HuskiesEveryone is selling AI agents this year. Few mention the second bill.
First the agent. Then the connectors so it can see your data. Then the pipeline that cleans that data so it can read it. Then the permissions, so it doesn't show payroll to a technician. Then the person who maintains all of the above.
The AI isn't the expensive part. Getting it near your data is.
An AI that needs an export is a project
If the assistant has to be fed a spreadsheet before it can answer, you've added a step, not removed one. The answer is only as fresh as the last export, and someone has to remember to run it.
The alternative is to put the assistant where the data already is. Same database, same records, same permissions as the person asking.
Watch it work
The second problem with AI at work is trust. An answer that appears out of nowhere is hard to act on. Was it the right customer? Did it count the rejected jobs? Did it change anything?
So Loki, the assistant in Huskies, shows its steps as it works:
- Which records it looked up, and with which filters
- What it did, step by step
- A stop button, if it's heading the wrong way
You ask in plain language, in Greek or English. "Which Northwind jobs are waiting for material?" "Book Crew A for Tuesday at 09:00." And you see how it got there.
Permissions come with it
Loki works with the same permissions as the person using it. If a technician can't see invoices, neither can Loki when the technician asks. Nothing is copied to another system to make it work.
Your AI doesn't need to connect to your data. It already sits on it.
That's the whole idea: the AI is part of the platform, on your own data, with no separate pipeline to build or pay for.