Delivery Playbook · Quality & AI
Delivery with and without AI
How our development approach differs with AI, stage by stage, and what stays exactly the same.
Owner: CTO · Last reviewed: · Next review: 28 April 2027
We have always built software with experienced engineers, written requirements, reviewed code and tested before release. AI changes how much of that first draft work people do by hand. It does not change who is responsible for the result.
Side by side
| Stage | Without AI | With AI | What stays the same |
|---|---|---|---|
| Discover | People read notes and documents, ask clarifying questions in short sessions, and write requirements by hand. | AI summarizes notes, spots gaps and conflicts, and drafts questions and acceptance criteria for people to confirm. | The customer, Delivery Manager and designer agree the problem, the baseline and the outcome. |
| Design | The Tech Lead and engineers draft options, diagrams and documents themselves, and the team reviews them. | AI lays out options and trade-offs and drafts diagrams and documents as a starting point. | The Tech Lead decides and records the decision. The design gate still applies. |
| Build | Engineers write the code and the unit tests by hand, and peers review each change. | AI drafts code and tests in small steps. The engineer reads, understands and owns every change. | Every change is peer reviewed, with at least two approvers, one of them senior. The definition of done still applies. |
| Verify | QC writes and runs test cases, often by hand first, and automates repeated checks such as regression over time. | AI proposes test cases, data and edge cases, and more repeated checks are automated. We check that AI-written tests fail when the code is wrong. | QC verifies each function, the customer accepts the result, and a named person approves the release. |
| Operate | Engineers read logs and alerts and trace problems by hand. | Where agreed with the customer, AI correlates alerts and logs, suggests likely causes and drafts incident notes. | Engineers decide and act. Consequential production changes are approved by a person. |
What changes for you
- Early drafts of requirements, designs, code and tests arrive sooner, so people spend more time deciding and checking and less time producing first drafts
- Work moves in smaller steps with tests alongside, so problems show up earlier
- You are told which AI tools we use on your project, and which data they may touch, in the engagement's AI-use policy
- Speed gains are measured on your project against a baseline we agree with you before we report them
What does not change
- People stay accountable for every change that ships, whoever or whatever wrote it
- Every change is reviewed by an engineer who understands it
- The four quality gates, the definition of done and the release approval apply in the same way
- Customer code and data are handled only as the AI-use policy allows
- AI never approves, accepts or releases anything
Where AI helps less
AI helps most when the problem is clear and the work can be broken into small, checkable steps. It helps less when requirements are unclear, when the system is poorly understood, or when the decision is a judgment call. In those cases the people on the team still do the thinking, and we say so rather than count it as a gain.
Related
Quality & AI
AI engineering
How AI supports the whole engineering workflow, who is responsible for what, and the rules that keep it safe.
Quality & AI
Measuring our AI speed-up
How we measure whether AI really makes delivery faster, and what we publish with every number.
Quality & AI
Quality gates
Four gates a change passes between idea and production, and what each one checks.
