." They're working up a quote since there's tariffs to sort out. The thing costs around 500,000 yen.

With this, even during a power outage I can generate electricity by exercising. And most of our household's electricity is my work-related usage, so covering that with my own workouts would be awesome. As an investment in future savings, I want to pull the trigger.

Mercari Sales Have Completely Dried Up

On the income side, nothing's selling on Mercari anymore.

I think everyone's busy with the new-life stuff that kicks in after April. Hopefully things pick back up once the weather gets better.

Duolingo 112-Day Streak — 113 if I Do It Today

For overseas relocation prep, I've kept my 112-day Duolingo streak going. Haven't done today yet, so once I do, it'll be 113.

That said, compared to the first month or two, my lesson volume and time have dropped. The habit is locked in, but the quality per session is slipping.

I want to maintain the baseline while using small steps to bump up quality and volume.

Wrap-Up

This week was rough. The illness has been hanging on and I'm still dragging it.

Even so, I kept going with everything I've been doing all year. I'll take that win.

There were plenty of moments where I almost caved. Moments of "is this really worth it?" But rather than white-knuckling through it, I experienced how a tiny decision can keep the usual cycle running on its own. It really drove home how important habit formation is.

My analysis of why I got sick: probably moved around too much on the weekend, plus I drank a small amount of alcohol three days in a row even though I didn't really want to. Running on fumes drained my stamina, and I bet my immunity dropped. The pollen and yellow dust probably didn't help either.

Next week, I want to start by hitting the sauna and resetting my body.

At work, I'll ride the VELO breakthrough for smoking while pushing the precision of AI-driven automation further. Unglamorous, but one step at a time.