In building AI agents @cline , we've identified three mind viruses Mind Viruses are seductive ideas that sound smart, but don’t work in practice.
1. Multi-Agent Orchestration
2. RAG (Retrieval Augmented Generation)
3. More Instructions = Better Results
Let's explore why!
@cline 1. Multi-Agent Orchestration
The sci-fi vision of agents (‘rear agent, quarter agent, analyzer agent, orchestrator agent’) sending out a swarm of sub-agents and combining their results sounds cool but in reality, most useful agentic work is single-threaded.
@cline Complex orchestrations rarely deliver real value and usually add confusion. Its hard enough to make the models work in a single thread let alone doing all this parallel orchestration logic that adds not just implementation complexity but model interpretation complexity.
@cline 2. RAG (Retrieval Augmented Generation) for Agents
RAG is a mind virus. It seems powerful on paper, but in practice even something as basic as GREP often works better, especially for agents.
@cline The RAG hype doesn’t translate to practical agent workflows because RAG leads to really scattered code that doesn't lead to a useful "understanding" for the model.
@cline Its almost always better for models to list the files and search through them with grep and then open and read the whole thing(Like a person). @cline was using this since forever but @ampcode and @cursor both shifted towards it.
@cline @AmpCode @Cursor 3. More Instructions = Better Results
The myth that stacking the system prompt with more and more “instructions” leads to smarter models is flat out wrong. Overloading the prompt confuses the model as more instructions often lead to conflicting advice and sensory overload.
@cline @AmpCode @Cursor You end up playing whack-a-mole with behaviors instead of getting useful output. For most frontier models today its just better to get out of the way of the models rather than the constant yelling at them to steer them a certain way. Measure your words(tokens) carefully.
@cline @AmpCode @Cursor Once again, all these ideas are very enticing and if you didn't spend all day tinkering with AI you would think they all make sense except that they don't. Our stances will them will change as underlying models improve.
