f18m

Tbh, I was expecting more from HN recommendations. Some are to be expected, like tech books that are kinda irrelevant in a post AI era.

Been to Trader Joe’s 3 weeks in a row now and I can’t find sardines anymore.

I tested it and the videos look great, congrats on that.

That said, I’m struggling to see a thriving business here, for two reasons. First, the moat looks shallow. Anthropic and OpenAI are already building in this space, and it’s hard to see what stops them from shipping something better and cheaper fully integrated with their ecosystem soon.

Second, I don’t think customer behavior around AI-generated content (and especially in your case) is settled yet and this affects many similar business across the board. The videos are good, but perception is a separate question from quality. If your buyers are content creators, schools, and e-learning platforms, they’re ultimately selling to their own customers, and they’ll need confidence that the content won’t read as cheap, low-effort or similar to everyone else.

This might be the golden era for starting a company, but building a wrapper around an LLM is risky unless something outside the software holds it up (proprietary data, distribution, a regulated position, real switching costs or any external component). Otherwise the model providers or a well-funded copycat can rebuild it.

Wishing you guys good luck! All the best!

A fun project I started about a month ago. Make has long been my go-to script runner and build tool, simple and ubiquitous, but as my projects piled up, I kept hitting the same problem: I'd write a useful makefile target for one project, then copy-paste it into the next, and so on.

MKPM offers a way to share makefiles within teams. It has some limitations for now, but may evolve in the future.

Learning itself hasn't fundamentally changed because of AI, period. Anyone claiming otherwise probably isn't thinking seriously about what learning actually is. What has changed is the demand for certain types of courses, especially computer science and programming courses, which has declined significantly. Honestly, this was entirely predictable.

People don't want learn tools anymore. Programming education experienced tremendous growth from the 2000s onward because coding became a valuable skill in the job market. Today, the market has shifted. Learning to code in 2010 is, in many ways, the equivalent of learning to use AI effectively for software development in 2026. The underlying goal is the same: acquiring the skills that employers and the market value.

What's happening to David is the same thing happening to many "learn to code" platforms. Their catalogs were built for a different era, and declining demand reflects changes in the job market, not a decline in people's interest in learning.

Maybe David is simply tired of teaching programming and changing a catalog that he put a lot of hard work for the last 20 years seems like a lot of work, and that's completely fair. But he could also adapt his expertise and content to fit this new reality instead of relying on a curriculum designed for a thriving market that no longer exists.

I also consider myself largely self-taught in the subjects I genuinely enjoy and work with today - despite my academic degrees. Looking back on my academic years, I realize I “vibeschooled” quite a bit (or the pre-AI equivalent of it)—mostly because I was studying things I didn’t care about, with my main motivation being just to pass. The point I was tying to make on the post is that “vibelearning” in the same way as “vibecoding” (where code is the actual output) is inherently contradictory if it doesn’t result in actual, retained knowledge. When you vibelearn the wrong way, it’s easy to confuse the appearance of learning with actual knowledge retention. Great point about "going back and forth" under a teacher guidance. I haven't thought about this before, but I'm sure this is a great way of reinforcing learning and broadening perspectives. It's another behavioral dynamic that's hard to replicate through tech alone.

Is this the equivalent of Lua for Rust?

Never used Supabase before but I'm very much comfortable with their underlying stack. I use a combination of postgres, PostgREST, PLv8 and Auth0 to achieve nearly the same thing.