Software Engineering in the Age of AI Agents.
Classical processes. Multi-agent systems. The future of how we build.
RUP was right about the value of structured process. It was wrong about the economics. Token deflation has collapsed the cost of specification, governance, and verification by five orders of magnitude. The math now works — and it changes everything.
A tour of the actual machinery: the six agents (Governor, Business Analyst, Requirements, Architect, Developer, QA), the artifacts they produce, the Progression Protocol, and real results from a brownfield reverse engineering run.
Three pipelines build the same library system. One invents a feature nobody asked for. This is what "show, don't tell" looks like — with real metrics, real code, and a surprise ranking system.
Everyone claims their multi-agent system "works." But compared to what? Three pipelines, nine metrics, five threats to validity, and one honest answer. Time to design an experiment that actually proves something.
The mapping between SDD spec formats and classical RUP artifacts isn't a coincidence — it's convergent evolution. How twenty-year-old process engineering turns out to be surprisingly good scaffolding for AI agents.
Multi-agent systems are powerful. They're also expensive. Meet the role that keeps them from burning through your budget while debating variable names. Token economics, circuit breakers, and why the cost problem is solving itself.
Write the spec. Let the machines do the boring part. What Specification-Driven Development actually is, who's using it, and why it's the missing piece to resurrect RUP.
What if the methodology we abandoned for being "too heavy for humans" was actually perfect — just waiting for non-human executors? Introducing AIRUP and the AI Governor.