The Revision Right
A capability is not personal merely because it knows you. It is personal when you can change what it is allowed to know, do, and carry forward.
01 / the fixed-system habit
Most software trained us to accept its updates as weather.
App People were expected to adapt to the application; Stack People need a capability that can adapt to their changed judgment.
That difference sounds ordinary until a personal AI capability begins to shape a real decision. A fixed application may change its interface, revise its terms, or add a feature, but the user’s role is usually to learn the new arrangement. The product is the stable center. The person is the moving part.
App-to-stack work reverses the direction of adaptation. A person assembles prompts, sources, routines, permissions, examples, and review habits around the situations they actually face. Those parts do not remain neutral. They develop a point of view by repeating priorities, defaults, and old decisions. A stack can become useful because it remembers. It can become dangerous for the same reason.
The revision right is a Stack Person’s practical ability to inspect, change, or retire the parts of a personal AI capability that no longer deserve to shape their work.
This is not a legal claim, and it does not mean a person may override an employer’s security rules or carry private information into a new setting. It is a design and institutional expectation. If a capability is supposed to serve someone’s judgment, that person must have a meaningful way to revise it when their judgment changes.
02 / the revision room
A living stack needs three doors, not one permanent setting.
Revision is broader than editing a prompt. It is the ability to decide which inherited parts still belong in the next act of work.
See what is acting.
A person should be able to identify the sources, preferences, instructions, automations, and permissions that influence an outcome. Invisible defaults turn a tool into an authority nobody can question.
Move a boundary.
They should be able to alter a standing instruction, narrow a permission, add a review moment, or replace a stale source. Change makes correction operational rather than merely regretful.
Let an old answer end.
They should be able to remove a routine that no longer fits, revoke access that no longer belongs, and stop carrying a past decision into a changed relationship.
03 / a small test
Can the capability take a new instruction from its person?
Try the test below. It is deliberately modest: not whether a system can be rebuilt from scratch, but whether its owner can name a changed condition and have that change alter the work it produces. The answer separates a personal capability from a personalized enclosure.
Changed condition: “This source is no longer reliable for this decision.”
Revision: remove it, record the reason, and route the next answer through a different check.
Authority: the person who bears responsibility for the outcome can make the change visible.
04 / the organization’s part
Institutions can protect their context without freezing a person’s method.
Organizations have legitimate reasons to constrain data, systems, approvals, and records. A revision right is not a shortcut around those boundaries. In fact, it makes them clearer. The organization can specify which materials must remain inside its environment and what requires a formal review. The person can retain authority over the general habits that make a capability more careful: checking a premise, labeling a draft, escalating uncertainty, or declining to automate a sensitive handoff.
That distinction matters for management. Without it, every improvement a person makes becomes either invisible private labor or presumed organizational property. The first arrangement depends on unpaid maintenance. The second treats a worker’s developing judgment as a container to be owned. Neither is a durable answer to work that is becoming more AI-mediated.
A stronger arrangement recognizes bounded revision. Give people clear places to adjust their personal methods; make the boundary around protected context explicit; and decide in advance who may change a consequential workflow when new evidence arrives. That is less theatrical than claiming to have an AI strategy. It is more useful.