AI
Researchers map the anatomy of AI coding prompts to predict development success
A new study proposes a three-part framework for understanding how developers craft prompts for AI coding assistants—and tests whether prompt structure predicts project outcomes.
As AI coding assistants become embedded in software workflows, the ability to write effective prompts has become a core engineering skill. Yet little systematic research exists on what makes prompts work or how they should evolve throughout a project. A new registered report from researchers proposes a structured framework to investigate this gap.
The Prompt Triangle model#
The work introduces the Prompt Triangle, a conceptual model that breaks down AI code-generation prompts into three interdependent components:
- Functionality and Quality — the requirements layer that defines what the system should do and how well
- General Solutions — architectural guidance that shapes the overall design approach
- Specific Solutions — implementation constraints that govern code-level decisions
This three-part structure builds on earlier work viewing prompts as hybrid artifacts that blend requirement specifications with solution guidance. The framework aims to make visible how developers balance top-level intent with concrete technical constraints.
Testing prompt patterns and outcomes#
The research team has registered a Stage 1 protocol for a confirmatory study that will test four core hypotheses:
- How prompts change shape and emphasis over the course of development
- Whether developer experience or background influences prompting patterns
- Whether alignment between prompts and traditional requirements engineering (RE) activities predicts success
- What temporal patterns emerge as projects progress
The study combines three data sources: a controlled experiment with 30 participants, voluntary community uploads of prompts and code, and mined datasets from public repositories. Researchers plan to use dual-coding (independent human review) to classify prompts and link prompt structure to development outcomes.
Why it matters#
As generative AI tools reshape how software teams work, understanding the mechanics of effective prompting has become a practical necessity. This research aims to move beyond anecdotal advice toward empirical evidence about what prompt patterns correlate with better results. The findings could inform tooling, training, and process improvements for teams adopting AI-assisted development at scale.
