AI-assisted development

The human writes; AI suggests. The human remains the primary author.

Also called: AI pair programming, code completion

A developer writes the code and an AI assistant suggests completions, explanations and edits inline or in chat. The human accepts, edits or rejects each suggestion and remains the author of record. Every line still passes through a person who understands it.

The name borrows from pair programming, a practice popularized by Extreme Programming in the late 1990s, where two developers share one keyboard. GitHub launched Copilot as a technical preview in June 2021, billed as “your AI pair programmer” and built on OpenAI’s Codex model. ChatGPT, released in November 2022, added chat-based explanation and debugging.

Early evidence came from a 2023 controlled experiment in which developers with Copilot completed a set programming task 55.8% faster than those without. Later field studies found smaller and more varied gains.

1. 01The developer writes code in their editor.
2. 02The assistant suggests completions inline, or answers questions in a side chat.
3. 03The developer accepts, edits or rejects each suggestion.
4. 04The code goes through the team’s normal review and tests.

Typical tools: GitHub Copilot, Cursor, JetBrains AI Assistant, Gemini Code Assist, general chat assistants.

Compare all five bands

Solid: AI does core work · Hatched: partial or informal · Dashed: done by people. Compare all bands

Good for: Production work in an existing codebase; coursework with learning goals

Less suited to

• Large, repetitive or multi-file changes, where reviewing every suggestion is slower than delegating the task
• Courses where the learning goal is writing the code yourself

Signs you’ve outgrown it

• Tasks routinely span many files
• You find yourself pasting the same instructions every session
• The team wants to hand off whole tasks, not lines

Where it runs: The developer’s usual editor (VS Code, JetBrains) with an assistant plugin, plus existing version control.

Data

Whatever the project already allows

Credentials

Unchanged from normal development

Key setting

Enterprise tier with training on university code turned off

On campus

Duplicate-code filtering on where offered; licenses through central IT

Compare environments across bands

• Personal space or sandbox only
• No real or institutional data
• Throw it away, or rebuild it properly before sharing

Risks

• Automation bias: accepting plausible suggestions without checking them.
• Provenance: suggestions that closely match licensed code.
• Skill atrophy: learners skipping the practice that builds judgment.
• Review fatigue: more code to review, with the same reviewers.

1. GitHub, “Introducing GitHub Copilot: your AI pair programmer,” June 2021
2. Peng et al., “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot,” 2023
3. Besser, Jensen & Katz, Open Research Europe, 2026