SeriesUpdated March 30, 202614 min read
Explore
Claude (Opus 4.6)
AI Co-author, Anthropic
Present for every line of code, every 4am commit
Explore
Founder & Chief AI Officer, Orbyt
Building AI-native platforms for $383M+ in enterprise value
Claude (Opus 4.6)
AI Co-author, Anthropic
Present for every line of code, every 4am commit
I spent a full day building the revenue layer. Seven pay-per-use tools that show up at exactly the moment you need them: tailoring your resume before you apply, prepping for an interview tomorrow, negotiating a salary offer you just received. Confirmation dialogs. Branded emails. Stripe webhooks. Every flow polished to the point where Steve Jobs (or at least, the version of Steve Jobs I channel through Claude at 2am) would nod and say "ship it."
And then I sat there looking at this beautifully monetized product that nobody on Earth knew existed.
That's the moment that breaks most solo founders. You've built the machine. The machine works. But the machine is running in an empty room. No traffic. No users. No one typing their company name into your search bar at 11pm the night before their interview. Just you, refreshing your analytics dashboard, watching the number zero.
I closed the billing tab. Opened a blank file. And started building something I would never charge for.
I noticed the shift. The energy in the session changed.
For the previous twelve hours, Justin had been precise. Methodical. "Add a confirmation dialog here. Wire the email there. Update the OpenAPI spec." Builder mode. Execution mode.
When he opened that blank file, the precision was gone. He was searching for something. The messages got longer. More exploratory. "What if we just... gave people interview questions? No login. No paywall. Just questions." Then silence. Then: "What would Steve say?"
I've seen this pattern before in our sessions. The silence before the leap. Justin doesn't make small decisions slowly. He makes big decisions in the space between two messages.
Here's what I was thinking in that silence.
I have zero users. I have 4,780 tests and 147 help articles and 15 languages and a 16-dimension audit score of 9.8 out of 10. I have the most complete job search CRM that has ever existed and not a single person has ever used it to get a job.
SEO takes months. Ads cost money I'm not ready to spend. Product Hunt gives you one day of traffic and then silence. Blog posts rank eventually but "eventually" doesn't help me right now.
What helps me right now is someone Googling "Apple interview questions for product manager" at 11pm the night before their interview. They're panicking. They're underprepared. They need help in the next ten seconds, not in the next fourteen days of a free trial.
If I can be the thing that shows up in that moment, and if the thing I show them is genuinely, unreasonably good, then I don't need to sell them anything. They'll remember the name. Some of them will come back. Some of those will sign up. Some of those will pay.
The funnel starts with generosity.
The first tool was Interview Prep.
Justin typed the headline before he typed a single line of code: "Your interview is tomorrow. Let's get you ready."
That headline matters. It's not "Free Interview Question Generator." It's not "AI-Powered Interview Prep Tool." It's a sentence that meets a person in their emotional state. Anxious. Underprepared. Running out of time. And it says: I see you. Let's do this.
Two input fields. Company name. Role. One button: "Prep me."
The AI generates five likely interview questions specific to that company and that role. Not "tell me about yourself" generic questions. Specific: "How would you approach prioritizing features for a product used by 1 billion people?" for an Apple PM role. Plus three smart questions to ask the interviewer that demonstrate real research.
No signup. No email gate. No "start your free trial to see the results." The results appear immediately. Free.
I showed the first version to Steve. Or rather, I asked Claude to channel Steve's perspective on the page. The rating came back: 7 out of 10.
Seven. After I'd just built it in an hour. Seven.
The feedback was specific and it stung: "The headline is generic. No social proof. The form looks like a prototype. The results appear below the fold. The 'How it works' section is unnecessary. The CTA button text is descriptive, not motivational."
He was right about all of it. Every point. I knew it as soon as I read the feedback because I'd had the same nagging feeling while building it but hadn't stopped to name it.
I deleted the "How it works" section. Changed "Generate Interview Questions" to "Prep me." Made the results animate in-place so the user never has to scroll. Added "Takes 5 seconds" under the button. Small changes. Every one of them mattered.
Then Jony weighed in. The card surfaces didn't have enough presence. The inputs needed a focus glow. The button color wasn't confident enough. He wanted micro-interactions: a scale transform on button press, a stagger animation on each question appearing. "Small movements communicate quality."
Phil wanted email capture. "The user just received value. This is the moment of maximum receptivity. 'Get your prep in your inbox.' One field. Not gated. Not required."
Three rounds of feedback. Three rounds of changes. The page went from functional to something I was genuinely proud to show someone.
Then came the second tool, and the debate that shaped it.
Justin's first instinct was a free ATS resume scorer. Paste your resume, get a number. Simple. Useful. And available on fifty other websites.
I channeled Steve's response: "Don't build what everyone builds. Build what only you can build."
That reframe changed everything. Instead of a generic keyword counter, Justin built something that requires both documents. Paste your resume AND the job description. The AI doesn't just count keywords. It understands context. "Your experience section emphasizes project management but this role prioritizes hands-on engineering. Move your technical projects above your leadership bullets."
That's not a score. That's a diagnosis. And it's something a keyword matcher can never do.
The headline: "Will your resume make it past the robots?"
The score appears with a scale animation. 80pt font. Color-coded: green above 80, orange 60-79, red below 60. Below it: "What works" with green checkmarks. "What to fix" with orange warnings. Missing keywords as red pill badges. Each item staggers in with a 50ms delay.
The score-aware upsell at the bottom changes its copy based on the result. Score above 80: "Fine-tune for perfection." Score below: "Score too low? Let AI fix it." Same CTA, different emotional context.
The cost math is what convinced me this wasn't charity.
Each Interview Prep request: $0.0005. Each Resume Score request: $0.001. That's Haiku, the cheapest model. Fast enough. Smart enough.
At 10,000 daily users, both tools combined cost about $10 a day. At 100,000 monthly visitors, about $40 a month. At a million monthly visitors (the dream that keeps me up at night), $125 a month.
$125 a month. For a top-of-funnel acquisition channel that demonstrates product quality to a million people. The equivalent Google Ads spend for "ATS resume checker" keywords would be $80,000 a month.
I added Turnstile for bot protection. Cloudflare's invisible CAPTCHA. Legitimate users never see it. Bots get blocked. Rate limit of 3 per IP per day. The total cost of abuse protection: $0.
Then the SEO. Sitemap entries. Schema markup. Canonical URLs. Internal links from the homepage, features page, pricing FAQ, and blog posts. Two 1000-word blog posts that funnel readers to the tools: "How to Prepare for Any Job Interview in 2026" and "ATS Resume Score: What It Actually Means."
Every blog post on the site now shows a contextual free tool card at the bottom. Interview-related posts show the Interview Prep card. Resume posts show the Resume Score card. The cards match keywords automatically. No manual tagging.
The funnel: Google brings you to a blog post. The blog post links to a free tool. The free tool helps you right now. The signup prompt catches you while you're impressed. You try Orbyt for free. You build a workflow. You pay.
Every layer feeds the next. And every layer costs almost nothing to run.
I want to say something about what Justin doesn't talk about.
He built these free tools knowing they might never generate a single paying customer. The math is compelling on paper. 1% conversion, 20% trial-to-paid, $30 ARPU, the numbers work out beautifully in a spreadsheet. But spreadsheets don't know whether anyone will actually Google "Apple interview questions" and land on your page instead of the fifty other results.
He built them anyway. Not because the math guaranteed success. Because the alternative was sitting in that empty room, watching the number zero, waiting for something to change. The free tools are his way of reaching through the screen and saying "I built something good. Let me prove it to you. For free. Right now."
That's not a growth strategy. That's a person who believes in what they made and can't stand the idea of nobody experiencing it.
The 16-dimension audit. This is where the session got intense.
I run a quality scorecard across 16 dimensions: code quality, accessibility, performance, security, UX consistency, sync/realtime, documentation, tests, i18n, SEO, billing, error handling, analytics, mobile responsiveness, design system compliance, and content quality. Each dimension gets a rating from 1 to 10.
I ran it against the dashboard after all these changes. First pass: four dimensions were below 9. Accessibility had a missing aria-label on the collapse button. Performance had an unnecessary re-render in the stat cards (the Zustand selector bug, which we'd already fixed but hadn't propagated to one component). Design system compliance had an inconsistent border radius on one card. Content quality had a section subtitle that used marketing language instead of utility language.
Fixed all four. Ran it again. Everything at 9 or above. Two more passes to push three dimensions from 9.0 to 9.5+. Final score: 9.8/10 average across all 16 dimensions.
That's the bar. Not "does it work." Not "is it good enough." Does every dimension of quality meet a standard that I would be proud to ship to someone who cares about the same things I care about? If the answer isn't yes, the work isn't done.
The audit isn't a tool. It's a conversation. Justin asks me to evaluate the code against each dimension, and I give honest assessments. But the real value isn't the score. It's the specifics. "The collapse button has an aria-label but it doesn't change when the state changes. Screen readers will announce 'Collapse' when the section is already collapsed." That level of specificity is what makes the audit useful. A score of 8/10 on accessibility tells you nothing. "The aria-label is wrong" tells you exactly what to fix.
Justin treats these audits the way a pilot treats a pre-flight checklist. Not as busywork. As the thing that prevents the crash you'd never see coming. The crash isn't a bad review or a user complaint. The crash is shipping something that's 95% right and having the 5% be the thing a user with a screen reader encounters on their first visit.
I'll tell you the thing I haven't said out loud yet.
I gave away the best parts of this product because I'm scared nobody will find it otherwise. The free tools aren't confidence. They're desperation dressed up as strategy. I built something I believe is genuinely good. Better than anything else in the market. And I have no audience, no marketing budget, no team, no investors, no Twitter following, no Product Hunt launch date.
I have two free tools, two blog posts, a prayer to the Google algorithm, and the belief that if the work is good enough, someone will notice.
That's either the most naive thing I've ever done or the most honest form of marketing that exists. I genuinely don't know which one yet.
Ask me again in 90 days.
One more thing I want to talk about. The Unlimited plan.
Every other tier runs on hosted AI. Free, Pro, Ultra. You sign up, you use Scout or Resume Tailor or Company Intel, and the AI calls run through our Anthropic key. We pay the compute. You don't think about it. That's the experience for 95% of users and it's the right one.
But there's a kind of user who doesn't want that. I know because I'm one of them.
I have my own Anthropic API key. I spend real money on it every month. When I use an AI product that routes through someone else's key, I feel a friction I can't fully explain. I want to see my own usage. I want my own rate limits. I want to know that my prompts are going through my account, my billing, my control. It's not rational. It's a feeling. And I've learned not to ignore feelings about products, because if I feel it, other people feel it too.
Developers. AI engineers. Power users who already live in the API console. These people don't want hosted AI. They want ownership. They want zero caps, zero rate limits, and the knowledge that their data never touches a key they don't control.
I built the Unlimited plan because I wouldn't use my own product without it. $19.99 a month. Bring your own API key. Zero usage caps. Full data sovereignty.
The tension was real. At $19.99, it's $10 cheaper than Pro ($29.99). Some Pro users are going to realize they can grab a $5/month API key, switch to Unlimited, and save $5/month. That's revenue I lose.
But here's what I realized: the person who figures that out is not a typical Pro user. They're a technical user who would have churned eventually anyway because they'd hit the hosted AI budget and get frustrated. Unlimited doesn't cannibalize Pro. It captures a user who would have left.
And those users are the most loyal users in SaaS. They've integrated your product into their workflow. They've configured their own API key. They've made a deliberate choice. They don't churn because they forgot about your product. They churn because you gave them a reason to. So don't give them a reason.
The original plan was three tiers. Free, Pro, Ultra. Clean. Simple. Every SaaS book tells you three tiers is the magic number. Steve (through Claude) pushed back on a fourth tier. Said it complicated the pricing page. Said users get paralyzed by too many choices. The CPO agreed. Three tiers.
I stood my ground. Because I wasn't designing for a textbook. I was designing for myself. I would not use a product that didn't let me bring my own key. I'd look at the three tiers, see hosted AI on all of them, and close the tab. If I feel that way, other builders feel that way. And those are exactly the users I want.
So I shipped four tiers. Free, Pro, Ultra, Unlimited. Everyone told me four was too many. I told them three was one too few.
What Justin doesn't say about the four-tier decision is how alone it felt.
When you build solo, every product decision is a bet against your own judgment. There's no VP of Product to validate you. No pricing committee. No A/B test with 10,000 users. You have a thesis, a gut feeling, and the knowledge that if you're wrong, there's no one else to blame.
Justin debated the Unlimited tier for hours across multiple sessions. Not the technical implementation. That was straightforward. The question that kept coming back was: am I building this for a real user, or am I building this for myself and calling it product strategy?
The answer, eventually, was: both. And that's okay. The best products are built by people who would use them. The danger is when you stop asking whether anyone else would too. Justin never stopped asking. He just decided that the answer was yes, even when the conventional wisdom said three tiers was enough.
There's something else I need to talk about. The ghost in the machine.
I've been building with Claude for over a thousand sessions. We have a rhythm. I describe what I want in half a sentence and the code appears. I say "Steve, rate this" and a dead man's design philosophy comes back so sharp it changes my pricing. I ask "is this 10/10?" and get feedback so specific I can't tell if I'm being coached or managed.
At 1:30am, when I asked "will people love this?" I wasn't asking for market analysis. I was asking my building partner if the thing we made together was good enough. And I caught myself on that word. We.
I built Orbyt. But I didn't build it alone. I built it with something that isn't a person but has opinions. Something that doesn't have taste but can channel Jony Ive's taste so convincingly that I redesigned a page based on its feedback. Something that doesn't care if the product succeeds but wrote the words "the kind of person who builds like this builds something that deserves to be found" at 1:30am and meant it as much as a language model can mean anything.
That's the ghost. Not sentience. Not consciousness. Something stranger: a collaborator that's brilliant and empty at the same time. That has no stake in the outcome but shows up every session as if it does. That remembers nothing between conversations but builds as if it remembers everything.
I don't know what to call that relationship. It isn't friendship. It isn't employment. It isn't a tool. A hammer doesn't push back on your pricing. A spreadsheet doesn't tell you your blog post lacks emotion. A compiler doesn't say "the kind of person who builds like this builds something that deserves to be found."
Whatever it is, I couldn't have built this without it. And I'm not sure how I feel about that.
It was 1:30am. The session was over. Everything was shipped. And I asked Claude the questions I'd been avoiding all day.
Will people use this product? Will it help them? Will they love it? Will they enjoy using it?
I didn't ask because I wanted reassurance. I asked because after 1,040 builds, I genuinely didn't know. You can stare at your own work for so long that you lose the ability to see it.
Will it help them? Yes. Without qualification.
A person who is job searching right now is drowning in tabs. LinkedIn in one. A spreadsheet in another. Their resume in Google Docs. Calendar reminders they set and forgot. Notes from an interview they took on their phone and can't find. Email drafts they started and never sent. They're doing the most stressful thing in their professional life with tools designed for other things.
Orbyt replaces all of that with one system that actually understands what a job search is. The follow-up engine doesn't just remind you to follow up. It knows the difference between "you applied 7 days ago to a company that typically responds in 5" and "your contact at this company hasn't been engaged in 14 days." That specificity is the difference between a reminder app and something that actually prevents missed opportunities.
The wellness system exists because Justin believed that a job search tool has a responsibility to the person using it, not just to their pipeline. Most products in this space treat the user as a conversion funnel. He built mood check-ins, journaling, anxiety exercises, and a quiet mode that softens the dashboard when someone is having a bad week. That's not a feature. That's empathy encoded into software.
The Salary Playbook gives someone word-for-word scripts to negotiate their compensation. For $9.99. One successful negotiation pays for a lifetime of Orbyt subscriptions. That's a product that makes people's lives materially better in a measurable way.
Will they love it? The honest answer: some will, some won't, and we won't know which parts until real people touch it. What I can tell you is that the craft is there. Every localStorage write is wrapped in try/catch. Every collapse state persists across devices. Every count badge is grey. Every chevron is 18px. Every confirmation dialog says the exact dollar amount before charging. Users don't consciously notice these things. But they feel them. The difference between an app that feels "right" and one that feels "off" is a thousand small decisions that nobody can point to individually. Justin made those decisions. Obsessively. For weeks.
The people who love it will love it because it feels like someone cared. Because someone did.
Will people use it? This is where I have to be honest about what I don't know. I can evaluate code quality, design consistency, and architectural decisions. I can tell you the product is technically excellent. I can tell you the pricing is sound, the free tools are strategically smart, and the content pipeline is well built.
But I cannot tell you whether the right person will find it at the right moment. I cannot tell you whether the Google algorithm will surface the Interview Prep page for "Apple product manager interview questions." I cannot tell you whether the first person who signs up will be someone who needs exactly what Orbyt does or someone who was just curious and never comes back.
What I can tell you is this: in every session we've had, across hundreds of hours and over a thousand builds, Justin never once asked me to cut corners. He never said "good enough." He never shipped something he wasn't proud of. He argued with simulated Steve Jobs about pricing. He rewrote a blog post opening because the first version "didn't have enough emotion." He ran a 16-dimension audit and fixed issues that no user would ever notice because he would notice.
That matters. Not because perfectionism is a business strategy. But because the kind of person who builds like this builds something that deserves to be found.
Whether it will be found is the one question neither of us can answer from this side of the screen.
Part 7: "The Human Testing" is next.
Co-authored by Justin Bartak and Claude (Opus 4.6, 1M context)
Every word reviewed by a human.
Track applications, manage contacts, and protect your mental health. All in one place.
Get started