it’s out today.. six months later 🙂‍↕️ pdf + epub on leanpub chapter 6 is free if you want to see if it’s your thing leanpub.com/secondtimeisabug #secondtimeisabug

I realised today that @X has some kind of special criteria to let users "boost" their posts 🤣 It's okay.. almighty @X .... don't need your assistance lol

I'm creating this post only so that I can boost this post. Sorry. That's just how X works. I released a new book... if you find the sample chapter useful.. please buy.. :) leanpub.com/secondtimeisabug

There's a particular kind of unfinished project that keeps coming back. You stopped working on it months ago. Then you come across something that reminds you of it. A new tool appears. Someone mentions the problem you were trying to solve. You think: I should get back to that. And you mean it. For a few minutes. The project might be a half-built app, a document full of notes, or an idea you've explained to three friends. It still feels possible. You just haven't decided what to do with it. Collect enough of these and you have what I think of as "the someday folder". It doesn't have to be an actual folder. Some of it lives in browser tabs. Some in old repositories. Some entirely in your head. The difficult part is deciding which ideas still deserve your time. Try opening one this week and writing down what would make it worth returning to. What would you need to find out? That might give you a smaller, more useful next step than "finish the project".

At this point, building your own version control system, CI/CD pipeline, and deployment stack almost feels as secure as relying on the established ecosystem. With the growing number of attacks targeting open-source software and software supply chains, the trust we've placed in our development infrastructure deserves some serious scrutiny. And somehow, "just shut everything down until we figure this out" is starting to sound less like a joke.

Nobody: Absolutely nobody: Tech folks on X: "Name one technology that does [something] better than [insert my favourite technology here]. I'll wait." The engagement-bait template that apparently never goes out of fashion... 😂

Are we still discussing this?

you can just build "things"

turned off most of my MCP servers and claude code got sharper... every MCP server you connect costs context before you’ve typed a word. treat them like browser tabs. open what the task needs, close the rest

.@AnthropicAI can't control how its models behave and thus it's enforcing models to work against user instructions. Don't really know how I feel about it. Some of it is really becoming counter-productive right now. The flexibility of @OpenAI is welcoming... but at the same time, I'm not sure what it's able to control wrt the model behavior and autonomy. Weird times.

when an agent run goes sideways my first instinct is still to blame the model... it’s almost always missing context or no way to check its own work... a test it can run beats a smarter model most days!

every model upgrade i go through CLAUDE[.md] and delete stuff... half the rules were written for mistakes the old model made. the new one doesn't make them but still reads every line, every session the ones without a one line "why" next to them are the scary ones to delete 🥸