The working method
How We Work with AI
AI is useful here because it can help explore options, build reviewable changes and catch problems. The question, direction, judgement and responsibility remain human.
Working Rule
Use AI to increase the quality of your thinking, then inspect what it produces before you rely on it.
Five checkpoints
A Simple Collaboration Loop
This is the public-facing method behind the project. It is intentionally understandable without exposing private records, prompts or internal systems.
- Human context comes first.
Start with the real goal, constraints and what you currently believe.
- AI expands the options.
Ask for alternatives, challenges, drafts or a bounded implementation that can be reviewed.
- Important claims are checked.
Use primary or reliable sources when accuracy, safety or cost matters.
- A human reviews and decides.
Test the result, correct weak assumptions and choose what should move forward.
- The learning is recorded.
Keep the useful decision, correction or milestone so the next step begins with better context.
Team Human · interactive map
Choose What Stays Human
AI is most useful when its role matches the purpose. Choose a goal to see one possible division of work. This is a planning aid, not a score or an automated decision.
Mode 01 · build capability
Learn or Practise
Use AI to create useful friction, feedback and another explanation without handing away the ability you meant to develop.
- 01
Human purpose
Name the skill you want to keep. Make your own first attempt before asking for help.
- 02
Source and context gate
Bring the teaching material, your attempt and the level of help that still leaves useful practice.
- 03
Bounded AI support
- Questioner
- Coach
- Critic
Ask for questions, hints, counterexamples or a different explanation rather than an unexplained final answer.
- 04
Human return
Close the answer and explain or perform it yourself. If you cannot, retry with less assistance.
Keep in view: a polished answer is not the same as retaining the ability you intended to practise.
Mode 02 · save effort
Complete Routine Work
Delegate more when the goal is completion, while keeping the original, the acceptance check and a usable fallback under human control.
- 01
Human purpose
Define the result, constraints, privacy boundary and what “done” will look like.
- 02
Source and context gate
Use the current input, share only what is needed and protect an unchanged original.
- 03
Bounded AI support
- Worker
- Checker
Draft, organise, compare or automate a bounded step that can still be inspected and reversed.
- 04
Human return
Sample the work, test the important edge cases, then accept, revise, reject or undo it.
Keep in view: speed counts only after the output passes its check and a fallback still works.
Mode 03 · consequences matter
Make an Important Decision
Use AI to widen the view and stress-test assumptions, not to hide uncertainty or transfer responsibility for a consequential choice.
- 01
Human purpose
Name who is affected, what matters and what only an accountable person can decide.
- 02
Source and context gate
Use suitable current sources, protect sensitive context and keep unknowns visible.
- 03
Bounded AI support
- Research aide
- Counterargument
- Risk check
Compare options, surface assumptions and identify what still needs direct evidence or qualified help.
- 04
Human return
Open the sources, verify what matters, seek qualified help where needed, then decide or stop.
Keep in view: AI can widen the options, but it cannot accept responsibility for the consequences.
Do not count agreement as proof. Different AI roles are review passes, not independent witnesses. Important claims still need suitable original or authoritative sources.
Clear roles
What Stays Human and What AI Supports
The line is not perfect, but making it visible helps each contribution stay understandable.
Human contribution
Direction, Judgement and Accountability
- Choosing the goal and why it matters
- Providing lived context and preferences
- Reviewing wording, design and important claims
- Testing what reaches the live website
- Making the final call and owning the result
AI contribution
Exploration, Implementation and Checking
- Generating options and challenging assumptions
- Drafting reviewable text and code
- Running bounded checks and finding defects
- Organising evidence for human review
- Supporting repetitive work without owning the decision
Practical boundary
AI Output Is a Starting Point, Not Proof
Good collaboration includes disagreement, corrections and uncertainty. If the evidence is weak, the honest label is unknown, unverified or planned.