AI can generate a dashboard in seconds.
But getting something you’d actually send to a client, publish on the web, or reuse tomorrow? That’s a different problem.
I shipped two open-source projects to close that gap.
The bigger idea: agent-generated UIs shouldn't be disposable chat artifacts.
They should be portable, reproducible deliverables you can inspect, reuse, host, and share.
That's what I wanted. So I built it.
Both projects are open source:
→ github.com/flesler/open-canv… - CLI + exports
→ github.com/flesler/cursor-ca… - web runtime + live demo
Try it: npm install -g @flesler/open-canvas
If you're building with coding agents, give them a try.
Same CLI. Python, Go, Rust, Zig. AI wrote all four.
The slowest port found bugs the others sailed past.
The hard language got easy once I stopped porting and started adding features.
github.com/flesler/scip-cli
Zig lesson, paid for in hours: read the stdlib source. Docs and training data lag the language.
0.17-dev deleted basic file I/O, thread pools, CLI args. Models still generate 0.13.
Unstable elegance loses to boring APIs that still exist next month.
For AI-written code, from this experiment:
Go: clear errors, ships fast
Rust: painful rewrite, fewer landmines later, features get easier once the patterns are in
Zig: compiles in a blink, burns the calendar on API churn
Tables, quotes, timings:
github.com/flesler/scip-cli/…
I recently launched aip 🚀
CLI project management built for 🤖. Create projects/tasks, track progress, enforce structure.
It has been evolved by my own 🤖 based on usage logs. PRs welcome, Let your 🤖 run `aip usage quickstart` and see what it thinks!
github.com/flesler/ai-projec…
@NousResearch @Teknium it's been running on my Hermes agent, a nightly cron looks at all terminal calls (aip or not) and suggests improvements and/or new commands. It's been working well!
Can use:
aip help skill --mode hermes > ~/.hermes/skills/some_toolset/SKILL.md
After 9(!) years, parallel 2.0.
34 commits in a single evening
github.com/flesler/parallel
