The human role
Intent, boundaries, judgment, responsibility.
The human decides what matters, what is acceptable, and what the result must protect.
Human + AI
Humans define the goal. AI explores the space.
Working with AI is not handing over the work. It is designing the conditions under which useful work can happen.
The AI role
Exploration, generation, comparison, speed.
AI expands the number of possibilities we can inspect, but it does not decide which future is worth creating.
Division of responsibility
Better output starts with a better frame.
H1
Define the goal
State what should change, for whom, and why it matters. A useful goal is an outcome, not a feature request.
H2
Set the constraints
Name the limits that protect quality and trust: budget, time, accessibility, performance, privacy, legal risk, brand, and technical reality.
H3
Choose the trade-offs
Decide what can bend and what cannot. Faster is not always better. More options are not always more useful.
H4
Own the decision
Review the evidence, make the call, and remain accountable for what reaches people. AI can advise, but it cannot carry responsibility.
A1
Explore the space
Generate many plausible routes, edge cases, questions, and alternatives more quickly than a team could inspect manually.
A2
Make and compare
Draft copy, interfaces, code, tests, and scenarios, then compare them against the agreed criteria.
A3
Surface consequences
Reveal conflicts, missing information, accessibility gaps, technical risks, and second-order effects before they become expensive.
A4
Accelerate iteration
Turn feedback into another version quickly, while preserving the human-defined goal and boundaries.
Constraints are a design material
A strong AI brief says more than “make this better.”
- Who is this for, and who could be excluded?
- What outcome would count as success?
- What must never happen?
- What evidence is required before shipping?
- Which decisions need explicit human approval?
- What happens when the system is uncertain or wrong?
What this changes
Prompts become briefs
The important work happens before generation: defining the problem, criteria, risks, and non-negotiables.
Outputs become proposals
AI output is material to inspect, test, combine, reject, or improve. It is not automatically the answer.
Automation gets boundaries
The more autonomy a system receives, the clearer its permissions, escalation rules, and stop conditions must become.
AI can make almost anything. The human task is deciding what should exist.