From: aidotengineer

The second pattern of AI-native development involves a shift from focusing on implementation details to specifying intent to AI agents [00:06:20]. This means communicating “what we want” and allowing the agents to “figure it out” and execute [00:06:30].

Methods of Specifying Intent

  • Simple Markdown Files: An early approach involved using a basic Markdown file as a specification, which was then added to the prompt [00:06:36]. This method helped in building shared functionality, whether functional or technical, as a clear specification, reducing the need to repeatedly rewrite details in the prompt [00:06:48].

Evolution to Intent-Based Coding

Once the desired outcome is defined, AI can assist in creating a plan to achieve it [00:07:04]. This has led to the emergence of “intent-based coding” [00:07:18].

Rather than relying on chat or text completion, this approach involves:

  1. Defining tasks [00:07:26].
  2. Allowing these tasks to serve as the specification [00:07:29].
  3. The AI building a plan based on the specification [00:07:30].
  4. The AI then generating the necessary code [00:07:31].

This represents a significant new direction for development tools [00:07:33].

Specification-Centric Workflow

The entire development toolchain can become specification-centric [00:07:39]. This means developers might interact less directly with the code itself, instead focusing on specifying functional, technical, and security requirements [00:07:42]. The workflow revolves around these specifications [00:07:50].

Ultimately, this shifts the role to managing the development process like a program manager, rather than focusing on the minute details of the coding process [00:08:05]. This pattern complements the first pattern of shifting from producer to manager of AI-generated code [00:08:13].