Multi-agentic Software Development is a Distributed Systems Problem (AGI can't save you from it)
Summary
The article argues that multi-agent software development with coordinating LLMs is fundamentally a distributed systems problem, formalizes a consensus-based model, and links it to impossibility results like the FLP theorem and the Byzantine Generals problem. It then discusses mitigation approaches such as verification, tests, and potential failure detectors, and points to additional distributed systems literature for context.