PAN Lab
Levers
A lever is a choice an organization makes about how it uses its AI system. Some levers put a check between the system and the people who act on its output. Some limit what the system may write or send. Others change the people who use it: their training, their habits, and their time.
In the PAN Lab, each lever pushes on named parts of a network: the people using the system, the model, the records, and the pathways between them. You pull a lever and watch which way it pushes. Each lever below has a page of its own with what it is, what it pushes on, which pressures it answers, and where you can pull it.
Levers
Every lever the Lab offers
Open a lever to read what it pushes on, which pressures it answers, and which networks offer it.
Verify output
An independent check reviews what the autonomous agent produces before anyone acts on it.
Upgrade model
The organization upgrades the model so that it makes fewer errors.
Route more work through the assistant
More of the day's work goes through the assistant, so it carries more of the load.
Let it keep working records
The assistant keeps its own working record, so it can build on earlier tasks.
Review the riskiest first
People concentrate their review on the cases with the most at stake.
Mark AI-written records
The records show which entries a machine wrote, so readers can weigh them accordingly.
Gate record entries
Nothing enters the permanent record without a person's sign-off.
Keep prompts neutral
People ask the system neutral questions, so it is not nudged toward the answer they expect.
Pause AI on alarms
When an alarm fires, the system stops shaping decisions until a person clears a review.
Escalate checks
When monitoring flags trouble, people check more closely right away.
Keep skills sharp
People practice the work without the system, so their own judgment stays strong.
Understand the system
The organization pays for continuing study of what its AI deployment actually does.
Vet connections
Systems connect only where someone has explicitly authorized the connection.
Store less data
The organization writes less down and keeps it for less time.
Peer sharing rules
Rules govern what people and agents pass to one another, and peer review becomes routine.
Assign a challenger
Challenge becomes a scheduled duty, so a claim cannot rest on one colleague's confidence.
Check with a second model
A different second model checks the first model's output, so shared blind spots show up as disagreement.
Check copied records
The organization checks copied records against the original before anyone acts on them.
Review on schedule
The organization reviews the deployment on a fixed rhythm and retunes the controls at each review.
Gate vendor updates
The contract gives the organization notice, inspection, and exit rights over its vendor's model.
Require sign-off
Someone accountable must sign off before the system goes live, and before it stays live.
Know the tool
Everyone who uses the system learns what it is, what it cannot do, and where its output may not go.
Train verification
Staff learn to check automated output against its source, for errors of reasoning as well as of wording.
Incident loop
The organization logs what goes wrong, learns from it, and fixes the step that let it through.
Sort the queue first
The assistant takes a first pass at each request, so people reach the right desk sooner.
Keep the door open after hours
The service answers when the office is closed.
Train the staff
Everyone who touches the system learns its limits, and then drills the skill of checking its output.
A lever that puts a governance pattern into practice links to that pattern in the Practice Library. Each pressure the Lab can apply has a page too, and each of those pages lists the levers that answer it.
A lever is easiest to understand when you pull it and watch the network move.
Open the PAN Lab