Is anyone working on local-first development tools? Does it sucks for everyone or just me?
Between using symlinks & handling Vite's caching, I feel like I’m trying to build a number of utility libraries to use across projects and had to reason about this messy, brittle workflow.
@gavinmcfarland Too long for x... even writing this little explainer gives me a little hope that I might not be crazy. Maybe we need to build better local-first development tools.
https://t.co/eTQDiAd53Q
bringing the scene to life with clouds rolling through the landscape
ported some of my volumetric cloud work from 2023 to webgpu and made it run in screen space as an effect
on top of that I finally figured out the issue I had with normals, the terrain looks way better now
@DavidKPiano having a hard time understanding if it’s true or if your meta-commentary/sarcasm about the current state of tools is peaking.
In any case, keep going 🫡
Why I don’t understand is how folks take all these agentic CLIs & code generation stuff to generate something as boring as UI… to generate dashboards & SaaS interfaces.
We have this new set of tools, that gives us extreme lateralization, to build almost wherever you want, and you use it to create the most boring, safest ripoff you could possibly made… are u joking?
Adds a 3D scene, renders ASCII.
OffscreenCanvas on worker.
w/ FPS throttling.
w/ character & color controls.
Color analysis & perf. controls on a worker.
Rendered on a browser based terminal.
Made w/#threejs
@_baku89@junkato Congrats to both of you! Coming from Baku, who has been so relentless at researching the design & engineering space makes it extra special 🫡
Super excited that our paper, co-authored with @junkato, on parameter-tuning widgets for creative software has been accepted to #UIST2025. It’s my first academic paper ever, which makes it way more special. #tweeqjs
https://t.co/pjwcjumLmF
If you have a piece of Three.js or R3F that no longer works, send me a link.
I’ll run it with my workflow, and I’ll reply with a codesandbox or codepen. If it fails, I’ll put the plan and the code on https://t.co/0nyFr7PLOR
Ok, need to share that I made a bunch of tools to work better with Claude Code…
One of the best use cases I found was using agents to outline a plan to refactor projects that I stopped working on because of dependency hell.
I worked on this issue for many months, developing strategies, techniques and data.
Yesterday I put it all together for the first time, and the refactoring and procedural update of the stack used on this app was completed in 2 or 3 rounds.
And I’ve been thinking about turning this into a proper agentic workflow, with a HTL to review, approve or request more actions.
One trick that consistently makes aware how to write better code is this:
- Write pseudo code (scrappy & capture the data and main fns)
- Write the initial version as a skeleton and ask an LLM to fill the gaps.
- Try to identify if the code can benefit from using an established design pattern.
- Create a version a more complex version of it.
- Find a library that does that but 100x better.
- Realize that you need to remind yourself to ALWAYS check if someone already solved the problem before writing any code.
- Out of pride use your overly complicated version.
- Fall asleep while your laptop drains, only to wake up the next morning asking yourself why I was doing in the first place?