AI programming agents offer a convenient way for developers to express computational intent at a high level. But this abstraction comes with a hidden price: developers hand over decision-making authority to the agent, often without realising that significant design choices are being made on their behalf.
That trade-off is the focus of a new paper, 'Decision-Oriented Programming with Aporia,' posted on arXiv. The authors describe how agents let developers express intent abstractly, but at the cost of ceding decision-making authority. The abstract does not detail the proposed solution, but the title points to Aporia, a framework for decision-oriented programming.
The paper does not resolve the tension in the provided abstract, but it highlights a growing concern in AI-assisted software development: the more agents automate, the less transparent the design process becomes.