AI-editable websites

Natural-language change requests, inside a controlled delivery workflow.

For suitable systems, Tudouke can turn a request into a structured change, build it in isolation, validate the important behaviour, and publish only an accepted result.

The workflow

AI assists each stage. The contract keeps the stages aligned.

The purpose is not to let a model make unchecked production changes. It is to make change requests easier to express while retaining engineering and release discipline.

  1. 01

    Describe

    Explain the change in ordinary language, in English or Chinese.

  2. 02

    Confirm

    Clarify the intended user and business outcome before implementation.

  3. 03

    Contract

    Represent required capabilities, constraints, and acceptance expectations structurally.

  4. 04

    Build

    Create the candidate version in an isolated preview environment.

  5. 05

    Validate

    Check the important workflows and report what is covered or still unresolved.

  6. 06

    Release

    Publish through the project's accepted release process, with rollback boundaries where provided.

What remains controlled

A request is not the same as a release.

The process preserves distinctions that matter when software affects real customers and business records.

Intent and implementation
The requested outcome is retained separately from build diagnostics and technical evidence.
Preview and production
Candidate work is reviewed away from the live service before release.
Supported and unsupported capabilities
Unknown or unsafe capabilities are not silently converted into something easier.
New project and revision
A new system and a change to an accepted system keep different operation states.

Safeguards

Designed to fail visibly at important boundaries.

A useful workflow must preserve required capability, protect sensitive material, and avoid publishing when coverage is not proven.

  • Structured requirements remain the authority for what the build must support.
  • Raw destinations, credentials, tokens, and provider authority stay protected.
  • Required downstream references can remain as opaque, non-sensitive identifiers.
  • Validation checks real routes, forms, records, roles, and user-visible outcomes where applicable.
  • Uncertain capabilities remain required until execution or validation proves coverage.
  • Release controls depend on the project; AI assistance does not imply unrestricted deployment access.

Project fit

The right level of automation depends on the system.

Start with the change you need. Tudouke can assess whether a controlled AI-assisted workflow is appropriate, or whether a conventional improvement is the better answer.

Discuss a suitable workflow See the redesign preview