We're hiringCareers at Moonlit Labs
Research

01 · Business & AI

Where agents break inside organizations

Where do AI agents create real leverage inside a company, and where do they quietly fall apart?

Area
Business & AI
Status
In progress
Models in scope
ClaudeGPTGrokGemini

The question

Most agent demos run in a clean room. Real organizations have messy handoffs, unwritten rules, and people who need to trust the output before they act on it. This note follows agents from the demo into the day-to-day workflow.

Approach

  • Map the workflows where agents are being introduced
  • Observe the handoffs between people and agents
  • Run the same tasks across frontier models and compare behavior
  • Document failure modes as they appear, not after the fact

What we're watching

  • Which tasks hold up once the inputs stop being tidy?
  • Where do people stop checking the agent's work, and should they?
  • What does a graceful failure look like, and who notices it?
NextEvaluating enterprise agents under real conditions