How do you see regarding business tasks. Not coding. Tasks like researching, summarizing, writing texts, routing to the next steps etc. Would in those cases a small LLM run in protected environment be sufficient?
And I also think about AI agents, especially the runtime environment. I could imagine in a few years, AI agents have an unique identifier and moving them from A to B would be painful or maybe even impossible.
Then it comes to several aspects: What does the AI agent do? Coding? Business tasks? Which model does he use? In which environment is it running? On which cloud?
I was thinking about using an open source AI agent runtime. But the questions about the models and the cloud vs. hardware remain. There are even European companies that host models on a cloud.
But I guess this is more for small businesses. What do you think?
For a bigger European enterprise, what would be the scenario for AI agents that do business tasks? Would it also be a 2 layer architecture?
Thank you Bianca, these questions are most interesting.
AI for business tasks:
A) Researching - Yes, absolutely. Perplexity does it with monthly costs or through API calls. More importantly, a company CAN build and host its own AI search and stop using 3rd party solutions
B) text summarizing and generating - Yes for non-confidential documents , and local models for anything else. Yes , it can be costly. Yes, it is more safe.
C) Routing - you don’t text a frontier model for this, a local AI routing hub can do it on a computer/ server.
About AI agents, I think what’s important is to approach them not as replacement of workers (hence, humanizing them) , but rather accept the fact that their performance is tied to a certain period, and not indefinitely.
It’s almost like with cloud and virtual machines - treat them as a herd, not cows. The auto scaling groups, the automated patching , etc etc - all are based on this principle.
Maybe this approach is valid for AI agents. So, with their short time-to-live, the question is not “how to migrate them from system A to system B” but rather “what context must be set for a new agent , so it can adequately perform its task”
It applies to the next “incarnation” of the agent in the same place A. Or B. Or Z.
All the questions about what AI agents does , how, where, when etc etc - all this defines an AI agentic operating model, which must be coupled with AI-augmented company’s operating model.
The sovereign AI is a local model, local agents, on local machines. Using local software. Under local licensing agreements , if external.
Thanks again for the opportunity to dive deeper into the subject. There so much to do!
Very interesting!
How do you see regarding business tasks. Not coding. Tasks like researching, summarizing, writing texts, routing to the next steps etc. Would in those cases a small LLM run in protected environment be sufficient?
And I also think about AI agents, especially the runtime environment. I could imagine in a few years, AI agents have an unique identifier and moving them from A to B would be painful or maybe even impossible.
Then it comes to several aspects: What does the AI agent do? Coding? Business tasks? Which model does he use? In which environment is it running? On which cloud?
I was thinking about using an open source AI agent runtime. But the questions about the models and the cloud vs. hardware remain. There are even European companies that host models on a cloud.
But I guess this is more for small businesses. What do you think?
For a bigger European enterprise, what would be the scenario for AI agents that do business tasks? Would it also be a 2 layer architecture?
Thank you Bianca, these questions are most interesting.
AI for business tasks:
A) Researching - Yes, absolutely. Perplexity does it with monthly costs or through API calls. More importantly, a company CAN build and host its own AI search and stop using 3rd party solutions
B) text summarizing and generating - Yes for non-confidential documents , and local models for anything else. Yes , it can be costly. Yes, it is more safe.
C) Routing - you don’t text a frontier model for this, a local AI routing hub can do it on a computer/ server.
About AI agents, I think what’s important is to approach them not as replacement of workers (hence, humanizing them) , but rather accept the fact that their performance is tied to a certain period, and not indefinitely.
It’s almost like with cloud and virtual machines - treat them as a herd, not cows. The auto scaling groups, the automated patching , etc etc - all are based on this principle.
Maybe this approach is valid for AI agents. So, with their short time-to-live, the question is not “how to migrate them from system A to system B” but rather “what context must be set for a new agent , so it can adequately perform its task”
It applies to the next “incarnation” of the agent in the same place A. Or B. Or Z.
All the questions about what AI agents does , how, where, when etc etc - all this defines an AI agentic operating model, which must be coupled with AI-augmented company’s operating model.
The sovereign AI is a local model, local agents, on local machines. Using local software. Under local licensing agreements , if external.
Thanks again for the opportunity to dive deeper into the subject. There so much to do!