AI
AI in your business, on your terms.
Every system we build has a defined place for AI. You decide what goes there, and you can change your mind.
The shift
Business software kept the records. Now it can do part of the work.
For years, a business system was where work got written down: the order, the quote, the customer call. People did the work and the system kept score.
A model can now read the inquiry, draft the quote and file the note. Questions once settled in a meeting now have to be settled in the software.
What this means
The questions that matter are plain ones.
- Who approves before anything is sent?
- Where does the data go?
- What happens when the answer is wrong?
- Can you switch it off next year?
A system is ready for AI when each of these has an answer you can test.
Your choice
Four ways to fill the same place.
- What it is
- A model on your own servers.
- Fits when
- The data has to stay inside the company, or use is high and steady.
- Where the data goes
- Nowhere. It stays on your servers.
- In our products
- ZEUS PODCAST translates on a local model. ZEUS ANiME localizes subtitles on one, off by default.
- What it is
- A model from the provider you choose.
- Fits when
- You want the strongest models and pay by use.
- Where the data goes
- To that provider, under your own contract.
- In our products
- ZEUS PODCAST transcribes through a cloud service.
- What it is
- A module we write, with no model in it.
- Fits when
- The rules can be written down and the result must be the same every time.
- Where the data goes
- Nowhere. It stays in your system.
- In our products
- ZEUS FRANÇAiS conjugates verbs with a rule engine.
- What it is
- No AI module. Your staff do the step; the system routes, reminds and logs.
- Fits when
- The judgment has to stay with a person, or AI can wait.
- Where the data goes
- Nowhere. It stays in your system.
- In our products
- ZEUS OFFICE runs with no AI features.
The rest of the system stays as it is: records, permissions and workflows. You can switch later.
AI-ready
What AI-ready means in a system we deliver.
An AI-ready system is one you can connect an AI assistant to safely, on the day you choose.
-
Documented interfaces.
Core records can be read and written through an interface, within permissions.
How we verify itInterface documentation, tested with you.
-
A standard connection for AI assistants.
Off by default, switched on when you ask.
How we verify itA general-purpose assistant connects and completes one read and one write.
-
The same permissions as the person.
An assistant sees and changes only what that person could.
How we verify itThrough a low-permission account, it cannot reach restricted data.
-
Approval before high-impact actions.
Deleting, sending and bulk changes wait for a person. The list is agreed with you.
How we verify itWe trigger each listed action; none runs without approval.
-
A record of every action, and one place to disconnect.
You can see what was done, by whom, and stop it.
How we verify itEvery action appears in the log. Once disconnected, the assistant can no longer read or change anything.
-
A model you can replace, or remove.
Changing the model leaves your data and workflows untouched.
How we verify itEvery business function still works with the model switched or removed.
In practice
How we do this in our own products.
-
- Brand
- Website
- Site plans
- Contracts
Reads: Site plans
One assistant, one folder.
In ZEUS IRiS, the assistant in a project chat reads that project’s folder and nothing else.
-
Proposed decision
Move the review to Thursday.
ConfirmNot nowExpires in 30:00
It proposes. A person confirms.
Before ZEUS IRiS records a decision, it asks. With no answer in 30 minutes, nothing is written.
-
Done. The quote has been sent.
Stopped: no matching action in the log.
Only what happened is reported.
A reply that claims an action the system did not take is stopped before it is sent.
-
- 0:12Welcome back to the second session.
- 0:18Today we look at three short examples.Edited by a person
- 0:25Let’s start with the first one.
Model update
Human edits are kept.
In ZEUS PODCAST, a line corrected by a person is never overwritten by the model.
Engineers call this layer the harness: everything around the model, from tools and permissions to approvals and logs.
In other systems
We check this in systems we review.
Our readiness review checks how people work with agents: how a request reaches the agent, whether it asks before acting, whether corrections reach it while it runs, and whether it reports results truthfully.
Readiness reviewTell us about your business challenge.
rieslin.z@zeusinsights.net