AI Agent Use Cases
Views —

AI Agent Use Cases

Added today

44All 189

Prompt to Prepare a Shopping Cart from a Recipe Photo

The author reports that Grok Bot receives a recipe photo, organizes the ingredients and quantities, and creates a shopping list by excluding ingredients already available at home.

Grok Bot
Life & Hobbies

1–20 of 189

Dev

Porting an Image Restoration Model to Run in the Browser

This is a report of how the author modified an image restoration model using Claude Code to make it runnable in the browser, fixing errors along the way.

Claude Code

How it works

  1. The author prepared the public model code, ONNX and WebGPU runtime environments, and selected sample images to check in the browser.
  2. Claude Code was used to repeatedly modify the PyTorch and CUDA-based model for ONNX and WebGPU execution, addressing errors and checking browser results each time.

Results

The author reports porting the model execution code to fit the browser environment, resolving errors, and sharing the process and public results.

Notes

There is an actual report and public result from the author. The finished tool itself is not an agent; the process is an example of agent usage.

Source

Simon Willison의 Moebius 이식 제작기 Original post ↗

Example prompt · Claude Code

Goal: Port an image restoration model to run in the browser.

Input and environment: [model repository], [sample input], [target browser], [original output for comparison].

Reference original: https://simonwillison.net/2026/Jun/22/porting-moebius/

Prepare an implementation or workflow that can be followed in a test environment, based on the published workflow. Do not guess internal code or settings not disclosed in the original; if alternative connections are needed, let me know.

Connections and materials to prepare: Prepare public model code, ONNX and WebGPU runtime environments, and select samples to check in the browser.

Agent operation to implement: Use Claude Code to modify the PyTorch and CUDA-based model for ONNX and WebGPU execution, repeatedly fixing errors and checking browser results.

Scope of operation: There is an actual report and public result from the author. The finished tool itself is not an agent; the process is an example of agent usage.

The result is image restoration code that runs in the browser. Please also organize the required setup steps and actual execution method. After checking the tool's output, choose the next task, and separately note any materials you could not access or tasks you could not execute.

Verification: Compare the original and browser outputs using the same input, and check for conversion errors, unsupported operations, or memory shortages. Show the input used for verification, the actual output, and items not yet checked separately.