Chase AI teaches practical uses of AI tools and builds workflows for coding assistants. His Claudex Loop project separates planning, adversarial review, implementation and inspection across Claude Code and Codex.[2][3]
Start with Claudex Loop →
Links
YouTubeCareer
Contributions
Separate building from the final inspection
Claudex Loop uses one coding assistant to build and another to inspect. Its full workflow gathers requirements, hardens a plan through review, implements approved work and checks the resulting changes. Either provider can host the process, so the choice of builder does not determine which model grades the result.[4]
Tie approval to the artifact that was reviewed
The workflow records a plan’s content hash and the inspected change set. A changed plan or later code changes need a new review. Failed executions and exhausted review limits remain unfinished outcomes; a blank reviewer file cannot count as approval. These records make it possible to see what a review covered, without treating a model’s clean report as proof of correctness.[4]
Tools, writing & talks
Tools
Start here
Claudex Loop
github.comA cross-model planning, building and inspection workflow with recorded review coverage.
Read the four phases and the approval rules before installing. Running the loop requires both documented coding CLIs and their accounts; try a small change with an explicit verification command first.
Creator and workflow packager[2][4]Original source
Writing, research & talks
- Chase AI channelyoutube.com
Tutorials on practical AI tools, agents and automation.
Educator and presenter[3]