Research Package 001

Primary Research Question

Why is working with an AI model fundamentally different from working with traditional software?

Initial Hypotheses

H1. The largest performance difference comes from workflow design rather than prompt wording.

H2. Durable project artifacts outperform long conversational history for long-lived engineering work.

H3. Engineers with accurate mental models of LLM limitations adapt more effectively than those relying on trial-and-error prompting.

Evidence to Gather

• Transformer architecture fundamentals
• Context window limitations
• Retrieval-augmented generation
• Human-computer interaction research
• AI-assisted software engineering studies
• Knowledge management practices
• Developer productivity studies

Competing Viewpoints

• Prompt engineering is the dominant skill.
• Model quality dominates workflow quality.
• Long conversational memory is sufficient for most work.
• Multi-agent systems provide little benefit.

Each viewpoint will be evaluated against available evidence.

Deliverables

• Literature map
• Concept map
• Mental models
• Lesson 1