Verifiable code delegation

How can you entrust a task to Codex while retaining control over scope, tests and validation?

  • Lovable
  • Claude Code
  • Codex

1960 € per session, excl. VAT

In short

This seven-hour day, excluding breaks, establishes a process for delegating development to Codex: selecting a task, preparing its context, formulating acceptance criteria, reviewing changes and checking tests. Each participant compiles a delivery dossier suitable for review, without assuming deployment to production.

Why this training?

An overly broad request for a fix can lead to unnecessary changes, conceal a functional assumption or move a defect elsewhere. Defining the files involved, exclusions and expected behaviours makes delegation to Codex controllable. Generated code is not proof of resolution. The day distinguishes between a proposal, test execution and a human decision, in order to measure the result's compliance rather than judge only its presentation. Running case study: The fictional organisation Avelune Location uses an equipment booking application. A defect allows a return date earlier than the departure date. Using a training repository containing only fictional data, participants delegate the fix, its tests and its documentation, without changing the pricing rules. Transfer to the workplace: On returning to work, choose a limited fix in an authorised repository, reuse the task contract and have a peer review the changes. Compare criteria met, regressions detected and rework needed before expanding delegation.

Objectives of this training

  • Break down a functional request into bounded, verifiable tasks.
  • Write instructions that include context, constraints and exclusions.
  • Have Codex generate or modify code for an authorised case.
  • Check the differences and run the relevant tests.
  • Document deviations, rework and the validation decision.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Developers wishing to delegate fixes and limited enhancements.
  • Technical leads and quality contacts responsible for reviewing assisted code.

Prior skills

  • Be able to read and modify the language used in the training repository.
  • Have practical experience of version control, code review and automated testing.
  • Have already used an AI assistant with structured instructions.

What you will work on

  1. Choose the task to delegate

    Objective: Define the boundaries of the booking fix. Human responsibility, dependencies, exclusions and risks of cross-cutting changes.

  2. Prepare a safe context

    Objective: Provide only useful context. Repository structure, conventions, existing tests and secrets management.

  3. Write the delegation contract

    Objective: Turn the defect into verifiable instructions. Expected behaviour, edge cases, constraints and requests for clarification.

  4. Steer the changes

    Objective: Obtain a proposal limited to what is needed. Code generation and modification; targeted iterations rather than a complete rewrite.

  5. Verify independently

    Objective: Establish evidence of compliance. Review of differences, relevance of assertions and search for regressions.

  6. Decide and hand over

    Objective: Deliver a dossier that a reviewer can use. Acceptance, rejection, known limitations and observed quality indicators.

What you take away

  • Task contract with acceptance criteria and exclusions.
  • Set of code changes and associated tests.
  • Completed review checklist and record of test results.
  • Delivery note with limitations and the human decision.

Method and assessment

  • Short inputs followed by immediate practice on the running case study.
  • Individual work followed by work in pairs, with participants reviewing each other's outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Check access to Codex as authorised by the organisation and consult the official documentation for the environment used. Prepare the training repository, run its initial tests and remove secrets, identifiers and sensitive data.

Observable criteria: Each requirement is linked to a test or an explicit check. ; Changes remain within the defined scope; each exception is justified. ; Tests are actually run and their results retained, including failures. ; No secrets or sensitive data are transmitted in the instructions. ; The final decision distinguishes between validated fixes, open deviations and limitations..

Frequently asked questions

Do you need to use all three tools?

No. Codex is central; Lovable and Claude Code are only positioned as other possible coding assistance tools, depending on permissions.

Can you bring a work repository?

Only an anonymised and authorised version, without secrets. The fictional repository remains available.

Are the capabilities identical across environments?

Do not assume so: consult the official documentation for exact access arrangements, settings and procedures.

Will the result be deployed to production?

No. The deliverable is assessed in a training environment; any real integration follows the organisation's validation procedures.

Last updated: