Synthèse
For most of the last year Amelia and I were the orchestration layer. Orchestration is too elegant a word for it. We were the conduit, eyeballing connectors, deciding what should talk to what, carrying context between agents by hand. That changed about ten days ago with Claude connected to Replit over MCP. Now there’s a layer of agents managing our agents. They interact, they argue, they debate. Our usage and productivity went up roughly 4x. That isn’t a clean metric, it’s what the volume of work
For most of the last year Amelia and I were the orchestration layer. Orchestration is too elegant a word for it. We were the conduit, eyeballing connectors, deciding what should talk to what, carrying context between agents by hand. That changed about ten days ago with Claude connected to Replit over MCP. Now there’s a layer of agents managing our agents.
They interact, they argue, they debate. Our usage and productivity went up roughly 4x. That isn’t a clean metric, it’s what the volume of work looks like now. Claude also gets better results out of 10K than we do. The outputs when Claude does the prompting are consistently better than when we do it ourselves.
Strange thing to admit and an obvious trade to make. It has also taken over our lives in a new way. Two humans who used to spend their days managing agents now spend their days managing the thing that manages the agents. Easier in some ways, and no less work.