When your AI needs to do better
A Claude Code skill that reads your AI session logs and treats your frustration as a signal of workflow friction, then hands you a concrete fix.
What is “do better”?
Do Better is a Claude Code skill that reads your own AI session logs and treats frustration as a signal. Where other tools track efficiency — tokens, speed, usage — Do Better watches for the friction moments your logs reveal, like the tone that creeps in when something isn’t working, finds where they keep recurring, and hands you a concrete fix. Built by John Weng, founder of Bricolas.
Does anyone else get frustrated with their AI tools?
Everything’s working, and then suddenly it’s not. It forgets something important, or botches a task it nailed five minutes ago. And because you know the AI won’t lash back (and maybe you’ve had a long day), you take it out on the thing: “How do you not know this? We JUST did this. Look it up.”
Just me? Maybe.
But here’s what I keep coming back to: that flash of frustration is data. It’s the moment something in how you and the AI work together quietly broke. A context you assumed it had, an instruction that didn’t land, a loop you keep repeating. The irritation is the smoke. There’s usually a fire worth finding.
Here’s the thing I noticed: there are plenty of tools that measure how efficiently you use AI. Most coding tools have them built in now. But none of them, that I’ve found, watch for the human signal. None of them notice when you got frustrated and ask why.
So I built one. It’s called… do better. It watches for those friction moments in your own session logs, finds where they keep recurring, and hands you a plan for what to change. It’s not another dashboard of metrics; it tells you what to do differently.
It’s built for Claude Code, but point whatever AI tool you use at it (it assumes you’re in something integrated like Claude Code or Cowork, not claude.ai). Want to try it? It’s right here: github.com/jdubdevs/do-better. Made it better? Ask your Claude to build in your ideas, then send it back so others get it too. If you’re on GitHub, you can post it right on the repo; if not, just write me.
I use AI as a non-technical leadership development professional, and I genuinely want in on the vibe-coding trend. I built Aperture with Claude Code, I lead the AI for Leadership Practitioners Community of Practice, and I like helping people work a little more easily. This isn’t the money-making part of the business. Just a passion project.