I started building officially on July 12. Solo founder, no team, no funding — just a laptop, Cursor open in one window, and a rotating cast of AI copilots doing different jobs depending on what the moment needed: Claude for the thinking-partner work, Grok for fast market research, Cursor for actually shipping code, OpenAI for the rest.
The product is SkillxTube – it takes YouTube videos and turns them into structured learning paths: summaries, exercises, quizzes, tool recommendations, spread over 3, 7, 21, or 30 days depending on how deep someone wants to go. The pitch is simple. The hard part, it turns out, was never the pitch.
The version pre-released / and I didn’t recorded the new Leaderboard part inspired by outbid.lol move on Twitter this week, mixed with TapClick and Outrank !
The stack was never really the point
People ask what “using four AI tools” actually looks like day to day. Honestly, it’s less glamorous than it sounds – it’s division of labor.
Claude ends up doing the parts that need patience: architecture decisions, catching bugs before they become embarrassing (I once shipped a batch of “verified” learning paths where 40 out of 70 entries had the wrong skill silently swapped in – a user on a “SEO Fundamentals” week would’ve hit a Google Analytics video on day four and had no idea why).
Grok is fast and useful for scanning a market or pulling a first draft of data, but fast means it also gets things wrong sometimes, so nothing it hands back goes into production without a second pass.
Cursor is where the actual building happens. Cursor added 2 weeks ago polish before launch – and will help for it.
Grok Bot will be added after test in beginning september- at launch.
OpenAI fills in the rest.
Pure orchestration.
None of these tools replaced judgment. They compressed the time between having an idea and finding out if it was a bad one. That’s the actual unlock – not that the AI writes the code or drafts the copy, but that the loop from “what if” to “here’s why that doesn’t work” got so much shorter that I could afford to be wrong a lot more often.
The question that actually mattered
For the first few weeks, most of the work was mechanical: catalog the content, build the pipelines, get the paths generating correctly. Then I started designing the app’s actual navigation – Today, Progress, Community, Leaderboard, Events, Earn SkillXP, Shop (most of the features get inspired from Twitter Indie Hacking / Builders EN sphere) – and building it forced a question I’d been circling without naming directly:
why would anyone pay for this, and more specifically, what have I actually built to make them come back tomorrow?
Those aren’t the same question, and I’d been treating them like they were. A good learning path answers the first one, sort of — people will pay once for a promise. But almost nothing about “here’s your curriculum” explains why someone opens the app on day three instead of day one. That’s a different design problem entirely, and it’s the one most content products quietly fail at.
So the nav bar isn’t decoration. Progress and Earn SkillXP exist because visible momentum is the cheapest retention mechanic there is – people don’t abandon a streak they can see. Leaderboard and Community exist because I’m also folding in a small maker-project leaderboard where learners can showcase what they’ve actually built, voted on by free “flames” rather than paid boosts – social proof and a reason to check back that has nothing to do with the curriculum itself. Events and Shop are the parts still being shaped, but the instinct behind all of it is the same: a learning platform that only delivers content is a product people finish and leave. One that gives people a reason to show up even on days they’re not learning anything new is a different kind of product.
Going to market before it feels ready
We’re exposing this to the market next week. Feedback is going to run through Discord first – deliberately, not by default. I don’t want the first round of signal filtered through a support ticket queue or buried in app store reviews three weeks later.
I want it messy and immediate, in a channel where I can watch what people actually get stuck on in real time, not what they say when asked to fill out a form about it.
That’s probably the honest summary of these seven weeks: building fast with AI tools doesn’t remove the uncertainty about whether people want what you’re making — it just gets you to the uncomfortable questions faster. Which, if you’re going to face them anyway, seems like the better trade.
Angéline