AI-Assisted Development

This content was generated with the assistance of AI and should be verified against the official documentation of each tool and framework before being relied on in production.

This section’s bibliography lists the reference material consulted while preparing these pages.

This guide targets GitHub Copilot in VS Code and JetBrains (with the Copilot CLI and Copilot on github.com) and Claude Code (CLI, IDE extensions, desktop and web), with details checked against each product’s own documentation as of 2026-09-28. The version baseline checked on that date is VS Code 1.139 with the GitHub Copilot Chat extension (which now ships every Copilot feature in VS Code), GitHub Copilot CLI 1.0.88 and Claude Code 2.1.283 — see VS Code 1.139 release notes, GitHub Copilot CLI — Releases and Claude Code — Changelog. Every page reuses DocsAssistant API, a small Spring Boot + Python sample repo that serves the same DocsAssistant running example as Vector Databases & RAG and LLM Foundations & Prompting — a Spring Boot controller under src/main/java/com/example/docsassistant/, its tests under src/test/java/…​, and a Python ingestion script under ingest/. Recurring tasks (add an endpoint, write tests, fix a failing test, refactor, review a PR) are shown in both tools side by side wherever both support the concept.

Every team, and every developer, sits somewhere on a spectrum. At one end is vibe coding: describing what you want and accepting what the assistant produces with little review, useful for a disposable prototype but risky for anything that ships. At the other end is AI-assisted engineering: the assistant drafts, but a human states constraints, reviews every diff, and owns the result exactly as if they had typed it — the working assumption for the rest of this guide.

Levels of autonomy

Assistants act at increasing levels of autonomy, and each level hands the assistant more of the loop:

  • Completion — ghost-text and next-edit suggestions as you type, accepted or rejected line by line.

  • Chat — a conversation, inline or in a side panel, that proposes an edit you still apply by hand or with one click.

  • Agent — the assistant plans, edits multiple files, runs commands and iterates against a goal inside your editor or terminal, asking for approval at configured points.

  • Background / cloud agent — the assistant runs unattended in an isolated environment (for example, assigned a GitHub issue), pushes a branch and opens a pull request for review.

flowchart LR A["Completion"] --> B["Chat"] B --> C["Agent"] C --> D["Background / cloud agent"]

Two effects show up as autonomy increases. The 70% problem: an agent often gets you most of the way to a working feature quickly, and the last stretch — edge cases, integration, real polish — takes disproportionately longer and still needs a human who understands the code. The knowledge paradox: the more an assistant writes for you, the less familiar you are with the resulting codebase, exactly when reviewing it carefully matters most. Both are reasons to keep a human reviewing every diff, however far up the ladder the tool is working.

Reading path. New to AI coding assistants: start with The AI Coding Tool Landscape, then Getting Started with GitHub Copilot or Getting Started with Claude Code, then Prompting Coding Assistants. Comfortable with chat and completion: move to Agent Mode and Plan-First Workflows. Running agents day to day: the Advanced pages below cover background agents, parallel sessions, CI automation, review and security. Leading a team: finish with Team Practices & Adoption.

What’s covered

Basics

Intermediate

  • Agent Mode — Copilot agent mode and Claude Code’s equivalent loop, how agents gather context, approving and steering tool calls, and reusing existing MCP servers.

  • Plan-First Workflows — explore → plan → implement → commit, spec-first workflows, plan mode in both tools, and reviewing the plan as a quality gate.

  • Managing Context — what an assistant sees, instruction files, session commands, scoping with @ / #, subagents for research, and token and cost awareness.

  • Choosing Models in Your Assistant — the Copilot model picker and Claude Code’s /model, matching models to tasks, and cross-checking with a second model.

  • Testing & Debugging with AI — generating and critiquing tests, test-driven debugging loops, and verification loops as the agent’s success criterion.

  • Refactoring & Documentation — refactoring recipes, the "majority problem", docstring/README/ADR generation, and legacy modernisation.

Advanced

  • Cloud & Background Agents — assigning issues to the Copilot cloud agent, Claude Code on the web and in GitHub Actions, and choosing bounded tasks these agents handle well.

  • Parallel Sessions & Worktrees — running several agents at once in git worktrees, splitting work by module, and a merge-conflict strategy.

  • Headless & CI Automation — claude -p and the Copilot CLI non-interactively, GitHub Actions integrations, scheduled runs and safe CI permissions.

  • Reviewing AI-Generated Code — a review checklist, Copilot code review on pull requests, Claude Code’s /review, and deterministic gates as the backstop.

  • Securing AI-Assisted Development — vulnerabilities that show up disproportionately in AI code, prompt injection, sandboxing, and secret scanning.

  • Team Practices & Adoption — golden rules, shared prompt/skill libraries, measuring impact without vanity metrics, and policy, IP and ethics.

  • Prototyping with AI — app builders and screenshot-to-code, the throw-away vs. evolve decision, and the path from prototype to production.

Cheat sheet

  • Cheat Sheet (PDF) — a single-page, printable summary of every concept in this sub-section.

Related pages

Bibliography

The following sources were consulted while preparing this sub-section’s pages. Every code example, command, flag and setting name was verified against the official documentation and current tool versions — never copied from a book — and where a book’s example was outdated, the page in question says so in prose and shows the current approach.

Requester-provided books, each a consulted reference for concepts and the practitioner’s angle only, never the source of an example:

oreilly.com answers automated fetches with an HTTP 403 response, but every publisher page above is a live, reachable page when opened in a browser.

Official documentation, grouped by tool:

GitHub Copilot (GitHub Docs):

VS Code (Visual Studio Code documentation):

Claude Code and the Claude platform (Anthropic documentation):

Other tools and frameworks:

Specifications and standards:

Papers and engineering articles:

These books are consulted references, not the primary source of any example on this sub-section’s pages. On any discrepancy between a book and a tool’s own official documentation, that documentation is authoritative.