Making assisted delivery reliable
How can the delivery of an application developed with AI assistance be made verifiable and reproducible?
- Lovable
- Claude Code
- Codex
3500 € per session, excl. VAT
In short
Over two days, totalling 14 hours excluding breaks, design a prototype delivery pipeline for an application developed with AI assistance. You prepare the repository, control generated changes, run tests and simulate a reversible delivery. The result forms a pilot dossier, with no commitment to production deployment.
Why this training?
A working prototype is not enough for regular delivery: changes that are difficult to review, implicit dependencies and incomplete tests make regressions difficult to diagnose. AI assistance speeds up code generation and modification, but replaces neither human review nor evidence of execution. The training organises these checks before any pilot decision. Running case study: The fictional organisation Avelune Rechanges is preparing an internal parts request application. A prototype exists, but assisted changes sometimes break quantity validation. The team must prepare a traceable, tested and reversible pilot delivery using fictional data. Transfer to work: Have the dossier reviewed by technical and security leads, then propose a limited pilot: an authorised environment, a validation owner, quality thresholds, monitoring and explicit stop conditions.
Objectives of this training
- Define the stages and responsibilities of a pilot delivery.
- Structure a repository and verifiable change instructions.
- Build tests covering critical user journeys.
- Run a reproducible verification pipeline.
- Simulate delivery and a rollback to an earlier version.
- Measure quality and justify a pilot decision.
Who is it for and what are the prerequisites?
Roles and tasks concerned
- Developers already using AI assistance to modify code.
- Technical leads and delivery contacts overseeing application prototypes.
Prior skills
- Be able to read, modify and run an application with its dependencies.
- Have practical experience of version control, branches and change review.
- Have already used AI assistance for coding.
What you will work on
Scope the pilot
Objective: Set the delivery conditions. Scope, responsibilities, blocking thresholds.
Diagnose the prototype
Objective: Identify weaknesses. Architecture, dependencies, configuration, data.
Organise the repository
Objective: Make changes traceable. Branches, versions, reproducible instructions.
Control assisted changes
Objective: Limit uncontrolled changes. Targeted instructions, code differences, human review.
Secure the configuration
Objective: Separate code and secrets. Variables, least privilege, sanitised logs.
Establish critical tests
Objective: Detect a regression in business functionality. Unit tests, integration, edge cases.
Assemble the checks
Objective: Automate the selected checks. Installation, analysis, tests, stopping on failure.
Put the checks to the test
Objective: Verify their ability to block. Injected regression, diagnosis, rerun.
Prepare a release
Objective: Identify a reproducible deliverable. Version, dependencies, configuration, release notes.
Simulate delivery and rollback
Objective: Demonstrate reversibility. Post-installation checks, data compatibility.
Measure quality
Objective: Inform the decision. Passed tests, open issues, duration of checks.
Defend the pilot
Objective: Make the case for conditional authorisation. Residual risks, owners, monitoring, stopping.
What you take away
- Prototype repository and execution instructions
- Verification pipeline configuration and test reports
- Review checklist and risk register
- Simulated delivery and rollback procedure
- Pilot decision dossier and quality dashboard
Method and assessment
- Short inputs 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 the tools authorised by the organisation from among Lovable, Claude Code and Codex; consult their official documentation for available capabilities and terms of use. Prepare an isolated environment, a training repository and fictional or anonymised data; exclude real secrets.
Observable criteria: A second run reproduces the checks from the repository instructions. ; Each critical user journey has an executed test and a retained result. ; A generated change is reviewed and linked to a requirement. ; No real secrets appear in the repository, instructions or logs. ; Rollback is demonstrated and the pilot decision cites the remaining issues.
Frequently asked questions
Is an actual production deployment planned?
No, only a prototype and an isolated simulation.
Must all three tools be used?
No. The choice depends on authorised access and the technical context.
Can you bring internal code?
Only with authorisation, without secrets and with fictional or anonymised data.
Do tests replace human review?
No. They provide partial evidence, supplemented by review and risk analysis.
Last updated: