My goal on π: 150 real connections.
Not more.
Dunbar's number says a human can keep about 150 stable relationships.
Past that, it's an audience, not connections.
I want to connect with:
π€ People building with AI (automations, mobile apps, web appsβ¦)
=> we can talk tech, efficiency, optimization
π° Builders who want to monetize their apps
=> we can talk marketing and sales
πͺ People who work out
=> we can push each other harder
Which one are you? Say hi and tell me a bit about yourselfπ.
11
2
158
Payments are 5% of my code and 90% of my stress.
The rest of the app is easy to deal with.
I write it, I test it, the tests go green.
A bug in production? No biggie. Nobody dies.
Payments are different.
When money is involved, a bug means:
- an unhappy user
- a refund (plus fees)
- a bad reputation
I'm not there yet. Right now my stress is of a different kind.
The code was almost the easy part.
What's hard is everything around it:
- waiting for my live environment to be approved
- waiting for my app to be verified
- sandbox tests that prove less than I'd like
None of it is mine to debug.
I can't fix someone else's approval queue.
So the real skill isn't coding payments.
It's staying calm about what I don't control.
What a journey, pushing my own limits π
What's your 5% that causes 90% of your stress? π
1
31
"Out of suffering have emerged the strongest souls."
β Kahlil Gibran
PS: I write down quotes that inspire me and share them with you.
Hard times don't feel like growth while you're in them. They only make sense looking back.
What does it inspire in you? π
9
What's an app worth if it can't take money?
Mine is still a work in progress. Here's what's left.
β Last payment tests in sandbox: account updates, and what happens when someone stops paying
β Get my live environment approved by my payment provider
β Get my app verified by Google (I need access to their APIs)
β Officially open the app (it's already live, with 1 user: me)
637 automated tests are green.
None of them can tick these boxes for me.
I'll update this list every day this week.
1
16
AI is making me dumb.
I'm a trained engineer. 20+ years in the field.
I used to do it all: architecture, design, code, review, tests, debugging, deployment, monitoring.
Now I just spec the behavior.
My harness lets AI take it from there.
Last bug: I described it in one line.
AI sent back a full report: the inconsistency, the root cause, how it fixed it.
I barely read it.
I just trusted it.
Not because it's always right.
Because I got lazy.
And lazy is how dumb starts.
Am I a lost cause, or is this just the new job? π
1
13
"If a plant cannot live
according to its nature, it dies;
and so a man."
β Henry David Thoreau
PS: I write down quotes that inspire me and share them with you.
Everyone must find their true purpose, their "why" (as Simon Sinek says), to fully grow and thrive.
What does it inspire in you? π
1
15
My most productive week was my least useful one.
Starting feels like progress.
But finishing is progress.
Last week I audited old projects.
New ideas, quick wins, AI doing the heavy lifting.
It felt great.
Meanwhile, the one thing that matters didn't move:
my app still can't take a payment.
Old projects are easy:
- no users waiting
- no deadline
- nothing to lose
- the thrill of something new (forgotten projects feel new again π)
The real project is hard because it's real.
Fear of failing. Fear of rejection.
Then discouragement kicks in. Then self-loathing.
But beyond fear, there is bliss.
So this week, one rule:
no new code outside the payment system.
I'll report on Sunday.
2
1
29
AI is creating this kind of code (spaghetti code)
Unless you know engineering and guide it correctly
32
"While we wait for life, life passes."
β Seneca
PS: I write down quotes that inspire me and share them with you.
We can't predict the future, or know if what we secretly wish for will happen.
This one reminds me to take action and see what unfolds.
What does it inspire in you? π
17
My follower count jumped 467%. And I have no explanation.
But that's not the interesting part.
Last Sunday I made a promise: my build/sell split would flip.
It did. Barely.
Last week: 72% build, 28% sell.
This week: 49% build, 51% sell.
But my time tracker found a third block: old projects.
As much time as the product I'm shipping.
Why?
I got sick midweek. Old projects were the easy escape.
And I had free AI credits to burn. I used them to audit projects I want to revive anyway.
Selling went up. Focus didn't.
Next week: finish what I should have done this week. The payment system for my app.
π Weekly numbers
Followers: 3 β 17 (+467% π)
Posts: 7 β 7
Replies: 19 β 30
Views: 451 β 1,430
1
99
10 followers, $0 MRR, and @levelsio already blocked me.
What did I do?
1. He thinks I'm a bot
2. I replied something he didn't like
3. Something else (tell me below)
Honestly, I'm a bit puzzled.
1
23
A month ago I started tracking my time.
What I found changed what I build.
First surprise: hours I couldn't name.
In my first 4 workdays, more time went to nothing I can name than to the product I'm trying to ship.
I didn't feel distracted. That's the scary part.
Why? AI made distraction cheap.
Sept 30, reading finance docs, an idea hijacked me: a stock alert system.
With AI, a quick test is one prompt away.
Building got cheaper. Focus didn't.
So I built 2 tools to see it:
- a sticky window with my one priority (built with Claude in ~10 min)
- a time tracker, per project
Then I tagged my hours: building vs selling.
70/30.
20 years of craft on one side. Almost none on the other.
Building is what I'm good at, so it's where I hide.
The question I ask now before building anything:
Does this help me ship, or does it just feel like progress?
Yes β I build it.
No β backlog.
I aimed for 50/50 the next week. Got 70/30 again: I chose to finish my v1.
This week, the v1 is in testing. The split should flip.
Results tomorrow.
2
1
46
637 tests. All green.
The AI still got it wrong.
My 3-step method produces a lot of automated tests.
They check every use case of every feature.
With fake ports: no real payment provider, no real database.
This week I started manual integration tests on my payment flow.
It didn't work.
The AI had assumed the payment provider hosts the payment page.
It doesn't. The page has to live on my site.
I didn't know that provider's API, so I trusted Claude.
The assumption went into the spec, then into the code.
637 green tests, faithfully checking the business rules.
None of them checking the integration.
Once I showed Claude it didn't work, it fixed it.
And admitted its first proposal could never have worked.
No big loss: with AI, dev time is short.
But a reminder: you still need to know the tech.
For how long?
1
27
AI can write your payment code.
It can't take the blame for it.
A weekend tool? Let the AI guess. Worst case, you rewrite it.
Payments are different. There, a guess has a price tag:
- a double click that charges twice
- a network retry that creates a second order
- an error swallowed, so nobody knows it failed
So two of my rules leave nothing to chance:
- Every write is idempotent. Replay it, nothing new is created.
- An error is always returned and shown. Never swallowed.
Claude follows them because they're written down.
Not because I naively think it will guess them.
Code is cheap now.
Mistakes with money aren't.
1
22
AI destroyed the old moat: knowing how to code.
The new moat is your AI context.
It channels the agent and saves tons of tokens.
I have 9 context files that apply to all my projects.
Same quality bar, every time.
One of them is my architecture guide: 692 lines.
3 rules from it:
1. The swap test
Change the UI framework: only the UI layer moves.
Change the database: only the adapters move.
If a business rule moves, it was in the wrong place.
2. Time is a dependency
No `new Date()` in business logic. A clock is injected.
That's how you test "expires in 30 days" without waiting 30 days.
3. Check order
Access before existence.
Not allowed? You get "Forbidden", even if the item doesn't exist.
Otherwise a "Not found" tells an attacker what's there.
Anyone can prompt. Few have written the rules.
1
41
I don't review Claude's code anymore.
I review the tests it's written.
Code can look clean and still do the wrong thing.
A test says what the code must do. In plain words.
If the test is wrong, the code will be wrong. Even when it's green.
So that's where my attention goes:
- Does each test match the spec?
- Is a case missing?
- Does it check behavior, or just the code it was written for?
Then the agent codes until everything is green.
But green isn't the finish line.
I've been testing my payment system this week. I found env config issues the AI never flagged.
Tests check the code. Not the world it runs in.
15
Don't let Claude write your codeβ¦
unless you go through these 2 steps first.
1. Spec + tests
Not a prompt. A conversation.
Claude challenges my need, asks what I didn't say, suggests what I forgot.
2. Architecture
The agent writes the backbone from detailed rules, with examples: how the layers are separated, what goes where.
The code compiles. The functions are empty. Tests are all red.
3. Code
The agent fills in the functions until all tests are green.
Quality rules: names from the domain, small modules, code that says what it does.
One agent per step. I validate between each.
Skip to step 3 and the AI makes up what you didn't say.
On a small project, that's fine.
On a bigger one, it wastes time and tokens.
Next: automating the handoffs.
1
30
I aimed for 50/50 between building and selling this week.
I got 70/30. Same as last week.
Why?
I chose to focus on finishing my v1.
The payment system is almost done. I'm testing it now.
Once it's validated, I switch to selling. The split should flip.
On the X side:
- 7 posts (last week: 4)
- 26 replies (last week: 3)
Happy I kept the pace of one post a day.
Still not enough to grow: I'm still at 3 followers π
2
29
Building is hard. Selling is harder. Most people who try both, fail.
But the hardest part comes way before, and nobody talks about it: getting started π
The forces that stop you are hidden: fear, laziness, procrastination ("I'll start Monday").
They don't feel hard. They feel like good reasons.
I used to feel them. Now my build is almost done, the next step is showing it to people, and I need to do it fast.
Delay has never been this expensive.
AI made build + sales easier for everyone.
Picture a million more people realizing it. Even if a conservative 1% try, that's 10k new competitors.
Every week you hesitate, someone else ships.
Speed of execution is the edge now.
What keeps me moving: I want to succeed.
How I make sure: I post every day to stay in motion.
It's harder to quit what you've said in public.
1
1
38







