A long weekend overflowing with tokens, so I turned my pain points into tools
Introduction
When it comes to AI subscriptions, I've always kept a 100/20 rule between Anthropic and OpenAI: 100 USD for the main one, 20 USD for the backup. In the last cycle Anthropic's performance dipped, so I moved my 100 USD over to OpenAI.
Recently, several large refactoring projects needed to move forward at the same time, and the rule finally broke. For about 20 days, I'm on 100/100 with both, like a Super Star from Mario ⭐
I already hate letting agents sit idle on a normal day. Now I'm keeping myself swamped, constantly checking whether they have something to do peak slave-driver boss
Over the last four-day weekend, on a whim, I turned the problems I'd been meaning to solve into tools.
In the blink of an eye, two OpenAI quota resets were gone 🫠
Tangent: point at it, change it
A whole paragraph, and it still changed the wrong spot
A while ago, my wife bought the Bambu X2D we'd been eyeing for ages and started putting it to work. Besides learning Fusion along the way, I also tried bringing AI in to help.
But every time I discussed changes with the AI, it never quite got there. I'd write a whole paragraph just so it knew I meant "the curve on the second corner, bottom right," wait forever, and it would change the wrong spot again 🙄
While digging around, I found that Fusion now supports MCP, so an agent can reach in and work directly. Still, two pain points remained for me:
- Fusion takes forever to launch on my machine.
(M1... 😅)The meshes agents produce are heavily fragmented, so even selecting something takes ages whenever I want to make a change. For models that aren't precise geometry, like anime characters or figurines, it's still better to leave them to AI tuned for the job, like Meshy. - There's still no way to tell the agent where I want the change. That's the real problem.
So, just point at it on the web
Pokémon web games have been popping up everywhere lately, one after another daring Nintendo's legal team, and it reminded me how far web performance has come.
Rather than building yet another Fusion 2.0, why not use WebGL on the web to work out exactly what to change with the agent first, and only then have it produce a measurable solid, ready to export as STEP or STL?
That's how the Tangent prototype was born. It's still being polished, but it has already become one of the tools we use every day for modeling.
Bangs: a toolbox living in the notch
My screen is only so big
While waiting for Tangent to generate models, I usually have music on and keep working with other agents.
But switching over to Spotify again and again kept breaking my flow. My screen is teeny-tiny as it is, and now a music player wants to squeeze in too 🙄
Operation: Kill Setapp
That reminded me of my slow-burning plan to replace everything I use from Setapp.
Earlier this year I built Coopy to manage my clipboard. Back then I wrapped it as a Tauri app, and it never felt quite native, so I kept making small compromises while working. Still, I put up with it for most of the year. Hard to complain about your own kid 🤡
Now that I have tokens to spare, I threw everything into one pot and made it a notch tool: Bangs.
Notch tools aren't anything new, but in this era where everyone reinvents the wheel, being able to add whatever I want, whenever I want, is just more convenient.
Somewhere along the way, it ended up more polished than Tangent 🤣
Wrapping up
With tokens overflowing, I cleared out the pain points stuck in my daily routine, but these two tools don't carry the same weight.
Bangs is, honestly, yet another notch tool; being able to tailor it to myself is what makes it fun. Tangent is the one where I see a real opening, and also the hardest to get right. Having AI build a model for you is pretty ordinary by now. What actually gets stuck is every change after that: whether it lands exactly where I mean, with as little back-and-forth as possible.
If they happen to solve your problems too, feel free to sign up, and I'd love to hear your suggestions and feedback later on 🙏 (or come be a lab rat with me 🐭)
A few days of Super Star left. Agents, back to work 🏃
for zhTw version: Token 滿溢的連假，把痛點都做成了工具