AI coding practice
Vibe coding,
done properly.
A practical playbook for AI-assisted software delivery. We use AI to write code faster, inside a controlled engineering process where people stay accountable for the result.
Four phases
For every meaningful change.
01
Plan
Agree on the outcome and the limits before any code is written.
Leaves behindA short brief and build plan
02
Constrain
Give the AI our project rules, approved tools and reusable patterns.
Leaves behindRules and reusable patterns
03
Iterate
Build one small, testable piece at a time, then check and save it.
Leaves behindA working change and handoff
04
Verify
A person reviews the change and the test results before release.
Leaves behindAn accountable release decision
The principle
AI can write the code.
The team owns the outcome.
Generated code is a draft until evidence supports release. We treat every AI-assisted change like any other change: scoped, reviewed, tested and signed off by a person.
$ dcp change --plan "add maker-checker to payouts"
brief ........ ok
rules ........ loaded (12 patterns, 3 approved tools)
iterate ...... 4 slices, 38 tests passing
verify ....... awaiting human review
▶ release decision: reviewer approved · evidence attached
Illustration only. Not a real tool output.
Five team habits
That reduce rework.
Save progress
Commit working changes before the next big edit.
Check packages
Confirm identity, upkeep, security and license.
Protect data
Approved tools, test data, no secrets in prompts.
Review by risk
Extra checks for money, identity or sensitive data.
Measure results
Track delivery time, review effort and defects.
Let's talk
Adopt AI delivery with guardrails.
We can set up the practice inside your engineering teams: rules, tooling, review gates and measurement.
