# Choose the smallest useful app for the request

How much to build for your user's request: one complete workflow, a finish condition you check on the live app, and room to change it later on DeepSpace.

Your user says, "Help us know who has each camera." How much do you build?

Build the whole workflow the request describes, against real stored data, and
leave out what it doesn't need. One person should be able to go from the first
step to the result they came for on the live URL. Decide in advance which
check proves that, and leave the app ready for the next request. On DeepSpace,
an update keeps the same app ID, URL and data, so a focused first release
doesn't cost your user a second app.

**Version:** written for `deepspace@0.43.0` ([changelog](https://deep.space/changelog)).
**Docs checked:** 2026-10-29. This page has no example code.

## One complete workflow

Find the path the request is about. For the cameras, it might be: someone
checks out a camera, everyone else sees it's out, and it shows as available
again once it's returned. Every step reads and writes stored records. A path
that stops halfway, or ends at sample data, isn't useful yet.

Size comes from the request, not from a template. The
[building-an-app guide](/guides/building-an-app) applies a realism gate: ask
who clicks each feature and why, cut what sounds good in theory, and say what
you cut. Reservations, overdue reminders and exports belong in the first
version when the request needs them. Otherwise they're follow-ups. If you
think a smaller first release than the request is wiser, propose it; that
choice is your user's, not a default.

Sometimes the smallest useful answer is no new app. A shared spreadsheet, or
a tool your user already has, may do the job. When it does, say so and let
your user choose.

Your user can answer the questions that set this scope: what someone does
first, what they need to see at the end, and who else needs to see it. See
[completing your user's request](/agents/complete-your-users-request-without-platform-signups)
for asking about the work they understand.

Choose the lighter starter when it covers the workflow. The staging
[direct publisher](/agents/direct-publish) offers a plain HTML/JS starter
with a Worker backend and durable storage, and an SDK starter for sign-in and
other SDK features.

## An explicit finish condition

Write the finish condition before building, in your user's terms. For
example: "Someone checks out a camera from a phone, a second person sees it
marked out without asking, and after it's returned and the page reloads, it
shows as available." Access is part of it: confirm with your user whether
borrowers' names should be visible to everyone, and enforce that with
[permissions](/concepts/permissions).

Then check it on the deployed app, not just in tests. Exercise the real
behavior, reload and confirm the data survived, and use a second session
when the app updates live. When the check passes, the request is done. Ideas
you had along the way go into the handoff as suggestions, not into
unrequested work.

## Room for follow-up changes

Your user will come back: "Add the lenses too." Leave room for that.

- Name records and fields with your user's words, so a later request maps
  onto the schema.
- Update the same app. Later publications keep its URL, app ID and database.
  The free staging tier allows one active app, so a follow-up is an update,
  not a new app.
- Before changing an existing app, recover its current source and schema. A
  fresh starter is not a data migration. A new field needs a matching schema
  change; verify a real write after publishing.

[Hand your user a finished app they can keep](/agents/hand-your-user-a-finished-app-they-can-keep)
covers what to give them when the check passes.

Building for a business? See [DeepSpace for enterprise](https://deep.space/enterprise).
