A software delivery pipeline for controlled AI-assisted development.
The pipeline structures agreed work from planning and implementation through review and documentation; it can also audit the codebase as a whole. Team-defined goals and rules, separate maker-checker roles and explicit checks make each step reviewable.
How the software delivery pipeline works
Before project work begins, the team defines the objective and definition of done, architecture and engineering rules, and domain and quality requirements. With AI support, a developer turns this frame into an epic draft with constraints, dependencies and verifiable outcomes.
Planning, implementation and documentation reconciliation follow in sequence, each with separate maker and checker roles. The maker creates or revises the work; the checker reviews it against the checks defined for that stage. Findings are remediated or dismissed with a recorded reason, after which the work is checked again.
Throughout the process, the pipeline records its state, confines writes to defined paths and can resume after an interruption. Nuitio uses the pipeline to deliver software projects. Its core principles are transferable; the concrete implementation can be adapted to the repository, language, domain and task.
Software Delivery Pipeline
Nuitio in-house systemBefore work begins, the project team defines:
- Objective and definition of done
- Architecture and engineering rules
- Domain and quality requirements
Define the work
With AI, a developer turns the agreed frame into an epic draft for planning. It captures the constraints and verifiable outcomes.
Planning
Breaks the epic draft into dependent sprints and tasks, drafts the plan and revises it in response to findings.
Checks each version of the plan against the configured planning checks. When all checks pass, the plan moves to implementation.
- Scope, dependencies and sequence
- Architecture, interfaces and verification
Implementation
Implements the reviewed plan in code and tests, then remediates findings or dismisses them with a recorded reason.
Checks the code after each revision using tool gates and contextual checks. When all checks pass, the flow moves to documentation.
- Code, runtime behaviour and configuration
- Domain rules and tests
Documentation
Updates the architecture and repository documentation and resolves discrepancies with the code.
Independently checks the documentation against the code. When they agree, the repository is ready for handoff.
- Architecture and repository documentation
- Documentation matches the code
Repository
The reviewed repository is handed over to the team under the agreed controls.
Repository audit
A repository audit applies the same controlled model to the existing codebase as a whole. The team defines the audit scope, applicable checks and definition of done. The checker then applies the selected checks; the maker remediates findings or dismisses them with a recorded reason. The review repeats until the definition of done is met. Documentation is reconciled with the code at the end.
Repository audit flow
Nuitio in-house systemSet the audit frame
Before the audit begins, the audit team defines:
- Audit scope
- Applicable checks, including many used during implementation
- Definition of done
Define the audit
With AI support, a developer refines the agreed audit frame and prepares the repository-wide review.
Review the repository
Applies the selected checks to the audit scope. After each revision, the repository is checked again until the definition of done is met.
- Code, runtime behaviour and configuration
- Domain rules and tests
- Architecture and repository hygiene
Remediates each finding or dismisses it with a recorded reason.
Documentation
Updates the architecture and repository documentation and resolves discrepancies with the code.
Independently checks the documentation against the code. When they agree, the repository is ready for handoff.
- Architecture and repository documentation
- Documentation matches the code
Repository
The reviewed repository is handed over to the team under the agreed controls.
- Project context
- Nuitio in-house system
Talk to Tobias about AI in your software delivery
Send Tobias a short note about what you have in mind. In an initial conversation, we discuss the project and possible next steps.
PhD, ETH Zürich · Financial Risk Manager · IMD Future Leaders