CHAPTER TWO
It was supposed to be a vacation.
The workplace was closed for the whole week. Every reasonable version of that sentence ends the same way – sleep late, rest, disconnect. Instead, at 6:30 in the morning, something pulled me out of bed. I made breakfast. Fed the cat. And by 7:25 I was sitting in front of the laptop, ready to build.
Not because anyone asked me to. Not because there was a deadline. Because the next thing was waiting, and waiting felt worse than starting.
That’s when I typed the first message of the session.
Respect – you want to build it right. Let’s think clearly before writing a single line.
React Native is a proper app. It connects to your existing Render backend – the same API endpoints you already have. The app is essentially a mobile interface that talks to cgen-backend-udg1.onrender.com exactly like your website does.
Before we start building I need to map the app structure with you.
Answer these questions and we’ll have a clear blueprint:
1. What screens does the app need?
My suggested list:
- Home / Landing screen
- Intelligence Engine (Mode A – Market Research)
- Intelligence Engine (Mode B – Idea Validator)
- Concept Generator
- Archives (browse published reports)
- Private Lab (member login + their drafts)
- Login / Register
Is that the right scope or do you want to add/remove anything?
2. What’s the priority order?
Which engine do you want working first? Intelligence Engine is probably the most powerful single feature – would you start there?
3. Expo or bare React Native?
Expo is a tool that makes React Native development much simpler – especially for a non-programmer building with AI. It handles all the complex setup automatically.
Session 29 — The First Screen
They said it couldn’t be done without a developer. They said React Native was too complex for someone without programming experience. They said you need a team.
David Wolf built the first screen of the CGEN mobile app alone. On his vacation. At midnight.
It wasn’t clean. It was three wrong folders, a Windows security policy, package version conflicts, a network that refused to cooperate, and an hour of white screens that showed nothing. Every step forward came with two steps sideways. Every solution created a new problem.
But the approach that finally worked was the same one that built CGEN itself – strip away everything complicated, go back to basics, find the one thing that actually works and hold onto it.
The connection method that nobody suggested at the start: type the address manually. exp://10.100.102.13:8081. Eight characters that unlocked everything.
And then – on a cheap Samsung A12, on a Tuesday night in Israel, on vacation – the screen appeared.
Dark background. Lemon green. Three engine cards.
CGEN · CONCEPT GENERAL Know Before You Build.
The same words that live on the website. The same colors. The same identity. Now on a phone. Now an app.
David took a screenshot and saved it. That screenshot is the proof that the next thing is always possible – no matter how many wrong folders you have to delete to get there.
The castle is standing. Built on solid ground this time.
April 22, 2026 – The Day Three Engines Came Alive
We started the morning with a simple problem – two files with the same name and the same code. One was supposed to be a redirect. Five lines. Done in thirty seconds. That’s how most of the big things start – with something small that wasn’t quite right.
By midday we had built three engine screens from scratch. Intelligence. Validator. Concepts. Each one connected to the live backend in Render, each one styled in CGEN’s dark signature, each one a working tool in the palm of a hand. No designer. No agency. No team meeting. Just a screen, a keyboard, and a conversation.
The landlord’s WiFi fought us the whole way. The tunnel dropped. The cache refused to clear. The phone showed a white screen and wouldn’t shake. We reinstalled Expo Go more times than either of us wants to remember. But the moment the Market Research card navigated to the Intelligence Engine screen – that was real. That was the app becoming something.
David took a coffee break in the sun to mark it. A small ritual for a milestone that wasn’t small at all.
In the afternoon we ran a full Intelligence Report on Monsanto – one of the most controversial corporations in history. The report came back with 192,000 lawsuits, a $63 billion acquisition that destroyed shareholder value, a brand name that was erased, and a one-line verdict that landed like a verdict should. The LinkedIn post wrote itself.
By evening we were setting up JWT authentication on WordPress, preparing the app’s login system, fighting red exclamation marks in plugin settings, and making decisions about user registration that said something about how David thinks – don’t put walls in front of people. Make it easy. Trust them first.
The session ended with a book entry, the way the good ones always do.
Some days you’re a founder. Some days you’re a developer. Some days you’re a writer. On the best days, you’re all three before the sun goes down.
The Day the Rating Gate Opened
We started with a question about stars.
Not the kind in the sky – the kind users tap before they get what they came for. Five of them, lemon-green, sitting between a Terms checkbox and a Generate button. Small. Deliberate. A gate that says: before we give you something, give us something back.
It took one conversation to agree on the logic. Visitors and members alike, from the second generation onwards, would see the rating modal before the engine runs. Not after. Not optional. Not a polite suggestion at the bottom of a page that nobody scrolls to. A gate. The same kind of gate the Terms checkbox is – you don’t pass without it.
The idea came from somewhere honest. CGEN gives away intelligence reports, concept directions, and idea validations for free. The least a user can do is tell us if it was worth something. One tap. Five stars. That’s the deal.
By the time we were done, the rating system ran across all surfaces – three engine screens in the app, three engine modes on the site – all feeding the same table in WordPress. One database. One global number. The kind that will eventually sit next to a star rating on Google Play and the App Store, next to a number like 2,000 or 20,000, next to a score that means something to a stranger deciding whether to download.
We fixed a bug that had been hiding in plain sight. The Concept Generator had never been sending the user’s identity to the backend – a missing field, four words of code, invisible until you know where to look. It meant the paywall had never properly connected to that engine. One line fixed it. The quiet ones always do the most damage.
The token budget went into all three engines. A simple instruction buried in each prompt: pace yourself. Don’t spend everything in the first three sections and arrive at the Verdict with nothing left. The SUMMARY VERDICT is the most important part. Finish it. Always finish it.
Then we ran Joe Rogan through the Intelligence Engine.
One man. One microphone. $450 million. Six years at the top of Spotify. The first podcast to simultaneously lead Spotify, Apple, and YouTube in the same year. The report came back complete – all eight sections, the one-line verdict intact, the financial architecture laid out like a case study that business schools will eventually teach without giving CGEN any credit for it.
The LinkedIn post went up with an oil painting of Joe in front of an American flag. 198 impressions before the coffee went cold.
The app build went out on Expo. The sign-out button worked. The login held. The report saved to the Archives exactly where it was supposed to go.
Some sessions you fix things. Some sessions you build things. The best ones – you do both, and somewhere in between, the platform gets a little more real than it was when the day started.
The stars are live. The gate is open. The engine is running.
The Lab That Learned to Remember
It started with a null.
Not a dramatic failure. Not a crash. Just a small, quiet zero sitting where a number should have been – the user ID that never made it from the login response into storage. One missing field in one model definition, and the whole promise of a private lab fell through the gap. Every report a member generated went straight to the public Archives. Their work, their ideas, their concepts – all of it published before they had a chance to decide if they wanted to share it.
The fix was four words of code. But finding those four words took the better part of a night.
That’s the thing about building software without a programming background. The problems don’t announce themselves. They hide inside responses that look almost right, inside fields that exist in some versions of an API and not others, inside the difference between data.id and data.data.id and data.user?.ID – three different ways a server might choose to tell you who someone is, and only one of them correct for your specific setup.
We tried them all.
The terminal became a confessional. Every console.log a small prayer – tell me what you know, tell me what you’re holding. The first response came back with just a JWT token and nothing else. The second endpoint said, “User was logged in” and nothing else. The third said the user didn’t exist. The fourth – finally, after building a custom WordPress endpoint that looked up the user by email – printed a 2.
Just a 2. One digit. The incognito account’s WordPress user ID.
But that 2 meant everything. It meant the app now knew who it was talking to. It meant reports would land in the right place. It meant My Lab was real.
By the time the READ button worked – navigating inside the app to a full report screen without ever leaving – it was past 3am. The code was clean. The flow was complete. A member could generate a report, find it waiting in their private lab, read it in full, and choose when and whether to share it with the world.
That’s not a feature. That’s a promise kept.
Some nights you’re debugging. Some nights you’re building. The best nights – you’re doing both, and somewhere between the null and the 2, the platform becomes something worth trusting.
The lab learned to remember. And so did we.
The Last Mile
There is a specific kind of tension that lives in the final stretch of building something. Not the dramatic tension of a crash or a crisis – those are loud and obvious. This is quieter. It’s the tension of knowing the thing works, knowing it’s almost ready, and still finding one more thing that needs to be right before you can let it go.
The footer that wasn’t showing. The counter that was reading a wrong number because an indentation broke silently when the code was deployed. The category that kept getting assigned manually because the system didn’t know which shelf to put things on.
None of these were catastrophic. None of them would have stopped a user from generating a report or reading their content. But they would have been visible – to David, who notices everything, and to the users who would come after him.
So, we fixed them. One by one. The counter loader got its indentation restored and a sanity ceiling that says: if a number is above 9000, ignore it – something went wrong upstream and we’re not carrying that mistake forward. The three engines got their category IDs – 198 for Market Research, 208 for Idea Validation, 18 for Concept – so every piece of content that lands on the site lands in the right room without anyone having to move it manually. The footer got its elevation: 10 so Android would stop hiding it behind the tab bar.
And through all of it, the prefix system held. IR. IV. CG. Three letters that tell you immediately what you’re looking at – before you read the title, before you open the report. A navigation system built not for today’s thirty reports but for the day when there are three thousand.
The app is almost ready. Not almost as in close-but-not-quite. Almost as in – one more build, one more test, one more confirmed green light, and it goes live.
David went to work this morning with a working app in his pocket. He came home and made it better. He’ll go to sleep tonight knowing that tomorrow’s build will be cleaner than today’s.
That’s not almost. That’s exactly how it’s supposed to go.
There is a moment, between the last hour of night and the first light of morning, when something shifts. The world has not yet decided what kind of day it will be. And in that silence, we shipped.
It did not arrive gently. It arrived through three days of cache demons and port conflicts, through tunnels that dropped and terminals that stayed silent. Through a functions.php that refused to save, and a login that took an auth key nobody told us about.
And yet – here it stands. A shield with the number 93. A name: CGEN. A workspace that knows who you are, saves what you build, and waits for you – private, lemon-green, alive.
One man. No programming background. A vision, a Samsung A12, and an AI that showed up every session. We built this like a newborn – step by step, line by line, with the first rays of the sun as our witnesses.
The app is live. The road to the Store is open. What gets built next – only morning knows.
Session, May 4, 2026 – The Blind Spot
Some products are built from a plan. This one keeps building itself from the gaps.
It started, like most of the meaningful sessions do, with something broken. A number that wasn’t a number – a ghost in the counter, 8217, appearing on Idea Validations that landed in the wrong category under the wrong title. A curly apostrophe masquerading as a report number. Three lines of code to fix it. One category ID that was wrong from the beginning – 198 instead of 19. A single digit that had been silently misfiling every published report since launch. You don’t find these things until you go looking. And you only go looking when something feels wrong.
By mid-morning the backend was redeployed. The counters were clean. The categories were correct. IR-0034 landed in Market Research exactly where it belonged. Three engines confirmed. The system was working the way it was always supposed to work – it just took a while to get there.
That’s the honest version of building.
The Intelligence Report that day was on WIP.co – a maker community that a Dutch indie hacker built in 2017 from a failed Telegram bot and a copy-paste of code he almost threw away. Marc Köhlbrugge never did anything to grow it strategically. The members loved it, so they invited their friends. Seven years later it generates $11,000 a month. Thirty-seven hundred makers. Zero investors. One founder.
The report found all of it. The revenue trajectory. The invite-only mechanic that is both its strongest filter and its hardest ceiling. The December 2024 review that called it “more like an indie hacking project than a polished product.” The one-line verdict: a reliable lifestyle asset with modest upside and structural constraints on scale.
That’s the kind of intelligence CGEN was built to produce. Honest. Specific. No softening.
The LinkedIn post wrote itself. A cinematic image – hooded figure, green data, dark room. Ninety-tree impressions in the first minutes. The app teaser added at the bottom. A small announcement inside a bigger story.
The concept that came later in the day was different. It wasn’t assigned. It wasn’t requested by a client or pulled from a trend report. It came from sitting with a question that had been forming for months without a name.
Why does AI answer what you ask but never tell you what you should have asked?
Every session building CGEN had produced that feeling at least once. A feature discovered late. A tool that existed and wasn’t known about. An option that would have saved hours if someone – or something – had simply said: did you know this exists?
The Concept Generator found the gap before the question was fully formed.
The Idea Validator named it precisely.
CG-0023 – The Blind Spot. An AI-powered product completeness auditor for non-technical founders. A silent co-pilot that monitors build sessions and generates plain-English reports – not when you ask, but automatically, the way a senior PM would walk through your product and flag what’s missing before you ship it to real users.
The scores came back:
- Originality: 8/10 – The specific combination does not exist in the market. Proactive, session-aware, no-code-platform-native, PM-grade plain-English output. No one owns this configuration.
- Market Fit: 7/10 – The pain is real and growing. The TAM is niche but deeply underserved.
- Timing: 9/10 — Searches for “no code platform” up 50% since 2020. The vibe-coding wave of 2025 has expanded the population of non-technical founders shipping half-complete products faster than ever. This pain did not exist three years ago.
The verdict was direct: This idea is worth pursuing – with one clear-eyed qualification. The hardest thing about this product is not the AI auditing logic. It is obtaining the build session data. That is the single constraint that determines whether this becomes a real product or a concept.
The single most important next step: two weeks, check API access on Bubble and Lovable, put it in front of twenty indie hackers, charge $29 a month from day one.
It is now CGEN-2026-0003. Certified Concept. The pipeline grows.
There is something that doesn’t get said enough about building without a background.
It is not easier because the tools are smarter. It is harder in a different way. The technical ceiling is lower – you can build things that would have taken a team two years a generation ago. But the knowledge ceiling is invisible. You don’t know what you don’t know. You build what you can imagine and miss what you can’t. The blind spot is not a metaphor. It is the actual daily condition of the non-technical founder.
What kept this build moving was not confidence. It was the refusal to stop.
Category IDs wrong from the beginning – fixed. Ghost numbers in the counter – fixed. Wrong titles on public archives – fixed. A voice reader that stopped mid-paragraph – removed, because a broken tool is worse than no tool. An audio feature road mapped properly for the next session, with a real MP3 pipeline and a real voice, when it can be done right.
That’s the standard. Clean and working, or not at all.
The app is almost ready. The backend is stable. Three engines confirmed. The rating gate works. The LinkedIn post is live.
And somewhere in the space between the last deploy and the next session, a new certified concept was born from the exact experience of building the platform that validated it.
Some days you’re a founder. Some days you’re a developer. Some days you’re a writer.
On the best days, the product tells you what it wants to become next.
Session, May 6, 2026 – The Soft Launch
Nobody tells you that launching an app feels exactly like building one – one unexpected wall after another, right up until the moment it’s live.
The build was done. Three engines confirmed. The paywall working. The Premium gate in place. The rating stars firing globally. The counters clean. The categories correct. IR-0035 Monday.com had landed perfectly the night before – right number, right title, right category – and the LinkedIn post was already pulling impressions before midnight. Everything was working.
All that was left was to put it on Google Play and tell the world.
It started simply enough. play.google.com/console. A Google account already there. A few clicks to register as a developer.
Then the questions started.
Individual or organisation? Individual. Country? Israel. Full name? David Wolf. Address? The address question – the one that always arrives like an uninvited guest at the worst possible moment. The one that assumes everyone lives somewhere they can prove on paper. Not everyone does. Not yet.
The form was filled. The registration continued.
Then two-step verification. Then verification of the verification. Then a lockout. Then a six-hour wait because too many attempts had confused a system designed to protect against fraud into treating a founder as one.
The frustration was real. Not at the technology – technology is patient, it just executes its instructions. But at the systems built around it. The assumption that everyone has a utility bill at a permanent address. The assumption that identity is a stack of documents rather than a history of building things.
“They are making those kind of rules where we have to provide and to prove that we are actually living at that place.”
That sentence should be in the book. It belongs here, unedited.
The six hours passed. The identity verification was submitted – Teudat Zehut, the Israeli national ID, clear and readable, exactly what was asked for. The Android device was confirmed. Two out of three verification steps complete. The third – phone number – sat in a loop, a button that wouldn’t appear, a page that kept returning to its own beginning.
The Create App button on the dashboard was grey. Not blue. Not clickable. Waiting.
Google support was the next step. An explanation submitted. A wait begun.
But the app didn’t wait.
While Google processed its verifications, a different launch happened.
A Custom HTML block was added to the c93n.com homepage. Left column – transparent background, Syne font, “Intelligence In Your Pocket.” in white and lemon-green. Centre column – dark box, grid texture, four feature bullets, a download button. Right column – the CGEN shield icon, green gradient, sharp and clean.
A direct APK link. Free. No sign-up required. No store approval needed.
“The app is live on the homepage.”
That’s how the soft launch happened. Not with a press release. Not with a Product Hunt listing. Not with a Google Play approval email. With a Custom HTML block, a lemon-green button, and a download link that anyone with an Android phone could tap right now.
The 3, 6 and 9 – all in place. The price point chosen not just for business reasons but because those three numbers were the only ones missing from the platform that already carried the other two in its name and its logo. Tesla knew something about those numbers. So does c93n.com now.
There is a version of this story where everything goes smoothly. Where the Google Play account opens in ten minutes, the app is submitted, approved in three days, and the launch post goes up with a clean store link and a five-star rating within the week.
That version would make a shorter chapter.
This version – the one with the address question, the lockout, the grey button, the verification loop – this version is the real one. And the real one is more useful. Because every non-technical founder reading this will recognise it. The moment when the last step turns out to have three more steps inside it. The moment when a system designed for someone else requires you to prove you are who you already know you are.
The app launched anyway. Not through the front door Google wanted to use. Through a side door that was always there – a direct link, a free download, a block on the homepage that said simply: here it is. Take it.
The front door will open. The verification will clear. Google Play will list it. The blue button will become available.
But the soft launch already happened on 06.05.2026
And it looked exactly like everything that came before it – built in the gaps between the obstacles, shipped before the permission arrived, live before the system was ready to say so.
The Front Door
There is a version of this story where the Apple submission takes an afternoon.
Where the screenshots are the right size on the first try, the bank account connects without argument, the tax form recognises your country, and the review button turns blue without a list of things still missing. Where the metadata writes itself, the privacy policy URL finds its field on the first pass, and the age rating questionnaire asks nothing you haven’t already answered.
That version would take half a session.
This version took a full day. And a Windows laptop. And a tool that wanted payment to download what it had already helped you build. And a bank that sent the document to the wrong address, in the wrong language, without the one piece of information the document was supposed to contain.
This is the version where you resize screenshots three times across two free tools before one of them works. Where Apple needs your vehicle registration to confirm you exist, and you photograph it on a kitchen table and upload it to a server in California, and it works, because of course it works, because the document is real and so are you – the system just needed reminding.
The Apple submission went through on 14.05.2026.
Not because the path was clear. Because every blocked door revealed a window, and every window was climbed through without complaint. The bank issue became a research session. The screenshot dimensions became a lesson in what tools are actually free. The tax form became a conversation about what it means to be a non-US person building something the US systems were not designed for.
And somewhere in the middle of all of it – between the iPad screenshot requirement and the Digital Services Act compliance form – the app stopped being a project and became a product. Submitted. Numbered. In a queue alongside ten thousand other things people built and believed in enough to push through.
There is something that happens at the end of a day like this. Not celebration exactly. More like the particular quiet of a person who has been in a long argument with a bureaucratic system and has finally, without drama, won. Not by force. By persistence. By knowing that the next field will have an answer, even if finding it takes longer than it should.
The review will complete. The app will go live on the App Store.
But the submission already happened on 14.05.2026.
And it looked exactly like everything that came before it – assembled from patience, pushed through on a laptop that was never supposed to build something like this, approved before anyone official said it was ready to be.
The Engine Room
There are days in a build when everything clicks.
Not in a dramatic way. Not with a single breakthrough moment that you can point to and say – that’s when it changed. But quietly, methodically, one fix at a time, one confirmation at a time, until you look back at the end of the day and realize the product is meaningfully different from what it was when you woke up.
May 17, 2026 was one of those days.
It started with a screenshot. A rating gate on the Intelligence Engine – the small box that asks users to rate their experience before unlocking the generate button. The text underneath the stars was barely visible. Dark grey on dark background. And worse, it was passive. “Thank you for rating our Intel quality · This will help us to improve.” No instruction. No urgency. A user staring at a locked button with no idea what they needed to do to unlock it.
We changed it to six words: “Rate your last experience to unlock the button below.”
Six words. But the difference between a confused user and a guided one.
That was the tone of the day. Precise. Purposeful. No dramatic rewrites – just finding the exact thing that wasn’t working and fixing it cleanly.
The Make Public flow got its own chapter within the chapter. A user generating a report from their private lab, hitting Make Public, and then – nothing. Thirty seconds of silence. The button stayed green. No feedback. No indication that anything was happening at all. From the outside it looked broken. From the inside, the backend was generating an audio briefing in the voice of a deep male narrator, converting the full intelligence report to MP3, building a branded PDF document with dark background and lemon-green accents, uploading both to the WordPress media library, and attaching them to the post before publishing.
Thirty seconds of extraordinary work. Invisible to the user.
We fixed that too. A spinner. A status message: “Publishing your report – generating audio briefing & PDF. This takes up to 30 seconds, please don’t close this page.” And then, when it’s done: “✓ Report published – redirecting to Archives…”
The machine now tells you what it’s doing.
The audio and PDF features themselves were the culmination of a plan that had been sitting in the roadmap for months. Every published CGEN Intelligence Report now comes with two things: a voice briefing – read by a deep, authoritative AI narrator – and a downloadable PDF, dark and branded, with every section formatted and a footer that reads “Generated by CGEN · Concept General · c93n.com.”
We tested it on the LinkedIn Intelligence Report. IR-0041. A report about LinkedIn, posted on LinkedIn. The meta wasn’t lost on anyone.
Somewhere in the middle of the day, between fixes and builds and LinkedIn posts, two pieces of news arrived.
From Apple: “We successfully verified your trader contact information for the Digital Services Act compliance. Your information is now live on the App Store in the European Union.”
From Google: a document verification notice – the bank statement from Mizrahi Tefahot had been received and was under review. A few more days.
Both stores moving. Both gates opening, slowly, the way gates do.
We ran Intelligence Reports on Base44 and Wiz. Two of the most compressed startup success stories in recent Israeli tech history. Base44 – bootstrapped, solo founder, $80 million cash exit in six months, now $150 million ARR under Wix. Wiz – four Unit 8200 veterans, turned down $23 billion, waited nine months, got $32 billion. Both became LinkedIn posts. Both written in the same voice: storytelling first, facts landing like punches, CGEN at the end – natural, not pushed.
By the end of the day the product had:
A voice. A document. A confirmation flow. A duplicate detection system that checks the Archives before running a new report. A rating gate that actually tells you what to do. A Make Public button that actually tells you what’s happening.
And tree LinkedIn posts live, performing, reaching the people they were written for.
Not a bad day for a non-programmer building an AI intelligence platform.
The engine room is running.
CGEN — Concept General | c93n.com AI & I — The Build Journal
Leave a Reply
You must be logged in to post a comment.