AI Driven Development
AI builds. People verify.
We call it ADD.
Why ADD
Now that AI writes the code, what is left for developers?
AI writing code is no longer an experiment.
Yet most organizations still fail to put it to work.
Two walls stand in the way.
-
Misconception
“AI will handle all of it.”
Setting direction and judging quality remain human work.
-
Doubt
“How can we trust code an AI wrote?”
Without answers on control and accountability, adoption stalls.
What is needed is not a new tool but a new division of work between AI and people.
Workflow
How the workflow changes
Before
With ADD
- 01Requirements =Unchanged Requirements
- 02Design →Changed Brief the AI
- 03Development →Changed AI generates code
- 04Testing →Changed Developer verifies and revises
- +Added Iterate with AI
- 05Release =Unchanged Release
What AI Does
What AI takes on
The more repetitive and structured the task, the faster AI handles it.
-
Requirements analysis
Document cleanup, feature breakdown
-
UI development
Component generation
-
API development
CRUD code generation
-
Database design
Draft ERD
-
Testing
Unit test generation
-
Documentation
README and API documentation
-
Code review
Improvement suggestions
-
Refactoring
Structural improvement
Accountability
Speed from AI, quality from people
ADD does not remove the developer. It raises the stakes of the role.
AI output does not ship as is. Judging what is correct, deciding what to keep and owning the result stay with people.
- Writing code
The developer had one job.
- Define the problem
- Brief the AI
- Verify the result
- Design the architecture
Design, judgment, verification and quality management carry far more weight.
QA Discipline
Built fast, trusted just as fast
Developing with AI speeds up the early stages.
That gain is often lost again at verification.
We hit that wall ourselves, and built a way through it.
Plenty of companies build fast with AI.
Few can make what they built trustworthy just as fast.
-
01
Verification by AI too
Every request that comes in is reconciled unattended overnight to confirm none was missed, and anything skipped is queued again automatically. The AI that builds and the AI that verifies are kept separate, so every result is cross-checked by a different model.
-
02
The bar for done is set in advance
What counts as done is defined as a gate before work starts. With one wording for verdicts and one owner for the record, “almost there” stops being an answer.
-
03
How far it was checked is on the record
Verification is recorded in three levels, static, specification and actual run. Results of an actual run are kept together with measured evidence such as logs, data or test output, so how far something was confirmed stays visible in the record.
-
04
Independent verification before conclusions
Critical calls are cross-checked through a second path before they are made. Irreversible work requires all three: independent verification, a confirmed rollback path and human approval.
-
05
Quality moves upstream
Automated checks run the moment code changes, then pre-commit rules and a pre-release quality gate filter in turn. Most issues are gone before QA ever sees them.
The know-how lives in automated rules, gates and checklists rather than in someone's instinct. It survives staff changes, and the mistakes we already paid for are not repeated on client projects.
Skills
Skills that matter
ADD calls for people who can direct and verify AI, not people who write the most code.
-
01
Requirements analysis
Defining clearly what needs to be built
-
02
Prompting
Giving AI instructions specific enough to leave no room for misreading
-
03
Architecture judgment
Deciding overall structure and technology choices
-
04
Code review
Verifying the quality and safety of AI-generated code
-
05
Verification design
Setting the bar for done up front, then automating the check
-
06
Domain knowledge
Connecting business requirements to technical decisions
In Practice
How we build
ADD is not a document at COMPUTERMATE. It is how we build.
We apply AI on top of 34 years on the manufacturing floor, and every solution we ship is built this way.
Our internal developer portal, byDP, runs the full ADD workflow in production,
including AI code review and AI quality management.
We use what we build before anyone else does.