Show Claude the Expected Output and Let It Write the Code That Produces It

Sometimes the clearest way to specify what you want isn't describing the logic — it's showing the result. Paste the expected output into Claude and it reverse-engineers the code that produces it.

> write an API endpoint that returns this exact JSON structure:
>
> {
>   "user": {
>     "id": 42,
>     "name": "Jane Smith",
>     "plan": "pro",
>     "usage": {
>       "api_calls": 1547,
>       "storage_mb": 234,
>       "limit_percentage": 61
>     },
>     "recent_projects": [
>       {"id": 1, "name": "Website Redesign", "updated_at": "2h ago"}
>     ]
>   }
> }

Claude reads the structure and writes the controller, query, resource/transformer, and any computed fields — matching the exact shape, nesting, and field names you specified. No ambiguity about what "usage data" means or how "recent" is defined, because the example makes it concrete.

This works for any kind of output:

> write a CLI command that produces this table:
>
> ┌──────────┬───────┬──────────┐
> │ Route    │ Method│ Middleware│
> │ /users   │ GET   │ auth     │
> │ /users   │ POST  │ auth,admin│
> └──────────┴───────┴──────────┘

> write an email template that renders like this:
> [paste the HTML or screenshot of the desired email]

> write a migration and model that lets me query data shaped like this:
> [paste the desired query result]

This approach is especially powerful for complex data transformations where describing the rules in English would be longer and less precise than just showing input and output:

> here's the raw data from the API:
> [paste raw response]
>
> transform it into this format for the frontend:
> [paste desired shape]

Claude reads both, figures out the mapping, and writes the transformation code — renaming fields, flattening nested structures, computing derived values, and formatting dates.

When the output is clear, the code writes itself — show Claude the destination and it builds the path.

via Claude Code