Skip to content
02 · Field Guide

Vibe coding, properly understood

Strong vibe coding is not one spectacular prompt. It is a controlled loop of intention, context, implementation, review, and correction.

What vibe coding is

Vibe coding is a conversational way to build software. You describe an outcome in natural language, an AI coding agent creates or changes the code, you inspect the result, and together you iterate.

It can feel magical when a page appears after one request. The useful skill, however, begins after that first result. Strong vibe coding is not one spectacular prompt. It is a controlled loop of intention, context, implementation, review, and correction.

Traditional hands-on codingDesigner-led vibe coding
Write each implementation detail directlyDescribe intent, behaviour, constraints, and quality
Read the code to understand every changeReview both the experience and the agent’s change summary
Debug primarily by editing codeReproduce the problem, provide evidence, and guide the agent to diagnose it
Own technical execution directlyShare execution with AI while retaining human accountability

Vibe coding is excellent for learning, prototypes, experiments, internal tools, and small products. It can also contribute to production work, but production software still needs engineering review, security, testing, maintenance, and clear ownership.

Your role is changing, not disappearing

AI can generate plausible interfaces extremely quickly. Plausible is not the same as purposeful.

The AI does not automatically know:

  • which user problem matters most;
  • what evidence supports a decision;
  • what your organisation is allowed to do;
  • what the brand should feel like;
  • which words may confuse or harm people;
  • which edge cases are important;
  • when a technically working result is ethically or experientially wrong.

Your value moves upward — from manually producing every artefact to framing the work, supplying evidence, directing decisions, and judging quality.

AI chat versus an AI coding agent

An AI chat assistant usually responds with advice, text, or snippets. An AI coding agent can inspect a project, create files, install packages, run commands, read errors, and edit multiple parts of the codebase.

AssistantCoding agent
Discusses the workActs inside the project
Produces suggestions or code snippetsCreates and edits real files
Needs you to transfer changes manuallyCan run and check the application
Best for thinking, critique, and preparationBest for implementation and debugging

Many products combine both modes. The important question is not the brand name. Ask: Can this tool only advise me, or can it act on the project?

Where each discipline contributes

DisciplineHigh-value contribution in vibe coding
Product designUser flows, interaction models, information architecture, states, priorities, and acceptance criteria
Visual designArt direction, hierarchy, typography, colour, spacing, responsive composition, tokens, and motion principles
Content designLanguage strategy, labels, instructions, empty/error/success states, tone, readability, and recovery copy
User researchEvidence, assumptions, hypotheses, test plans, behavioural signals, synthesis, and limits on what may be claimed

Everyone can prompt. The point is not for everyone to contribute the same prompt. The point is to bring different forms of expertise into one build.

The core loop

Do not ask an agent to build an entire complex product in one move. Large requests create large amounts of unreviewed decision-making. Small slices make quality visible and mistakes easier to reverse.

Study check
  • Explain vibe coding without using the phrase “AI writes the code.”
  • Name one decision AI can assist with and one decision your team must own.
  • Take a current project and identify its smallest useful interactive slice.