Skip to content
The AI Agent Factory

1.4 Continuity, execution and persistence

Approved by
Panaversity ·
6 min read

In everyday life. Your family doctor is the same doctor for years, even though each visit happens in a different exam room. Your records are kept in a file the office can pull up any time. The doctor, the room and the file are three different things.

Products mix together three ideas that a manager must keep apart.

  • Worker continuity is what makes a worker the same worker over time: its identity, its responsibilities, the context it keeps, and the recurring work it does.
  • The execution environment is where one piece of work runs. It is often temporary.
  • Artifact persistence is where results are saved, and how you get them back.

Each of the three can change without the other two. A standing worker does not need one computer to keep running forever. A temporary machine does not mean the worker forgets who it is. And a finished spreadsheet on a machine is not a saved spreadsheet until it is saved in a place where you can get it back.

Over time the worker stays the same while task environments come and go. Delivered results remain. Task 2's result, never delivered, may be lost.

Figure 1.3. Continuity, execution and persistence vary independently.

The third idea, artifact persistence, depends on where things are kept. One piece of work can touch three places:

PlaceWhat lives thereHow long it lasts
Your account or workspaceTranscripts, delivered outputs, documents, projects, schedules, memoryDurable. Some of it is personal, some is shared by your organization
The execution environmentWorking files, installed tools, work in progressOften temporary. Check each product
Your own computerYour folders, browser sessions and appsAs long as you keep them. Cloud work reaches them only through a desktop app, with your permission

Work that needs files on your own computer reaches them only through a desktop app that is open on that computer.

Three places for one piece of work. First, where the work runs: a cloud sandbox, a temporary machine in the AI vendor's cloud that holds working files, installed tools and work in progress, often temporary. Second, where your files live: your account or workspace, which holds transcripts, delivered outputs, documents, projects, schedules and memory and is durable, and your own computer, with your folders, browser sessions and apps, kept as long as you keep them. Third, what needs the desktop app open: cloud work reaching your computer. With the desktop app open, the work goes from the cloud sandbox through the app to your own computer and reaches your folders and apps, with your permission. With the app closed, there is no way in, and the work cannot reach the files on your computer.

Figure 1.4. Where the work runs, where your files live, and what needs the desktop app open.

The practical rule is short. Save anything you need later to a known place, and check that you can get it back. Know that place's retention and access rules. Maria's spreadsheet, in the chapter's opening story, failed this rule. Her summary survived because it was part of the conversation.

There is a second, deeper point. Durable is not the same as authoritative. Personal memory belongs to one login. A shared project belongs to a team. Neither is your company's official record of policy, and neither is the live state of the ledger. Chapter 4 adds three layers for that. KSoR (Knowledge System of Record) governs what an AI Worker may know. DSoR (Data System of Record) governs what it may do. SSoR (State System of Record) tracks where the work stands. These three names are this book's own terms, from open Panaversity specifications, and exam questions use plainer words for them: approved knowledge, permitted actions and where the work stands.

Check yourself

Question 1 / 8

0 answered

Which of these is durable account state?

Owner
Panaversity