That was my reaction to playing the first Sixward prototype. Even with limited features, I was shocked by how complete the experience felt. It actually worked.
I had expected considerably less. Thit had asked more than 100 clarifying questions. The conversation had become very long, with plenty of context drift, and I figured the whole thing might have disappeared down a rabbit hole. Instead, playing it brought me right back to the feeling of first playing Ishido (but with different shapes) on Paul’s PS/1 and seeing those “amazing” VGA graphics.
A game I had imagined was suddenly a game I could play.
And it is fun.
I’m starting to figure out the strategy: how to maximize points instead of just placing tiles wherever they fit. More encouragingly, the friends I shared it with kept playing. They weren’t just testing it once to be polite. They kept playing.
Sherry leveled up to Square and could already see how changing the geometry would demand different patterns and a different strategy. That was a satisfying moment. Thwe did it: a game worth playing. One I could not have made myself. Not even close.
My vision, thit’s execution, for the win.
Of course, the prototype came with some prototype behavior. Wildcards weren’t following the rules correctly. That was an easy enough fix, although frustrating while I was playing. I also accidentally swiped my browser a few times and closed a game in progress. Coming back meant starting with a new board. That needs fixing.
Getting the pictures right has been a different kind of challenge.
I recently listened to Freakonomics Radio episode 619, which discussed things AI couldn’t do at the time, including understanding what was in a picture. Thit has become pretty good at that. Keeping the image creation tool focused is another matter. It has less of an attention span than a child: drifting between ideas, seemingly starting over, and leaving out something already agreed upon.
Pictures helped communicate what I wanted. Apparently, they really do say a thousand words, but those need to be exactly the right thousand words. I’m thinking about how to build may way around this. Something to put into the briefcase, or maybe its own project one week. Not sure.
Back to sharpening the axe.
Meanwhile, I got a bit inspired, or maybe it was too much Kraken and Coke.
I should have stuck with Jack, but I can’t argue with the fully working v1.0 that I could play on Monday.
Once thit had several days of project context, it became very easy to ask for huge, complicated features with a few basic prompts.
I can always revert to a working version. Why not reach for the stars?
I ran a few more simulations using my test agent, found a few weird experiences, but it works. I could ship it today, but now I want to make it more than my first game, I want to make something that people will love to play.
There is a lot to polish about the user experience. A whole lot.
I think this is the only update this week, have a funeral to go to on Friday.
I had a lot of fun making and play-testing the games for Load Goblin. There’s something really satisfying about making something that’s actually fun to play.
I’m mostly a retro gamer. I enjoy games that don’t require a massive story arc, hundreds of hours to finish, and ten times that to complete everything accomplishments. Gotta catch ‘em all! Give me ten minutes of digital pinball and a chance to beat my all-time high score, anytime!
My brother Paul came home from the Army with an OG IBM PS/1 computer. That thing had VGA graphics! My Tandy 1000 had nothing on it. Two games I particularly enjoyed were Lemmings and Ishido.
Later, I got Ishido for the Atari Lynx, playing it on Uncle Karl’s early high-resolution color handheld.
I loved the strategy, the lore, and the almost Zen-like experience of this solitaire game. It reminded me a little of Shanghai (mahjong solitaire).
I liked it enough that I tried making a physical version using modeling clay. I designed the pieces, but never finished it.
Years later, I got a wood-burning kit for Christmas and thought I could make one using wooden Scrabble tiles. The sanding wasn’t going well, and I valued my fingertips more than finishing the project.
Twice half-baked. I do like a twice-baked loaded potato skin… 🤤
Where did the idea come from?
I wasn’t originally trying to reinvent Ishido. I just wanted to make a physical version.
Sometimes there’s something satisfying about tactile things. Like kindergarten, when everything was something you could pick up, touch, and play with.
Sherry gave me a wooden Sudoku set that I really enjoy. It’s much slower than my normal speed runs on Sudoku apps, but there’s something therapeutic about that.
My one real departure from Ishido was wanting symbols that meant something to me, making the experience a little more personal.
Now, for the first time, I have the ability to build the game in a format I never thought I’d be able to create myself: a video game.
Third time’s the charm? Or maybe trice-baked?!
What am I trying to make?
A game that’s fun to play. That’s it.
My other projects have been fairly niche or require some commitment. This one should be something anyone can open in a browser and start playing immediately.
A few minutes of entertainment, some brain chemicals flowing, maybe a new high score.
Thwe’re calling it Sixward. Six colors, six symbols, with a goal of getting a high enough score, or enough achievements to get to the next level forward.
Thwe’re starting with the ideas behind Ishido, but exploring some new mechanics involving colors, symbols, scoring, and harmonic combinations.
I haven’t even settled on whether to begin with traditional square stones or something different.
I love the idea of progression. Get a high enough score or achieve harmonic perfection, and unlock another way to play.
But first, we need to make the basic game fun.
What do I hope to demonstrate on Saturday?
Something people actually want to play again.
Success isn’t just getting people to click the link. It’s repeat players, games played, and maybe people inviting their friends.
You should be able to start playing without registering. Once you’ve unlocked an achievement worth keeping, you can create an account to save your progress.
If somebody enjoys it enough to share it with a friend, that’s a pretty good indication that we’ve made something worthwhile.
What could go wrong?
I’ve never designed game mechanics before.
We could easily disappear down some rabbit hole and come out with something that technically works but isn’t the experience I wanted.
My stretch goals are pretty audacious. I’d love to turn this into an iOS app (someday)/ And we’ve even discussed a spherical playing field, which is a little nuts to put my brain around. Something I can imagine, but could never actually code myself.
But I can’t skip ahead. We have to prove the basic mechanics work before expanding the experience.
Visual design is another potential trap. Working with AI on visual experiences has been challenging to say the least, and I don’t want to spend the whole week making things look beautiful at the expense of making a fun game.
Especially with a shorter Sunday than usual, we’ll need to stay focused. Go Blue, beat Houston! Go Lions, beat the Pack!
How should the story end?
With a game I love playing myself.
I could care less if anyone else plays it. But I bet if I enjoy it, plenty of other people will too.
And unlike Load Goblin, nobody has to take my word for it or own a physical Timex Sinclair 1000 to experience what thwe’ve built.
Last Sunday, I wrote that I wanted to build something people could use to connect through food.
Well, it’s Saturday.
Introducing More Than Ingredients
My expectations for this week were modest. I wanted a basic site where people could enter or import recipes, save them, share them, and use them in the kitchen.
I hoped thwe might get versioning working.
Instead, thwe built nearly every feature from my original vision. And then some.
More Than Ingredients is a digital recipe binder with friends.
You can add a recipe manually, import one from a website, save recipes from other members, follow chefs, and see what they have shared.
Recipes can be public or private. Private recipes can be shared through revocable links. Public recipes can be viewed through a direct link without forcing someone to create an account just to find out how much garlic goes into dinner.
You can mark recipes to try, add private notes, record that you’re making something, and answer one simple question afterward:
Would you make it again?
That still feels more useful to me than averaging star ratings.
Recipes have histories
Versioning was important to me from the beginning.
I copied Mr. Prisk’s pumpkin cookie recipe by hand when I was a teenager. I added chocolate chips and the missing baking instruction: Bake 20 minutes @ 350°. My mom changed the quantities, reduced the cinnamon, and added nutmeg to serve more than one Rob.
Which one is the recipe?
All of them.
More Than Ingredients lets someone create their own version while preserving its connection to the recipe that inspired it. Versions can have versions. Notes and contributions can travel through the family while still respecting who created, discovered, adapted, or contributed what.
The model underneath became surprisingly sophisticated.
The experience shouldn’t feel that way.
That became one of the themes of the week: elegance.
Not merely visual polish. Design, user flow, simplicity, and features that surprise and delight.
One of my favorite examples is the recipe editor. Paste a block of ingredients or instructions, and the system turns it into a structured, editable list. You can rearrange items, add headings, or make corrections without typing everything into separate fields.
Manual entry is still there if you want it.
The complicated part happens behind the curtain.
When something goes wrong
Importing recipes from the internet is messy. Some websites provide beautiful structured data. Others provide incomplete information. Social-media posts can be even less cooperative.
When an import fails, More Than Ingredients creates a temporary card in the member’s binder and sends the recipe to an administrator or moderator queue.
The member sees that it is coming soon. Someone behind the scenes repairs it, and the member receives a notification when it’s ready.
I think that is one of the most important things thwe built.
A failed import shouldn’t become the user’s technical problem. It should become our promise to finish the job.
Is anybody getting in?
The lowest point of the week was silence.
Joining a brand-new invitation-only recipe community is a big ask. When people didn’t immediately sign up, and Sherry wasn’t receiving invitation or password-reset emails, I couldn’t tell whether people were uninterested or the front door was broken.
Was the core mechanism of the entire site completely FUBAR?
Then Cole successfully created his account. So did my former boss at AdAdapted.
The front door worked. 😅
Cole also caught details thwe had missed. Separate username and display-name concepts made registration more complicated than necessary. He also suggested that choosing a different generic user icon should change its color.
Small changes, but they made the site feel more elegant.
There’s that word again.
Process leads to production
I’m shocked by how complete this product became in one week.
The process improvements carried forward from the first two projects helped me raise the stakes and accomplish more in less time.
Clearer briefs led to better builds. Testing created better requirements. Actually using the site exposed things planning could not.
Process leads to production.
The weak spot was UX feedback. The punch list became enormous, and describing visual problems through text is difficult. Screenshots and wireframes helped considerably.
Next time, significant UX changes need to become visual earlier: show the screen, mark the problem, and review the interaction before writing the code. Otherwise, one improvement can create three hotfixes and another opportunity to say “final final final.”
Image manipulation remains its own category of funny and frustrating. I’ll spare you the seventeen-version Made in Michigan logo saga. You can read that story instead.
Some scars deserve their own post.
It’s alive, so the list is growing
Using something naturally produces feedback.
I already have 10 post-launch improvements recorded.
That does not mean the product missed the finish line.
It means the more you use something, the more refinements you want to make.
More Than Ingredients also includes a suggestion box so members can tell me what isn’t working, what feels awkward, and what they wish it could do.
I expect the list to grow.
Always more things.
Now I need some people
More Than Ingredients is live, but invitation-only.
I want to grow it deliberately while I learn what happens when people other than me begin adding recipes, following each other, creating versions, and cooking what they find.
Once you’re inside, try one of my recipes. The Chicago Hot BBQ is REDONKULOUS!
Then share one of your own that you’re proud of.
Maybe it is a family recipe passed around for years. Maybe you created it yourself. Maybe it came from a friend, an index card, a binder, or a social-media post you were afraid you would never find again.
Add the recipe.
Credit the person behind it.
Tell us why it matters.
The goal was never simply to build another place to store ingredients.
It was to create a little more honest human connection over food.
That more or less explains how I ended up spending seventeen versions trying to rebuild my Made in Michigan logo while More Than Ingredients was being built in another window.
I have always liked the logo. I think I modeled it after something I had seen, but I made it from scratch. Unfortunately, the original was awful at small sizes. I wanted a proper vector version that would remain crisp and could easily change colors to match each weekly project.
How hard could that be?
(Yes, I vibe coded a way to compare the logos with a slider…)
For a while, the answer appeared to be “impossible.”
The M kept disappearing. The UP would vanish. The layers would not cooperate. The Upper Peninsula needed a clear margin, a thick white stroke, so it could sit over the letters without turning into an ink blot test.
Somewhere around version fourteen or fifteen, it finally looked right.
I was so proud that I pulled it up to show Sherry. I zoomed way in to demonstrate how crisp the new vector remained compared with the original.
That was when I noticed St. Ignace had moved approximately fifty miles west.
Thit had produced a beautiful, scalable, geographically inaccurate Michigan.
I used my finest and most descriptive words to explain exactly how the tip needed to align with one tiny nub. I was overly confident that this would settle the matter once and for all.
The next version shifted the entire Upper Peninsula! Menominee was now apparently connected to the Lower Peninsula by the Mighty Mac. This would make getting to Wisconsin a lot easier!
In one revision, Thit had changed the course of Ice Age glacier gouging history.
I laughed out loud for about two minutes.
The next version was perfect.
Seventeen. Seventeen versions, but thwe got it.
The MVP Was Okay
Meanwhile, the actual project had reached its original goal.
You could create an account, add recipes, import them from websites, save them, and use a cooking mode. It was what I had initially aspired to build.
But honestly, It was also pretty meh.
A recipe-saving website is useful, but there are already plenty of ways to save recipes. The more interesting idea had emerged while we were building: recipes are not fixed documents. People change them.
They add more giardiniera. They use less oil. They substitute what they have. They inherit a card from someone, write changes in the margins, and eventually make something related to, but different from, the original.
So we pushed for the stretch goal: recipe lineage and versioning, said no chef ever.
An engineer might, though.
A member could now copy a recipe, change the parts they made differently, publish their version, and preserve its connection to the original. Versions could have versions. Helpful contributions could travel through the family while remaining attributed to the person who shared them.
The build was ready very late Tuesday night. I decided not to deploy complicated database changes while tired and called it a night.
The following day, I deployed it with cautious optimism.
It worked. Ish.
You could create a version, change it, publish it, and let other people see how your version differed. The experience was still awkward and had a severe case of unnecessary padding, especially around the dietary restrictions, but the central idea worked.
My confidence was restored.
😅 Almost.
Take. My. Money.
Just about then, right as thwe found our groove, I hit my daily ChatGPT Work limit.
Full stop. No AI for you!
I was informed that I could start using Work again at 3:13 a.m.
WTF.
Seriously.
Take. My. Money.
My ChatGPT spending progression has now gone from $20 to $100 per month. I didn’t think I would need to upgrade. But here we are. At four projects a month, that is about $25 per website.
Heck of a deal.
At some point, the cost crossed below the value so decisively that upgrading stopped feeling expensive, but that deserves its own post.
Our First User Was an AI
To improve the dreaded ugly UX, I asked Thit to perform a user-experience review of the live site with some of these new (to me) fancy Pro features!
But how could thit see it, it was behind a invite only wall.
I created a tester@ account and let it loose inside my invitation-only recipe community. What could happen?!
Thit found a chocolate-chip-cookie recipe on King Arthur Baking (I need to look into this), imported it, saved recipes, tested links, entered cooking mode, and generally kicked the tires.
Then the AI, pretending to be a tenured UX designer at a major recipe publisher, produced an elegant and undoubtedly thorough sixteen-page report that I will never read.
I handed the report back to my regularly scheduled programming Thit and asked thit to extract the few changes that would actually matter before Thursday night.
This may be the new workflow:
One AI builds the product.
Another AI uses the product.
That AI writes a report.
The first AI reads the report because I will not.
I decide what is worth changing.
The most important result was not a button color or another round of padding removal. It was the user journey.
A new member does not follow anyone. Opening the site with “Newest from people you follow” gives them a beautifully designed empty room.
A first-time member should see the recipe library. They need immediate proof that there is something here worth exploring. From there, they can discover people and begin following them.
Once they have participated and actually have a feed, the site can show that feed first when they return.
First visit: explore recipes.
Early engagement: discover and follow people.
Returning visit: see what those people have added.
Deeper use: save, cook, contribute, and create versions.
While answering the interview questions for this post, I realized the requirement still was not quite right. It should not merely remember whether someone had visited before. It should show the feed first only when that person actually has one.
I opened another session and changed the requirement while writing the paragraph before this one.
The retrospective improved the product while documenting it. Huh.
Then a Human Arrived
Today, @lancer became the first outside human member.
He accepted his invitation. He logged in.
That is all I know.
I hope he explored. I hope he thought, “Holy crap!” I hope he didn’t hit a bug.
Maybe he opened it, became distracted, and went back to something important.
No recipes. No parade. No manufactured testimonial.
The first contribution from someone other than Sherry or me still came from an AI agent importing chocolate-chip cookies into an invitation-only recipe community.
The Axe Was Apparently Sharp
At the beginning of the week, I spent too much time changing how thwe worked. I sharpened the axe instead of hitting my first milestone, and I was frustrated about losing a day of real-world testing.
That work paid dividends.
Thwe can now develop requirements in one conversation, turn them into a structured build brief, hand the brief to a building session, and work on something else while it runs. We can have another session test the result, reduce sixteen pages of findings to a practical punch list, and send the important pieces back through the process.
Now, we are CRANKING!
Bring on the scope creep!
👆Famous last words.
The site is live. It contains real recipes with real stories: Mr. Prisk’s pumpkin cookies, J Ross’s Salade de Saumon et Pois Chiche, Chicago Hot BBQ, and the best recipe for Detroit-style coneys, which also taught us that uploaded photos need rotation and cropping because my hot dogs are currently sideways. Why does rotating an image 90 degrees mess with your brain?
People have asked for invitations. I have sent them. What I want most now is for those people to give it an honest try.
I am very proud of this one.
I think it is done.
Well, it will be as soon as I deploy the release package that is building right now.
Literally.
Now.
Version 1.7.3
Shipped.
P.s. this was the logo thit recommended for the site.. Just going to park this here.
Thwe made the first MVP build of More Than Ingredients by the end of the night on Sunday. My goal was to get it deployed, test the basics, and start adding the social features before inviting real people to try it.
I didn’t get it deployed Sunday. I couldn’t because I had blocked myself with changes to the deploy process. WTF.
Let me back up.
Thit interviewed me, and had developed a build package, that just sat there waiting for me to deploy it.
Thwe spent a lot the day sharpening the axe. That is usually a good thing. But not today.
Thwe organized ChatGPT project folders, fought its recent conversation list and less-than-stellar search, connected Google Drive, and created documents to capture our evolving development process.
Thwe even created an in-chat “note to self” workflow that writes follow-up tasks directly into a Google Doc. I can do this on any chat within the project! No more switching contexts to capture an idea. Just continue the stream of consciousness and add my coded phrase inline, Thit picks it up and adds to the to do list!
Pretty cool, actually.
Except I was supposed to be building, deploying, and testing a social recipe website.
I also distracted myself building the One Project Every Week website. Apparently, I wasn’t satisfied building one website in a week. Now I’m building two.
I did find out you can make a quick HTML page that you can fully preview inside ChatGPT that you can use to do useful stuff like generating 64-bit secrets, all without deploying any code. Useful. I need to remember that for something later.
Cool, but a distraction. Clearly I miss having Sherry around (she is visiting her parents in FL this weekend).
Lots of useful work. Very little chopping wood.
By Sunday night, I was frustrated. I’d introduced real-user testing as a goal for this week. More Than Ingredients is supposed to be social, and every day spent preparing was a day people couldn’t use it.
A recipe binder is nice, but it’s hardly a social network.
Monday: Finally, some wood chips
Getting the site running meant dealing with hosting, email accounts, passwords, configuration files, and the hundred little boxes you have to tick before a website actually works.
Some email configuration thwe’d discussed hadn’t made it into the original build. Invitations failed. Password recovery needed testing.
At one point, an invitation produced a 403 error.
But there he was. The Goblin!
One of our reusable components from Load Goblin had made the journey to More Than Ingredients. Nice to see a familiar face, even if he was delivering bad news.
Eventually, the email flows worked, and I could move on to something more interesting: loading my first recipe.
Chicago Hot BBQ Pulled Pork. My recipe, of course.
I started with the import-from-URL workflow. That exposed some issues with semantic ingredient tagging, which required changes to that one page site.
And encoding.
ENCODING!
How have thwe not solved character encoding by 2026? I’ve been dealing with this nonsense for decades.
I even found a phantom “[Ingredients]” ingredient. Turns out that one wasn’t a bug at all. The system supported ingredient subsections, like “Dry Ingredients” and “Wet Ingredients.”
An unexpected feature. I’ll take it.
With the recipe loaded, I successfully registered another of my email addresses through an invitation and saved Chicago Hot BBQ into that account’s recipe box.
It worked!
I fixed the earlier broken invitations, ran through some QA, and found surprisingly few problems. A few bugs, a few changes, but the core experience worked end to end.
I’d also sent invitations to actual friends. As of Monday night, though, my entire registered user base was still me and my various email addresses.
Not exactly viral growth.
More than a recipe box
Monday brought another reminder of why I’m building this.
I texted my former CEO to try and retrieve Mr. Prisk’s pumpkin cookie full recipe from their company Slack.
The original handwritten card I’d saved as a kid only had ingredients. I’d previously used AI to help develop the instructions, made the cookies for a company exchange, and shared the resulting recipe on Slack.
Now I had to ask my former CEO to get it back.
There’s a problem worth solving.
And I talked with my friend Lance, who was making my Chicago Hot BBQ. Again. He’s made it four times. I’ve made it twice. He’s already experimenting, adding more Potbelly giardiniera and oil to spice things up.
My recipe is becoming his recipe, too.
This is what was missing from other recipe sites. They are one way doors, best you can do is leave a comment or thumbs up. I wanted to build an interactive connected recipe site.
Recipes people discover. Recipes people share. Notes about what they changed. Eventually, different versions that tell their own stories.
So, with the MVP working, I revisited the someday maybe features backlog thwe’d deferred and asked thit what thwe should tackle next.
Discovery. Social features. Notes.
Thwe decided to go for it.
The first design was clean and functional, but it needed more personality. Thwe added a visual-direction checkpoint to our future build interviews: colors, style, inspiration. Another lesson learned.
And yes, more axe sharpening.
By Monday night, thwe had a substantially more ambitious social build ready to install and test. Not deployed yet, not proven, but ready.
Sherry was coming home Tuesday. I was nervous about installing the new build, excited about getting it into people’s hands, and thoroughly exhausted.
