A local ‘clone’ answers coding agent blockers for the user

The author says autonomous coding tools often stop when they face an unclear choice and ask the user. They built a local model that uses their profile and past decisions to answer low-risk blockers as they would. The system still asks the user when it is unsure or when the action is always blocked, such as force-push, production database work, deletion, secrets, or external sending.

Key points

  • The post focuses on coding agents that stop and ask the user when they hit unclear choices.
  • The author built a local model using Gemma through Ollama.
  • The model uses a user profile and past decisions to answer low-risk blockers.
  • New user answers are saved as precedents for future similar blockers.
  • Hard-rule actions like force-push, production database work, deletion, secrets, and external sending are not auto-approved.

Quick term guide

autonomous
The ability of an AI to complete tasks or make decisions without constant human guidance.
local model
An AI model you run directly on your own computer, with no internet connection or external service needed.
profile
A saved AI setup with a specific role, tools, and work style.
production
The live version of a service that real users use.
database
A large collection of organized data used for search and analysis.
coding agents
AI programs designed to autonomously perform tasks like writing or fixing code.
coding agent
An AI tool that writes or edits code from a person’s instructions.
logging
Keeping records of what happened in a system so it can be checked later.
Read original