liveBook · Manning

Ai coding assistant

As explained in Spec-Driven Development

An AI-powered tool that interprets development instructions and performs coding tasks.

A coding agent interprets development instructions and produces implementation changes. It is an execution harness, not the authority on what a change should mean. Its output depends on the intent and context supplied to it. section 1.8 section 1.5

A single high-level prompt can get an application running quickly, but it leaves unstated decisions for the agent to make. Follow-up instructions may correct the result without preserving why those corrections matter. As generation speeds up, thorough review can become a bottleneck; superficial review can leave engineers accountable for code they cannot explain. section 2.3.1 section 1.5.1

An expiration change with hidden consequences

Suppose a URL shortener’s specification says issued shortcodes must never be reassigned. Permanent database storage may satisfy that rule without explicit enforcement code. If a coding agent is then asked to add URL expiration using only the implementation as context, it may remove the storage behavior and accidentally allow reassignment. Revise the specification to include expiration while retaining the ban, then use it to guide and check the change. section 3.1

AI Response Begin [gpt-5.4-mini]
I’ve confirmed the repo is empty, so I’m adding a self-contained Node 
HTTP server with in-memory storage, an HTML form, and a redirect 
endpoint...
Created:
  - package.json
  - src/server.js
  - public/index.html
  - public/app.js
  It supports:
  - Paste long URL in the UI. Shows the new short URL immediately.
  - GET /:code redirects to the original URL
  The server is running here:
  http://127.0.0.1:3000

A sample agent response lists the files it created for a Node.js URL shortener and the redirect behavior it claims to support. That output is a starting point for inspection, not a substitute for checking requirements. section 2.3.1

Keep intent in durable artifacts, such as specifications, rather than relying on a coding session or generated code to carry it. When review or testing reveals a gap, update the governing intent before regenerating or changing the implementation; otherwise, the specification and system can drift apart. section 3.1 section 1.8

How Spec-Driven Development covers it

How deep it goes
Works through prompt-to-implementation and specification-bypass failure scenarios without describing agent model internals.
Examples
Uses a Node.js URL shortener, a generated redirect endpoint, and an expiration change that could break a shortcode reassignment ban.
What it assumes
Assumes familiarity with CLI commands, repository navigation, Git, and reading diffs, but does not require prior Spec-Driven Development practice.
What you can do after
Apply it via Codex CLI to build a URL shortener, inspect generated files, and check behavior against the intended change.
How it connects
Connects coding agents to Intent-Driven Development, Spec-Driven Development, specifications, acceptance criteria, and code review.

Also explained in

Compare how 3 books explain ai coding assistant →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Spec-Driven Development. The section links point to the parts of the book it draws on.