A Return to Code

A return to code is a cheap-custom-software loop that turns a clear ask into a one-afternoon program the market would never have funded. You describe the program, an agent builds and tests it, you correct it, and the keepable result can serve one person for one afternoon.

What got cheaper

Getting a traditional program to first-run used to cost a pile of ceremony: pick a framework, wire dependencies, build the interface, host it, then chase the errors. Agents flatten that pile. They walk a repository, run shell commands, read the failure, and keep editing — if the surrounding stack is Agent-Native Infrastructure, surfaces an agent can read and act on without a human click-path.

The loop that comes back is playful and direct. A wanted behavior becomes a file you can execute. That fast creative loop is Vibe Coding. Agentic Engineering is the professional quality system wrapped around it, and the two should stay separate concepts.

What the person still owns is narrower and harder: name the behavior you actually want; hold taste and product judgment; catch the agent’s wrong assumptions; decide when to stop; tell the difference between “good enough to keep” and “this needs real engineering.” Wanting a clear thing is the scarce skill. Cheap implementation is not permission to keep adding.

Why code, and where it fails

Think of a personal app store the way you think of a folder of spreadsheets — a shelf, not a company. You stop waiting for a startup to ship the exact workflow and you keep a rough tool of your own. A one-shot app is that tool when it serves one person, one job, or one afternoon. Disposable still counts. It does not have to grow into a product or a codebase you maintain. Some programs are jigs: a script you throw away after the job, the way a spreadsheet is thrown away after the quarter.

Code is the domain where a model cannot hide. Prose can sound finished while being wrong. A program either runs, fails a test, writes a log, or shows the user a broken screen.

Execution is a verdict. A one-shot you cannot run, cannot test, and cannot discard is not a return to code.

Models are not interchangeable. Some plan well, some edit well, some are just fast. Matching the model to the job, and stuffing the window with enough of the right material, is Context Engineering.

Knowledge work has the same economics. This wiki can grow small tools as they earn their keep — a markdown search command, a check that the source index is sound, a health pass for orphans and missing backlinks, a listing of still-uncompiled intake, a generator for ingest prompts. Build one only when it cuts real maintenance. That is the stop decision, applied here.

The same cheapness that makes the loop worth running is what breaks it:

• A jig dressed up as production software.
• Scope that grows because the next feature feels free.
• Taste that drowns in a stream of generated extras.
• A wide edit accepted with no test and no review.
• An agent trusted as if it could see the whole project, when the window still cannot.

A return to code is only a return when the thing runs and the operator can still stop. Cheap custom software that never executes, or that grows because growth is now easy, is just more untested text.

Related

• Vibe Coding — the fast creative loop this page’s economics sit under.
• Agentic Engineering — the professional quality system around that loop; the two stay separate.
• Software 3.0 — the broader frame: software as English plus models.
• Agent-Native Infrastructure — what the surrounding stack has to look like for agents to run commands and edit files.
• Context Engineering — giving the agent enough context to match model to task.
• LLM Tool Use — operator craft for calling tools from a model.

Sources

• Naval Ravikant and Nivi, A Return to Code, 2026-04-29.
• The phrase “vibe coding” was popularised in 2025 (Andrej Karpathy).