First fixes with Codex

How can you entrust a code change to Codex and then check for yourself that the fix meets the need?

  • Lovable
  • Claude Code
  • Codex

1250 € per session, excl. VAT

In short

In a half-day session lasting 3 hours participants use Codex on a small existing programme: understanding a request, writing instructions, examining changes and running tests. They leave with a documented fix, evidence of its verification and a reusable method for limited coding tasks.

Why this training?

An overly vague request for a fix can produce plausible code while altering expected behaviour. Defining the files involved, making accepted inputs explicit and providing examples helps to guide the work requested from Codex. The value of the fix is measured in the programme as it runs, not in the confidence of the generated response. The training therefore combines instructions, human review of changes and reproducible tests, with particular attention to secrets and unnecessary changes. Running case study: The fictional organisation ‘Avelune Quincaillerie’ uses a script that calculates order totals from lines containing a reference, quantity and unit price. The script currently accepts negative quantities. Participants must correct this behaviour without changing valid calculations, using only fictional data. Transfer to the workplace: In the following week, select a minor bug in an authorised project, reuse the instruction template and compare tests before and after the fix. Have the change reviewed before any integration into a shared environment.

Objectives of this training

  • Describe a bug with a reproducible example and an expected result.
  • Write an instruction defining the scope of the change requested from Codex.
  • Examine the differences between the original and modified code.
  • Run tests for normal, boundary and invalid cases.
  • Document the fix, observed results and remaining limitations.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Beginner developers wishing to use Codex for small fixes.
  • Technical staff who maintain scripts and can read their code.

Prior skills

  • Have already used an AI assistant for a simple request.
  • Be able to read a function, recognise a condition and run a script in the provided environment.

What you will work on

  1. Define the task and access scope

    Objective: Prepare a verifiable request for a fix. Code generation and modification based on instructions; responsibility for human review; file scope, authorised data and secrets. Consultation of official documentation for usage arrangements.

  2. Write and refine the instruction

    Objective: Obtain a targeted proposal rather than a general rewrite. Minimal context, expected behaviour, constraints, example inputs and a request for an explanation of the changes.

  3. Review and test the fix

    Objective: Verify the actual behaviour of the proposed code. Reading differences; testing positive, zero and negative quantities; regression checking of the total; distinguishing between a suggested test and an executed test.

  4. Repeat the approach independently

    Objective: Address a second limited bug and document the checks. Rejecting a negative unit price; preserving valid calculations; change note and return to the original version.

What you take away

  • Bug report and instruction for Codex.
  • Corrected script with annotated differences.
  • Test set with results before and after modification.
  • Human review note and recovery procedure.

Method and assessment

  • Short explanations followed by immediate practice on the running case study.
  • Individual work followed by work in pairs, with peer review of outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Check access to Codex, authorised by the organisation, and prepare an isolated local practice environment. Provide the fictional script, its data and its execution commands; remove any secrets or real identifiers.

Observable criteria: The instruction specifies the expected behaviour, scope and excluded changes. ; A negative quantity is rejected according to the rule defined in the exercise. ; Valid cases retain their expected results after modification. ; Each test includes an input, an expected result and an observed result. ; The review explains the modified lines and checks that the submitted elements contain no secrets..

Frequently asked questions

Do you need to be proficient in programming?

You need to understand the provided script and know how to run it; no complete application development is required.

Are Lovable and Claude Code also used?

The exercises focus on Codex. Lovable and Claude Code are only positioned as other possible tools, depending on internal permissions.

Can we work on internal code?

The session uses the fictional case study. Subsequent transfer requires authorisation and a check of data, secrets and confidentiality rules.

Is the deliverable put into production?

No. The fix is run in the practice environment; any actual integration requires the organisation’s checks.

Last updated: