Vibe coding · frameworks

Spec driven development: write the plan, then let AI build

Spec driven development means you write down what the software must do before any code exists. The spec says who it is for, what it does and how you will know it works. The AI builds to that spec, and you check the result against it. Less guessing, less rework.

Why a spec beats a clever prompt

Most AI builds go wrong the same way. You ask for a feature. It works. You ask for the next one, and the first one breaks. A spec stops that loop.

The AI stops guessing

An AI fills every gap you leave with a guess. A spec closes the gaps before it starts.

You can check the work

With a spec, done has a test. You are not just hoping it looks right.

You can start again cheaply

If a build goes off track, hand the same spec to a fresh session and rebuild that part.

Other people can help

A coach, a developer or a second AI can read the spec and pick up where you left off.

How spec driven development works, step by step

You can do this with any AI coding tool. The steps stay the same.

  1. 01

    Write the spec

    One page. Who uses it, what they can do, what the rules are, and what is out of scope.

  2. 02

    Turn it into a plan

    Ask the AI to break the spec into small tasks. Read the plan. Fix it before any code is written.

  3. 03

    Build one task at a time

    The AI builds a task and runs its own checks. You look at the result. Then you save it.

  4. 04

    Check against the spec

    Go back to the spec. Does the app do each thing it says? If not, that is the next task.

What goes in a good spec

The user and the job

Who is this for, and what are they trying to get done? One or two sentences.

The screens and the steps

What the user sees and does, in order. Plain words are fine.

The rules

Who can see what. What must never happen. What happens when something fails.

The data

What you store, such as users, orders or notes, and how the pieces link.

The done test

A short list you can tick off. If every line passes, the feature is done.

What is out of scope

The things you are not building yet. This stops the AI adding extras you did not ask for.

Context engineering: give the AI what it needs to know

Context engineering is choosing what the AI can see when it works. The right files, rules and examples, at the right time. A good spec with bad context still gives you a bad build.

A project rules file

A short file the AI reads every session. Your stack, your naming and the things it must never do.

Skills for repeat jobs

Save the steps for a job you do often, so the AI loads them only when it needs them.

Live connections

Let the AI read your database or your docs through an MCP server, so it works from facts.

Less is more

Do not paste in everything. Too much context hides the part that matters.

Agentic coding: let the AI do the steps

Agentic coding is when the AI does more than suggest code. It reads your project, plans the work, edits the files, runs the commands and checks its own result. You set the goal and review the work.

It needs a spec most of all

An agent that works alone for an hour will drift without a clear target. The spec is the target.

Give it a way to test

If the agent can run the app and the tests, it can catch its own mistakes.

Keep the steps small

One task per run. Review it. Save it. Small steps are easy to undo.

You stay in charge

Decide what the agent may do alone and what needs your sign-off, such as deleting data or going live.

Questions, answered

What is spec driven development?

It is a way of building software where you write the spec first. The spec says what the software must do and how you will check it. The AI then builds to the spec, one task at a time.

Is spec driven development the same as vibe coding?

No. Vibe coding is building by talking to an AI. Spec driven development is a method you can use while you do it. The spec gives your prompts a plan to follow, so the build stays on track.

Do I need a special tool for spec driven development?

No. A plain text file and any AI coding tool will do. There are kits that add templates and commands, and they can help. Learn the steps by hand first, so you know what the kit is doing.

What is context engineering?

Context engineering is choosing what the AI sees while it works: the rules, files, examples and live data. Good context means the AI works from facts about your project, not from guesses.

What is agentic coding?

Agentic coding is when an AI agent plans the work, edits files, runs commands and checks its result, with you reviewing. Tools like Claude Code and Cursor both work this way.

How long should a spec be?

Short. One page is enough for a small feature. If it runs past two pages, split the feature in two. A spec the AI can hold in view is better than a long one it skims.

Can someone teach me this on my own project?

Yes. Vibe Coding Coaching is 1 on 1 with James Killick, working on your own app. It is $1,950, $2,750 or $5,000 a month for 1, 2 or 4 sessions.

Want a coach to check your first spec?

Vibe Coding Coaching is 1 on 1 with James, on your own app. Bring your idea. Leave with a spec and a plan you can build from.