So the workspace can be disposable, but the work is still first-class. The only caveat is unsnapshotted filesystem edits: if the folder disappears before jj captures them, they're gone. Once captured in the repo, they're recoverable.
That makes temp workspaces in /tmp really nice for experiments or AI agents. I can use one myself, or have an agent use one for a separate task, and only decide later whether the result should be integrated.
jj workspaces feel much nicer for parallel work than Git worktrees, and the important part is not the shared repo. In jj I can just start working in a workspace, and the work becomes part of history through @ without first creating a branch or deciding to commit.
Era of a mayfly software. Everything has its application. In this case, it seems like a far more advanced one-time shell scripts.
👏
Go @npmx.dev!
These look beautiful 😍
Do you ever clean up your old GitHub repos and forks? 🧹 Wondering if anyone actually bothers or just lets them pile up.
This is really great! I subscribe to newsletters, but sometimes I find it hard to catch the emotion the author infuses into their writing. This manages to fill in that missing piece and adds a personal touch. I'm sure many of us will appreciate it - I certainly do.
Feeling that way isn't an exception but a rule. Comparing yourself to others is natural; what's more crucial is how you respond. There's no one right answer - sometimes I pause and reflect, other times I push myself harder. Ultimately, I improve. Use this as a fuel to go from good to great.
Oh, the luxury of just finding a calm place and reading a book. Like a lotion for an irritated mind.
Rewrote my agent coding skills from checklists to thinking instructions. Specific rules like "use X not Y" ground the agent and go stale as the language evolves. Guiding it to ask "is there a simpler way?" and self-research before deciding lets it go much further. Cognitive quality instructions?
One thing I'm realizing with code agents: the speed gain shouldn't become pressure to ship more. I want to use that time for quality. Better tests, thoughtful design, simplification. Boring is the goal.
GitHub stars inflation.
Reflecting on a pattern: each time we adopt new tools, we discover they weren't built for the new intermediary — search engines, app stores, algorithms — and rush to optimize. The product drifts a little further from the person it was meant to serve. "AI-first" codebases feel next in line.
I've gotten more from practical experience than from memorizing "best practices" — and that goes for prompting AI too. Concrete examples show you the tradeoffs. Best practices usually skip that part.
In practice, "when does this apply?" gets me further than arguing about whether it's the right way.
The era of mayfly software has begun.
Raw skill and knowledge are more important than ever before.
What if MCP had the same kind of progressive disclosure as skills? 🤔
Have you ever considered that your PM is “vibe coding” you
One useful addition to agents.md I made to better align CLI agents with the “reality” is: “use `date` command to get my current date and time when you need to work with dates.”
🤩💪🏻
Andy exemplifies effective web practices and design. Highly recommend the courses offered by Andy and his team.
Really digging the workflows the Wispr Flow provides. Added their blog to the list of favourites. Their approach reminded me a lot of this blog post wattenberger.com/thoughts/our... by @wattenberger.com
What are the good starting points to start hacking with atproto? 🤔
🧐 @tuple.app
Cloudflare 🫂
I mean - my usual usage depletes the budget much quicker, not like the amp itself is more expensive.
Noticed that amp became much more expensive (MCP might be contributing here, not sure 🧐). Tried their race mode for a change (haiku 4.5) - super fast, results are very close to what I need most of the time, and it is cheaper.
Deploy to any platform with one click 👀