How it works
One line of reasoning — visibility, reliance, change, impact, action — built on four distinctions that Heisterlux is being built to keep explicit.
The line of reasoning
Visibility
Start with an honest picture: which AI systems and AI-enabled services are used, in which applications, and through which Deployment Context. In the current scope this relies on information people in the organization record. Automatic discovery is not part of it and is not promised.
Reliance
For each use, record what the organization depends on it for, how critical that is, and the reason the reliance was considered justified — for example that a person reviews every output.
Change
Changes can happen at any layer: a model is retired, a licence moves to another plan, an application switches on a feature, or the use itself drifts.
Impact
Because the distinctions stay explicit, a change can be traced to the recorded uses it may affect. A dependency path or a model change does not by itself show that a decision is affected or no longer valid. Where evidence is missing or incomplete, that is shown as uncertainty and left to human judgement.
Action
The response should fit the change and the stakes, from noting that a change was seen to reviewing a critical reliance before continuing. Heisterlux suggests; people decide.
The four distinctions
- AI system or capability
- The model or AI capability that produces the output.
- Deployment Context
- How your organization obtains access to that capability — for example an organization-managed licence, a personal subscription or an API agreement. It describes the access route, not where the AI appears.
- Application or service
- The product or service in which the AI capability appears or is available.
- Organizational use and reliance
- What the organization depends on a concrete use for — a task or decision — and why that was considered acceptable.
How they relate
- One AI capability can be available in several applications, and one application can offer several capabilities.
- The same capability can be reached through different Deployment Contexts, with different terms and controls.
- An organizational use points at the concrete combination that is actually used. It is not tied one-to-one to any single layer.
- A change can touch any of them. Whether it matters depends on what the organization relies on — and that is for people to judge.
A host application such as a word processor, a meeting tool, a mail client or a design tool is where an AI capability appears. It is not how access to it is obtained, so it is not a Deployment Context.
An example
A support team drafts replies to complaints with AI. Described with the same four distinctions:
| AI system or capability | A general-purpose language model, in the version the vendor states. |
|---|---|
| Deployment Context | An organization-managed licence under the company’s business agreement. |
| Application or service | The helpdesk tool in which the team uses it. |
| Organizational use and reliance | First drafts only; an agent approves every reply. Accepted because a person checks each one. |
If the vendor retires that model version, a record like this could flag the reliance for review; whether it is actually affected is for people to judge. If the team stops reviewing drafts, the recorded reason may no longer hold — which is also a change.
