The Quality of AI-assisted Software Depends on Unit of Work Management

Let’s say your AI agent has a 5% chance of making a mistake. I’m not just referring to hallucinations—it could be a subtle mistake because it forgot to look up some documentation or you missed a detail in your specification.

In an agentic multi-turn workflow, which is what all coding workflows are converging to, this error compounds. If your task takes 10 turns to implement, you will have a (1 – 0.95)10 = 59.9% chance of success. Not very high.

If you don’t provide the necessary information in the context to do a good job, your AI will hallucinate or generate code that is not congruent with the practices of your codebase. It is especially brittle at integration points of your software system.