AI / MOVE FAST
Gather, compare, and verify.
Assign work that benefits from scale and speed to AI.
- Research and organize information
- Compare options and draft
- Review changes and test
- Prepare approval materials
05 / AI Product Building
Let AI move quickly;people keep finalresponsibility.
What AI can do is not the same as what it should be allowed to do. Parallelize the work and assemble decision evidence automatically, while pausing irreversible actions for human review.
Explore the operating model ↓AI / EXECUTIONDecompose → Context → Build → Verify
HUMAN / GATEReview the impact and decide whether to publish
01 / Operating model
Divide the work without dividing responsibility.
Keep repeatable tasks separate from decisions. Let AI prepare the inputs; people remain accountable for the outcome and its impact.
02 / Build flow
Structure the work before introducing AI.
Do not start with prompts. Separate the work, tasks, decisions, and exceptions, then align the source of truth and completion criteria before execution.
01
Decompose
Separate work, tasks, decisions, and exceptions
02
Context
Provide source data, inputs, and prior decisions
03
Constraints
Define scope, prohibitions, and completion criteria
04
Execute
Build in small steps and keep a record of changes
05
Verify
Check rendering, behavior, data, and exceptions
06
Learn
Feed failures back into the next rules
03 / Command center
Manage work by state, not by conversation.
Bring projects, current status, next tasks, blockers, pending approvals, and completion evidence into one operations view.
Exceptions that need human review
RUNWork AI can continue→
HOLDPause until criteria are met‖
ASKHuman approval required!
DONEComplete with recorded evidence✓
CASE 04 / AI COMMAND CENTER
Focus on projects that need a decision instead of watching every task.
An “In progress” label is not enough to resume work. Keep the purpose, attempts, next step, blocker, and approval scope in a project record.
View the case in Works →Read the implementation notes ↗
04 / Approval gate
Pause for approval before the final action.
Rather than stopping AI early, let it assemble the information needed for a decision, then return control to a person.
HUMAN GATE
Capability does not equal permission.
Let AI proceed
- Gather and organize information
- Prepare comparisons and drafts
- Test and review changes
- Prepare approval materials
People decide
- Publishing, posting, and external messages
- Purchases, contracts, and financial actions
- Sending personal or confidential information
- Changes to important rules
05 / Human review
Design the review process as AI scales up.
People reach their review capacity before multiple AIs reach their generation capacity. Distribute reviews throughout the work instead of waiting until everything is finished.
Do not leave all reviews until the end.
Break reviews into direction, structure, implementation, and rendering; check each while changes are still inexpensive.
Scale review depth to the deliverable.
Copy, UI, data, and external messages each require a different level of scrutiny.
Collect completion notices in one place.
Start from the review queue instead of searching through conversations.
Feed failures back into the rules.
Turn each correction into a constraint that future projects can follow from the start.
06 / Current status
What is working and what remains unverified.
Do not extrapolate results from personal use to enterprise adoption or productivity gains.
Used continuously on my own projects.
Completion criteria are checked separately for each task.
Irreversible actions are never approved automatically.
Not presented as proven results at this stage.
Next / Thinking