From: aidotengineer
Specifying intent to AI systems is identified as the second of four patterns in AI-native development, representing a shift from focusing on implementation details to defining desired outcomes for AI agents [06:17:00]. This means developers increasingly care less about the actual implementation and more about communicating what they want the agents to build [06:25:00].
Evolution of Intent Specification
-
Markdown Specification Files Initially, a simple markdown file could be added to a prompt as a specification, offering a basic way to define shared functional or technical requirements without repeatedly rewriting them in the prompt [06:39:00].
-
AI-Generated Plans Once the desired elements are defined, AI can assist in building a detailed plan. GitHub, for example, has added task-oriented capabilities that translate user intent into step-by-step actions [07:04:00].
-
Intent-Based Coding This approach moves beyond simple chat interactions or text completion. It involves defining tasks, allowing the AI to build a plan, and then generating the code based on that plan [07:18:00].
-
Specification-Centric Tools The direction of tools is becoming increasingly specification-centric [07:39:00]. In this model, the code itself may not be the primary focus; instead, the workflow revolves around specifying functional, technical, and security requirements [07:40:00].
-
Program Manager Role Ultimately, the process could evolve to a point where one primarily manages the process as a program manager, rather than directly overseeing the coding process [08:05:00].
This pattern indicates a shift for developers towards roles similar to QA or architects, focusing on defining and guiding the AI’s creation process through clear intent [12:51:00].