[4/25-5/1] 2026 Goal Progress

Health

Slightly Recovered, but Fatigue Won't Leave

The illness that dragged on all of last week finally started to ease up this week.

That said, I didn't really do anything about it. I never made it to the sauna either. And since I had baseball on Sunday, the physical fatigue just won't go away.

Recovering, or maybe not recovering — it's that kind of vague, in-between feeling.

Personal Gym Session #15: Pivoting to Heavier Weights

Hit my 15th personal gym session. Since I was tired and still had some lingering soreness from baseball, I shifted focus from reps to heavier weights.

• Bench press: 70kg for 3 unassisted reps, 2 with help. Then 60kg with 10kg assist for 1 set of 10
• Deadlift: 100kg, form-focused, 12 reps × 3 sets
• My legs were still tight so I skipped them, and instead did triceps for 15 reps × 3 sets

Up until now I'd been sticking to "Big 3 first" and avoiding arms, but right after working triceps I had this clear impression that my body got noticeably bigger. So I'm thinking I'll start gradually folding shoulders and arms in.

Quitting Smoking: Velo Is Just Too Strong

Honestly, the quit-smoking effort is iffy. No improvement at all.

Last week I wrote that Velo (the nicotine pouch) had "shed some light," but the nicotine is so strong it stings, hurts, and makes me feel sick. I bought the weakest version they had, and it's still pretty intense. It's going to take some time to get used to.

So I gave up on the idea of switching cold-turkey to Velo. Instead, I'll just gradually reduce the number of cigarettes through my own management. Slow and steady.

Sleep Is Stable, No More Oversleeping

Sleep-wise, the oversleeping is gone — I'm waking up at 7 or 8 AM.

Thanks to the constant fatigue, I fall asleep easily at night. A little ironic, but I'll take the win.

Work

The Per-PR Verification Environment Workflow Is Finally Done

This week's highlight: I finished building the workflow that spins up an environment for each pull request — the thing I started last week.

I built in hibernation to save costs, plus automatic resource cleanup once a PR is merged.

That said, it's only been a few days, so most of the team hasn't used it yet. Some evangelism is in order. Personally, it feels more than sufficient, so I want to push for heavy use and get the team to ride this development workflow I've defined.

Notion-Triggered Auto-Implementation: Already Works

And the Notion-triggered auto-implementation I mentioned last week? Already done.

The setup is simple:

• Write a Notion task using a template designed to be read by AI
• Pack in the resolution to a level where any human handed the task wouldn't be confused either
• Flip the AI-related data property
• Then just wait

This made local development almost unnecessary. PRs come in automatically; I review via comments, ask for fixes, and the AI commits the changes. Conflicts and Claude Code review results also get auto-committed where possible.

So, "What Do Humans Do?"

We're squarely in "what do humans even do?" territory. Here's my take.

Knowing *what* to build is still a human's job. So:

• Intent, concrete specs, "watch out for this," "don't do that" — the harness-style guardrails
• What I'd call context engineering, done with high precision
• No weird abbreviations or vague words — the kind of clarity business actually demands

That, I think, is what humans are being asked for now.

On top of that, humans should focus more on human-to-human interaction. Liven up meetings, generate ideas, and immediately roll those ideas into the AI workflow. That's the shape of work going forward.

And the more automation advances, the more load shifts to testers. Services with clients can't just "yolo deploy," so internal eyes-on QA becomes one of the high-value places. This was historically a domain that didn't fit the in-house-first mindset, so hiring and developing talent here is something we'll need to do properly.

The Headache: Too Many PRs Locking Up the Backlog

There's a struggle.

Pull requests are pouring in, and if you just merge them one by one, conflicts cascade, which means more Claude Code cycles and more context burned.

On top of that, people are firing PRs from their local Claude Code sessions. In that case, I want the person who opened the PR to take responsibility. Speed of "please resolve the conflict" or "this should merge but won't" is definitely up — but it feels like the homework keeps piling up and getting locked in place.

The mismatch between speed and processing capacity is surfacing as a new problem. A first-world problem, sure, but a real one.

Money

Second Profit-Taking of the Year

I sold my US stocks that were in the green. That makes the second profit-taking of the year.

Now I'm thinking about where to put the cash next. I have candidates, but it's a matter of timing and inventory at this point.

The Paid Note Is Going Well — Nerve-Wracking but Fun

The thing I'm planning to ship as a paid note is also going smoothly. Lots of nerve-wracking moments, but I've been clearing them somehow, so I'm having fun with it.

The specifics? I'll be writing those in the paid note, so I'll skip them here. Stay tuned.

Power-Generating Exercise Bike: No Quote Yet

The power-generating/storing exercise bike from the UK I talked about last week — the quote still hasn't arrived. Waiting on updates.

Duolingo: 120-Day Streak

For my emigration prep, I just hit 120 days on Duolingo.

That said, I feel like my daily learning volume has been going down, so I want to remember the original goal and put more time into it. Keeping the streak is great, but I shouldn't be satisfied with just that.

Wrapping Up

The most striking thing this week was that even with my body tired and run down, I managed to keep every one of my weekly habits going.

• The note (since January)
• My tech blog (and a regular blog too)
• Daily English study
• Weekly personal gym
• One more habit I'm saving for the paid article

I think I really pushed through. Keeping the lazy version of myself in check and getting it all done is something I want to give myself credit for.

On work: AI-driven development flows are now the foundation. Each piece feels well-built on its own, but once you connect them, there's interference and accuracy drops. There's a lot of waste I can see now.

But the end-to-end pipeline is built. From here, if I focus on tightening the connection points and improving the precision of each piece, things could get pretty wild.

I'll keep going.