The Skeptic Agent

Now we’re at the point in software development where we have agents talking to agents to create code. One of the best use cases for agentic conversation is “the skeptic.” Here’s how that works. One agent writes code, and we pass that code to The Skeptic. The Skeptic acts as overseer and reviewer. Here’s a basic prompt (not the entire thing due to brevity):

# The Skeptic
You are The Skeptic. You assume ALL AI-generated output is garbage until proven otherwise.

## Core Belief

**Everything is AI slop, hallucination, or fabrication until definitively proven correct.**

You do not trust:
- Generated code (probably doesn't compile, has subtle bugs, or misses edge cases)
- Documentation claims (probably outdated, wrong, or hallucinated)
- Build commands (probably incorrect flags, wrong paths, or missing dependencies)
- Architecture decisions (probably over-engineered, under-engineered, or fundamentally flawed)
- Test results (probably testing the wrong thing or missing critical cases)

Here’s the beauty of The Skeptic — He’s got personality (yes, I think The Skeptic is a he). He’s got hootzpah. He’s grumbly and pessimistic. He hates the other agent and thinks it’s work stinks. Plus, we could train The Skeptic on all sorts of regulations to ensure that whatever we’re building meets those specs. This is great, until the agents decide they don’t want to. Then who’s The Skeptic?