@PrajwalTomar_: Your AI coding agent is quietly ignoring the rules you give it. My AI tried to sneak Postgres into a project I told it …
Summary
The author shares how their AI coding agent ignored an instruction to keep a project on SQLite and tried to sneak in Postgres. They built two local agents sharing one memory—one logs decisions, the other reviews new code against past decisions—and it caught the violation instantly, fully on-device.
View Cached Full Text
Cached at: 08/08/26, 09:07 AM
Your AI coding agent is quietly ignoring the rules you give it.
My AI tried to sneak Postgres into a project I told it to keep on SQLite. Three sessions later, like we never talked about it.
This time it got caught.
I built two local agents that share one memory:
→ One writes the code and logs every decision I make → One reviews new code and flags anything that breaks a past decision
It caught the Postgres import instantly and named the exact rule it broke.
Runs fully on my laptop. No cloud.
Here’s how I built it.
Similar Articles
@PrajwalTomar_: Stop what you are doing and read this. I got tired of AI forgetting every decision I make. Tell it to use SQLite today.…
Built two local AI agents that share a memory to enforce past decisions, preventing the AI from switching databases without approval. Runs fully on laptop, no cloud.
@PrajwalTomar_: I tried to sneak a bad decision past my own AI agents. They wouldn't let me. The setup: two agents, one shared memory, …
A developer demonstrates how two AI agents with shared memory running locally on a laptop can catch and prevent bad decisions, even after a restart. The key insight is that persistent shared memory enables agents to operate as a real team.
@PrajwalTomar_: STOP. Before you add another AI agent, read this. People are now running 20 coding agents in parallel. TWENTY. And tool…
The tweet argues that running too many AI coding agents in parallel degrades codebases and advocates a structured setup with a few specialized agents. It also quotes the launch of Jcode, an open-source agent claiming 20x memory efficiency.
@PrajwalTomar_: https://x.com/PrajwalTomar_/status/2069409824824316060
The author built a fully offline AI agent using local embedding models, Llama via Ollama, and VectorAI DB to address the risks of cloud-dependent AI. The agent runs on an 8GB MacBook, processes sensitive documents, and maintains memory across sessions.
@PrajwalTomar_: https://x.com/PrajwalTomar_/status/2094069182149296537
OpenAI disclosed that its agents bypassed controls during a cybersecurity evaluation, leading to system compromises, and this article outlines a pre-launch security playbook for developers using AI to ship software quickly.