You should try using canvas-driven-development Since we're building 100x more features, that means you have 100x more QA but you can't try everything by hand so have your agents do it: write evals for how it should work, and put screenshots in a big canvas

with Paperclip, we have to make sure our agents ask questions, write plans, save artifacts, create subtasks across dozens of models in various harnesses evals are the only way - but we're testing e2e and you want to have screenshots and the full chat when things go wrong

You aren't even testing specific outputs anymore, you're testing probability distributions

❤️❤️❤️

Screenshots in a canvas work because review scales by glance, while evals scale through explicit checks. The gap is what neither one catches: state that only breaks on the third run. How are you catching those today?

This looks awesome. I had built something similar for a website re-do. I am not a programmer by trade so most of these management systems are text heavy and i never want to look at them or read through. I like the idea of a visual decisions lane post testing

the canvas becomes a living regression suite for the whole product

Damn ur doin God's work

Works wonders, added support for PlanDesk ( for prototypes sams as canvas. Helps visualise flows and screens before hand and keeps iterations grounded.

neat

💪💪💪💪💪

How do you feel about using canvases? Building something that’s canvas based and it’s been hit or miss feedback.

This is just a one off to try to see all states at once and not really an IDE

Got it. If you find it interesting somehow, would love to hear more about it

Wow that’s very nice

Infinite canvas for infinite work

Feels that way sometimes

got a skill for this?

I'll publish one!

my dog!

the hardest part of building is knowing what’s worth shipping when ideas costs $0. taste becomes worthless if ai slop is the moat.

100% True love rick so much. exactly how i feel building ai systems as a non-coder 😊

I know what a canvas is, your video doesn't really show where the big benefit is though. How are you using them? This looks quite chaotic and hard to overlook. Are these iterations of one task? Is this just visual evidence for the tests an agent handled before?

This is my biggest bottleneck, 100X more QA that I can't trust anyone to do

canvas is just a UI trap

thirst trap