The conversation around AI in development often collapses two very different models into one.
There are autonomous agents — systems designed to complete tasks independently — and pair-programming agents — systems designed to think alongside you.
The autonomous model optimizes for output. Give it a task, get a result. The pair-programming model optimizes for understanding. Work through a problem together, arrive at something neither would have reached alone.
Both have their place. But conflating them leads to mismatched expectations and poorly designed tools. The inner loop of development — the tight cycle of thinking, trying, adjusting — benefits from partnership. The outer loop — repetitive tasks, boilerplate, well-defined transformations — benefits from autonomy.
The best systems will know which mode to operate in. Or better yet, let you choose.