Skip to main content
Jig helps you build software with an AI coding agent, whether you are new to programming or an experienced developer. Start from an idea or improve an existing project, one manageable task at a time. You explain the result you want. The agent plans the work, writes the code, checks it, and keeps notes that future sessions can use. You try the result and decide what should change.
You need an AI coding app first, such as Claude Code or Codex, with your project folder open. Jig adds a way of working to that app. It does not include an AI model or replace your AI account.

Built for your level of experience

New to programming?

Describe your idea in everyday language. Follow the setup guide, let the agent explain the choices, and try what it builds. You do not need to know commands or design the code yourself.

Experienced developer?

Give your agent a repeatable workflow with project rules, reviews, and checks. Inspect the changes, run commands yourself, and choose how much work to delegate.
Both paths use the same framework. Start a new project or bring Jig into an existing codebase.

Spend tokens on the task, not the whole handbook

AI reads text in small pieces called tokens. Loading a project’s entire handbook into every task uses tokens even for pages the task does not need. Jig organizes project knowledge so the agent starts with the core rules and the documents relevant to its work. Other useful material appears in a short catalog and can be opened when needed. For a booking change, that might mean booking rules without reading an unrelated reporting guide.
Less unnecessary reading, more room for the work. Selecting relevant documents can save context space and input tokens compared with loading all project knowledge. The amount depends on the project and task; Jig does not measure or guarantee a token reduction.
See how Jig builds the agent’s context →

Your first conversation

Once Jig is set up, paste this into your AI chat and replace the example with your idea:
The agent asks about your users and what they need, helps you choose a small first version, and writes down the plan. When you agree on it, ask it to start the first task. Follow the complete example →
You can say “I don’t know — explain the options and recommend one.” You bring the purpose and the feedback; you do not have to arrive with a technical design.

What working together looks like

1

Describe a result

“Let visitors reserve a place in a class.” A concrete example gives the agent something to check.
2

Settle the important choices

The agent looks at what the change affects. Bigger or riskier work needs a design for you to approve before it starts building, unless you have explicitly enabled unattended runs.
3

Try what it built

Ask how to open the result, what was tested, and what is still missing. Give feedback in the same conversation; the agent can continue the task.
A finished task is not automatically a live website. Saving the work, merging it into the project, and publishing it are separate steps. Ask your agent what remains before people can use it. A team that tests on a staging server first can have each task wait there until it passes (Test on staging first).

Learn what you need, when you need it

From idea to first version

Example messages, the decisions you make, and how to ask for the first working result.

Day-to-day work

Ask for changes, continue tomorrow, check progress, and understand what needs you.

Let the agent do more

Choose how far the agent may take finished work. Start with the defaults and change them later.

For developers

Run commands yourself, inspect task state, and understand the files and checks behind the workflow.