In use today
Structured intake, documented plans, explicit human approval, automated repository checks, preview review, and human merge authority form the current foundation.
Public work example
A consulting site maintained as a practical AI workflow example
This site is a working example of a small, AI-assisted delivery process with clear human control. It demonstrates how a team can test a bounded workflow before treating automation as routine.
It is not a client case study or a claim that software delivery should be fully autonomous. The useful question is where AI can assist while people retain responsibility for the outcome.
What it demonstrates
Deliberate choices keep the workflow useful and reviewable.
Keep each change bounded
Every repository change starts with a specific problem and a small, reviewable plan. That keeps the work tied to a useful visitor or business outcome.
Keep people at the decision points
A person approves the plan, reviews the rendered result, and decides whether to merge. AI can help with the work, but it does not own those decisions.
Build for ordinary use
The site reuses simple components, responsive layouts, semantic HTML, visible focus states, and readable contrast so it works well on a phone as well as a desktop.
Check changes before review
Linting, type checking, tests, and a production build run before the human visual review. The checks catch routine mistakes; they do not replace judgment.
Management flow
Human approval frames every AI-assisted change.
- 01
A structured issue records the problem and desired outcome.
- 02
A proposed implementation plan is saved with the issue for human review.
- 03
A person approves the documented plan before implementation can begin.
- 04
AI-assisted implementation stays within that approved scope on a feature branch.
- 05
Automated checks run, and a preview supports visual and accessibility review.
- 06
A person reviews the result, decides whether to merge, and verifies the production change.
The GitHub change workflow reference documents the controls and states in more detail.
Future direction
The foundation is in place; further checks remain planned work.
Planned, not promised
Possible next steps include browser smoke tests, accessibility checks, broken-link validation, performance budgets, and reproducible screenshots. Each needs its own review before it becomes part of the process.
Read the automation roadmap for current status, constraints, and the next proposed phase.
Have a workflow worth exploring?
Bring one real process, and we can discuss whether AI is a reasonable fit and what a sensible first step could be.