Forem: Yoskee

Why I Started MODAY — The Story of a T-Shirt That Says \"I'm So Sorry\

Yoskee — Mon, 11 May 2026 12:25:58 +0000

Originally published at moday.me. Building MODAY in public.

Why I Started MODAY — The Story of a T-Shirt That Says "I'm So Sorry"

"I'm So Very Sorry" Printed on a T-Shirt, Right There on the Zoom Call

It was the day a client got stranded on a trip because of a typhoon.

It was meeting time. I opened Zoom, and there on the other side of the screen was the client wearing a T-shirt with "I'm So Very Sorry" printed on it in big Japanese characters.

"There's a typhoon, so I'm still away from home," they said with a laugh. "Is it okay if I'm dressed like this today?"

In other words, they were apologizing for joining from a trip location — and the apology was on the T-shirt itself. Before they could say it out loud, the shirt was doing the apologizing.

I laughed. Really laughed. This is good, I thought. The T-shirt instantly connected to their situation and softened the mood. What you wear can tell the story of who you are that day.

Remote Work Makes You Lose Track of Days

A little before that, I'd been feeling something. When you do remote work long enough, the sense of which day it is just fades away.

Tuesday and Wednesday blur together. You think it's Monday when it's actually Thursday night. Friday's relief gets diluted. Monday's heaviness becomes fuzzy.

You look at the calendar. But you don't feel the day of the week in your bones.

Just like that "I'm So Sorry" T-shirt explained that person's day in one line, I thought — what if you could wear the day of the week on your chest? That seemed fun.

Why It Had to Be for Engineers and Geeks

Making a day-of-the-week T-shirt the normal way wouldn't be interesting. A T-shirt that just says "Monday" already exists everywhere.

If it was going to land, I figured it would be with the people who do stupid things with style. The kind of people who can laugh after doing git commit -m "fix typo" three times in a row. The people posting questions on Stack Overflow at 2 AM. Engineers. Geeks.

Day	Print	Color
MONDAY	System Booting...	Black
TUESDAY	Compiling...	Black
WEDNESDAY	Deployed in progress	Black
THURSDAY	Running on caffeine	Black
FRIDAY	Build Successful ✓	Black
SATURDAY	Off the grid	Blue
SUNDAY	Rebooting...	Red

I kept the five weekdays all black, and switched colors just for the weekend. Saturday is blue, Sunday is red. I wanted there to be a small visual shift on those two days when you step away from work.

I Didn't Plan to Start a Brand

I normally do e-commerce consulting. Because of the work, I've been in the trenches of store building and operations for years. Still, I had zero intention of launching my own brand.

Growing other people's brands is fun. But taking the founder seat from zero requires a different kind of resolve. Inventory. Photography. Shipping. Customer service. Taxes. International shipping headaches. The domain of stuff you're holding alone explodes.

And to be honest, this is the first time I'm seriously using Shopify. As a consultant, I've worked more with other e-commerce platforms and just watched Shopify from the outside.

So I flipped all that over and decided to do it myself. That was March and April of this year (2026).

What Claude Code, Shopify, and Gelato Changed

If it had been January or February, I probably wouldn't have started.

Once March hit, I dove into Claude Code. Not just code — operating flows, Webhooks, image generation, translation, file management. I could run everything with it as my copilot. Even while getting my hands dirty with Shopify for the first time, Claude Code filled in the gaps I didn't understand. That was huge.

Shopify making Translate & Adapt free dropped the barrier for a 9-language rollout instantly.

I'd known about Gelato, but when I tested it again, I confirmed that Geo-Routing actually works in practice. Orders get printed and shipped from the facility closest to the customer's country. While sitting in Japan, you can sell to the world from day one.

"One person. Global. Zero inventory. Three months to launch." That moment became real. By mid-March, I was certain we could do it.

It's Not About the Thing — It's About the Experience and Story

A day-of-the-week T-shirt as a product, anyone can make. Send it to China, do a mass run at $3 cost per shirt.

You can't win on that. What MODAY sells is the embodied feeling of one week, baked into the shirt, and the goofball conspiracy between the people wearing it and the people making it.

That's why I'm planning to push the story behind the product way more than the product page itself. This article is the first one.

Reality: I Thought Two Days, But Four or Five Days Aren't Enough

To be straight with you, I was going to blitz the whole thing in two days during Golden Week. Then it'd just be tweaks.

In reality, I'm four or five days of work in right now and still not done. Store configuration, design, translation checks across 9 languages, Gelato product syncing, payment testing, terms of service, special commerce disclosures, shipping policy. It just keeps coming.

The actual development speed feels 3–5x faster thanks to Claude Code. But even that doesn't fit the two-day plan into five days. The distance from "building a store" to "having a store ready to sell" is further than I imagined.

Right now I'm targeting May 18, 2026 for launch. Nine days left. From here on, I just keep checking things off the list in order.

And 5/18 isn't the finish line. The real work starts after we open. Until the first order comes in, until the 9-language localization actually proves itself, until Gelato's Geo-Routing works exactly as planned — it's all still ahead. Building MODAY is only just beginning.

What I'll Write Going Forward

On this blog, I'm going to document everything about getting MODAY off the ground.

Why I picked the tech stack (that's the next post)
What I threw at Claude Code vs. what I did myself
Whether Gelato's Geo-Routing actually works in production
Where we got stuck on 9-language localization
The day we got our first order

I'll put it all out there — wins and fails. Not a polished case study, but a record from the trenches.

More to come.

— Yoskee

moday.me

Four Things I Still Can't Hand to AI

Yoskee — Mon, 11 May 2026 10:49:53 +0000

Originally published at moday.me. Building MODAY in public.

Four Things I Still Can't Hand to AI

I'm building MODAY — a one-person, day-of-the-week T-shirt brand — and I'm trying to hand every single task to AI. Three days in, four things refused to move.

The actual reason I'm doing this

You've probably read the obvious half of why this brand exists: a Japanese "I'm sorry" T-shirt going viral, days of the week blurring under remote work. That's the surface.

The other half is harder to admit out loud — I want to figure out how to build companies in the AI era before everyone else does.

My day job is e-commerce consulting. I see plenty of operations from the inside. So when generative AI broke open a couple of years back, one question started gnawing at me and never stopped: how much of a business can you actually let AI run?

Reading didn't answer it. Twitter didn't answer it. Papers didn't answer it. The question doesn't surrender to reading — it surrenders to a real business with real money moving through it.

So MODAY is a brand, yes. But it's also a one-person experiment at the right scale to actually measure things. Zero inventory. Solo operator. Global from day one. Three days to launch. Real money, real customers, real packages. Hobby projects and internal pilots let your judgment go soft. You can't measure anything without skin in the game.

The rule: no lines. Hand off everything. See what won't move.

Three days in, zero regrets

I haven't had the "I shouldn't have let AI do that" moment yet. Not for stack selection. Not for the design system. Not for the store, the webhooks, the translations, the copy. I've shipped Claude's output almost as-is across the entire build.

That sounds like a flex. It isn't. The honest read is darker — I've also handed Claude my judgment criteria. There's nothing left in me to regret with. If I'd kept my own opinions, eventually a "wait, I would've gone the other way" moment would arrive. It hasn't.

For better or worse.

And yet — four things stayed mine

For all the no-lines talk, four parts of this build refused to be delegated. By subtraction, here they are.

1. The decision to send a prompt at all.

The first sentence I type to Claude — "what's next?" — still comes from me. Every single time. I can't make it not.

When I can hand this off, I'll be at the next stage. I'm not there.

2. Bank accounts and payment processor approval.

Shopify Payments KYC. Stripe identity verification. Opening accounts for international payouts. These don't accept anything but a human. You show up with an ID, sign with your name and your face, and become the legal counterparty.

3. Service signups and putting a card on file.

Gelato. Render. fal.ai. Make.com. Anthropic. For each one I create the account, enter the card, and click "upgrade to paid." Every signup is my hand on the dashboard.

4. Generating the API key — up to the moment of handoff.

For Claude Code to call an API, the key has to exist. The "create new key" button is mine. The moment the key lands in .env, it stops being mine.

That's the entire list. Across three days, those are the only points where my hand physically moved. Everything else, Claude is running.

I want to hand off #1 too

Of the four, the one I most want gone is the first — the decision to prompt at all.

The version I'm aiming at: Claude Code shows up with "here's what we're doing next," I answer yes or no, and the direction of the business itself migrates fully to the AI side.

This is probably already possible technically. An agent setup can self-decompose tasks, implement them, and propose the next one in a loop. Plenty of builders are running variants of this today.

I haven't gone there yet at MODAY's stage. Part of me still wants to choose the direction of the first step. Honestly, that's the whole thing. Letting that go is my next move.

The other three aren't AI's limit. They're the system's.

This is the line I most wanted to write.

The bank, the KYC, the signups, the API keys — none of these stayed with me because "AI lacks the judgment." That's not why.

The real reason is institutional. AI has no legal personhood. Not as an individual. Not as a corporation. It cannot be a party to a contract. That's it.

Technically, Claude Code with browser automation could probably do all of these tomorrow. Fill the form. Upload the ID photo. Click the email link. Computer Use already exists. The blocker isn't capability.

The blocker is that even if Claude did the clicks, the registered party would still be me. Claude would just be filling forms under my name. The liability stays with the human.

So those three aren't where AI hits its limit. They're where society's institutions haven't caught up. The day AI itself can be the legal operator of a business, those three will move too.

Subtraction, not segregation

The "what humans do vs what AI does" framing will look obsolete in three years.

The moment you draw the line — "humans here, AI there" — the line becomes a constraint. And the question "how far can AI actually go?" becomes unanswerable from inside that constraint. You can't measure the limit if you've defined the limit upfront.

So MODAY isn't segregation. It's an attempt to hand off everything and check, by subtraction, what wouldn't move.

Four things stayed. Three are institutional. One is my own intent.

Time fixes the three. The one — I have to let go of that myself.

More soon.

— Yoskee

moday.me

I Asked Claude to Pick My Tech Stack — Now I'm Running on Services I'd Never Heard Of

Yoskee — Sun, 10 May 2026 10:33:31 +0000

Originally published at moday.me. Building MODAY in public.

The brand's thesis is "how far can AI take a launch?"

Up front: MODAY is half a T-shirt brand and half an experiment.

The real question I'm trying to answer is how much of a brand launch can I delegate to AI? I work as an e-commerce consultant by day, so I've watched the operational side of launches up close. But MODAY is a deliberate setup: one person, global market, three days, and AI doing as much of the judgment and execution as it'll let me hand over.

The tech stack got picked the same way. Not "I didn't know, so I asked AI." More like: I started with the assumption that AI would pick, and my only job was to define the criteria.

Here's what came out:

Layer	Service
Storefront	Shopify (Dawn theme)
Print-on-demand	Gelato
Translation	Translate & Adapt
Automation	Make.com
Webhook server	FastAPI on Render.com
Image generation	fal.ai (FLUX Pro)
Text / AI processing	Anthropic API

Of those seven, the only names I recognized going in were Shopify and Anthropic. The rest I met for the first time in conversations with Claude.

The whole stack is three days old.

I picked Shopify because of a news headline

The Shopify decision is the dumbest one to explain.

A while back I saw the Shopify MCP announcement go by. My immediate thought: "ok, so Claude Code can actually drive a Shopify store now." That was 90% of the decision.

I've watched Shopify from the consultant side for years, but I'd never built on it myself. Doesn't matter — if Claude Code can reach in via MCP, I figured I could ship without having to first become a Shopify person.

Japanese-domestic platforms (BASE, STORES) never made the shortlist. One reason: I wanted to ship globally on day one. Nine languages, multi-currency, real cross-border checkout. Shopify is the only platform that makes that look easy on day one.

So my actual selection criteria weren't "which platform is technically best." They were "which one works smoothest with Claude Code" and "which one lets me sell worldwide tomorrow."

I outsourced the POD decision to Claude on purpose

How do I print and ship the actual T-shirts?

Going in, I knew zero print-on-demand services. Printful, Printify, Gelato — never heard of any of them.

That wasn't a problem. The plan was always to outsource the selection. My personal knowledge of the global POD market is shallow. If I write down the business constraints and let Claude rank options against them, the answer is going to beat anything I'd reach on my own in a week of research.

The prompt I sent:

I want to sell custom T-shirts via Shopify with print-on-demand, fulfilled automatically. Target markets are the US, Europe, Asia — Japan is a secondary market. Which POD service should I use? Compare a few.

That was it. Nothing technical. No SDK languages, rate limits, webhook formats, or auth schemes. I couldn't have written those criteria if I'd tried.

The answer:

I compared the three major POD services. Bottom line: Gelato is the best fit for your use case.

🥇 Gelato (top pick)

Local print network in 32 countries, 140+ facilities

Local production in Japan, Korea, Singapore, Malaysia, Australia, India, China

Average 72-hour delivery to EU / US / Asia customers

Low customs risk (production stays in-country)

Ships to 200+ countries

Trade-off: smaller catalog than Printify, slightly higher unit cost

🥈 Printful (best for brand)

17 facilities across NA, EU, Asia

Tightest QC of the three; well-suited to brand-building

Trade-off: highest unit cost

🥉 Printify (best for cost)

90+ print providers, 950+ products

10–20% cheaper than Gelato in most categories

Trade-off: thinner Asian local-production network, no Japanese-language support

Recommendation: for global-primary, Japan-secondary, lead with Gelato. Local production across all three regions wins on shipping cost, customs, and delivery speed simultaneously.

Decided. The core of my fulfillment now runs on a service I'd never heard of and never touched.

If I'd done this comparison the way I would have as a consultant — reading three sets of docs, asking around — I'd have spent a week. Instead, near-instant.

With FastAPI, I'm not even at the "do you use it?" level

The stack table says "Webhook: FastAPI on Render.com."

I laugh a little when I write that. I don't actually know what FastAPI is.

Early in the build I needed something to catch Shopify order webhooks and forward them to Gelato. Claude Code said "I'll write it in FastAPI and deploy it to Render." A Python framework, apparently. I said go ahead.

It got written. It got deployed. It works. My total interaction with it is staring at logs in the Render dashboard.

(Side note: that piece is already retired. Something felt off mid-build and I swapped it out for a different approach. That's two posts down the road.)

When you decide to "go AI-first," you accept that you can't fully explain your own stack. That trade-off is the point, not a bug.

Smart, or reckless?

Honestly, I haven't sorted out which.

The upside is clear:

Options I'd never have considered show up on the shortlist on day one
"What I personally know" stops being a ceiling
Decision and implementation collapse into the same step — selection that should take a week happens immediately
A non-engineer running a global D2C brand becomes a real, not theoretical, project

The downside is just as clear:

When something breaks, I might not be able to read what's broken
I can't yet explain "why this stack" in my own words (that's literally why I'm writing this post)
If a vendor dies, my replacement decision is going to be slow

I won't know how this trade plays out for another six months. While it works, it's incredible. The day it stops working might be brutal.

The reason I'm running it this way anyway: I want a real answer to "how far can AI go?" measured against a real business. Hobby projects don't generate the answer. Real money, real customers, real shipping is what produces a number you can trust.

A three-day stack, going global

It's been three days since I picked this stack and started building. The actual hands-on-keys time is shorter than that.

Inside that window, the storefront went live, nine languages started flowing, the order webhook is firing, and Gelato product sync is moving. This is the v1 stack of an "AI-first brand launch" experiment.

Half of it will probably be replaced by year-end. FastAPI/Render are already gone. There are other suspect spots.

But as of today, this stack is on the brink of selling globally — and that's the part that matters.

Next post: which work I handed to Claude Code, and which work I had to do myself.

— Yoskee
moday.me