Eyeballs for my ai buddies

Feature Requests
1

As workflows are shifting away from code rolling and to beter designs, better planning, and better testing/execution… we really need ai eyeballs. I’m REALLY tired of copy pasting snipped screenshots of things…

context: I know we all have very different projects… here is my workflow/toolchains:

project: Unity Game
c#, and shaders hlsl, .compute code mostly. heavy lifting in shaders
raymarching 3d fractals via realtime audio cymatics.
it’s ALL visual… so instead of the coder/tester/etc I’m now the planner/tester/architech/ai-testmokey…

Most of my convos w ai now center around testing and being the ai’s eyeballs so that we can work through any coding whoops along the way… and with shaders – trust me there are many even for the ai code helpers. So the amount of hours spend typing up what my eyes are seeing like visual artifacts between fractal modifier regions that only appear from certain angles… and that’s one of the easier ones to describe. I rely on a combo of text and still image captures to try to get ai to understand the outputs of the code we sling together. madening at times… and other times the ai “sees” in a still image what I missed at 100fps. So basically… When ai helper eyeballs?

2

example of my meaning… https://www.youtube.com/watch?v=P_igJA4eOoI

in this testing of cymatics I found that a color inversion of sorts is happening at peak cymatic values somewhere through the rendering pipeline. trying to describe the exact nature of this “anomaly” I’d like to fix to ai is tediously rough … cant wait for a toolcall w “share screen” video feed into my ai’s visual cortex. I know there are all sorts of vision models out there. Even if the tool needed a one frame per minute limit at version 1 of eyeballs, would be a reclaim of about 4hours a day lately trying to find the right words to describe the ineffable to my ai help.

3

Unfortunately, AI cannot yet replace humans everywhere. A workaround can be sought: if the artifact is large enough, one can create a gallery and depict multiple frames in one or several images. Perhaps then the AI will be able to understand what is wrong. However, it is clear that a custom MCP needs to be written for such tasks.