You don’t got my skills has a whole new meaning now

Most open-source comparison checkboxes are written by marketers. The version I want is a feature backed by code. Show me where it lives. Show me how it works. Then let my agent study it.

I don't think 'vibe coded' should be an insult by default. Sometimes it means a domain expert finally built the thing programmers kept telling them was too niche.

Open-source repos are full of already-solved problems. Most of that knowledge is trapped in code nobody reads. Indexing it at the feature level is the bet behind the new Opensource.Builders.

I’ve had a people ask me why I haven’t switched to Fable? It’s because every SOTA model require you to prompt it differently and the mental capacity it takes to context switch between 5.5 vs. 4.8 vs. Fable 5 is too much. There’s a great value to being able to use one model and one model, but understanding it’s nuances.

The difference between running an LLM in your terminal vs. using their website like chatgpt.com is becoming more and more evident. 6 months ago, they were able to do 90% of what we were able to do in the terminal. Now it’s 50%.

Vibe coding gets mocked because a lot of it is rough. Fair. But rough doesn't mean useless. Some rough projects are early versions of real products.

There is a kind of builder who doesn't care if the architecture is fashionable. They care if the tool works for their business. AI is going to give those people a lot more power.

I hope Apple gives Siri the ability to create shortcuts and run them. It’s the equivalent of codemode on iOS and they won’t need to add MCP support then

The person who didn't code before AI may be more willing to ask for the weird thing. That's where some of the interesting software will come from.

The Skill Builder is the part of Opensource.Builders I'm most excited about. Pick features from real repos, generate a skill pointing at the exact code, hand it to your coding agent. 'Use this as inspiration' stops being hand-wavy.

Programmers have defaults. Preferred stacks, patterns, taste, scars. Useful stuff, but also limiting. Someone new to coding with AI often starts with the product in their head, not the implementation constraints.

Before AI, 'just read the source' was technically true and practically useless for most people. Now an agent can read it with you. That changes what an open-source directory should even do.

I honestly think vibe coders may find some of the best uses for AI. Experienced programmers know the rules. New builders are still asking for things they don't know are supposed to be hard.

Old directories told you what existed. The thing I want now is a map of which open-source project has the best implementation of the one feature I need, and a link to that exact file.

A lot of software is about translating lived frustration into a system other people can use. AI just lowered the cost of translation.

Calling something a Shopify alternative tells you almost nothing. Does it have variants? Multi-currency? Discounts at checkout? Apps? Themes?

If you know the workflow, the edge cases, the customer, and the pain, AI makes you dangerous. If you don't, it helps you build the wrong thing faster.

Refs are really letting them play these conference finals. I respect it but hope I’m not speaking too soon

The farmer's daughter who understands water usage, land, labor, equipment, buyers, and weird seasonal constraints has better startup ideas than most people sitting in a coworking space guessing at agtech.