How many AI Agents will we each have?
and why...
Jeremiah Owyang asked this great question over on LinkedIn and for opinions and logic; so here goes.
I think people will have one Primary Fiduciary Agent ('like an Orchestra Conductor’), then up to a hundred or so Fiduciary Action Agents that take their instructions from the Conductor. A proxy for that number may well be ‘the same number as the number of apps you have and use on your mobile phone’. In my case that’s seventy five.
They will then have one more Delegated Fiduciary Agent for each role in life in which they carry legal responsibilities. That would mean one for an employment role, one for a trustee of a charity, and so on. I say so because these delegated fiduciary agents will be more than capable of ‘signing things’ on our behalf within a role context, albeit with the guardails and trusted data approaches that would underpin this. Perhaps then one as a head of household, parent/ guardian or carer, although I think in practice they are less driven by regulation so would likely be absorbed into the primary agent with appropriate functionality.
The above are fiduciary agents, acting specifically on behalf of a single human-being with a duty of loyalty and duty of care to that individual. Then both the individual and their fiduciary agents will encounter and interact with thousands of other non-fiduciary agents. The proxy for this might be the number of websites a person visits over time. For example, I will not need to go the Scotrail app to know if my train is on time or not; my ‘get a train; fiduciary action agent will bring that data to me when I need it. Every click and second saved is a benefit…
My logic:
A person’s primary fiduciary agent has very priviliged access to data, and needs to be able to act across many fronts. So we will not want any more of these than we need. It’s a bit like a GP/ Medical General Practitioner; we don’t need or want more than one of those. Having more than one leads to silo-ed data and actions being taken.
And we know that if there is a specialist that needs to be brought in by the primary agent for a specific task then that is fine. But typically the interaction continues to be done through the primary agent. (e.g. test results are returned to the GP to communicate to the patient).
The fiduciary action agents are more specific, not generalist. They undertake specific workflows running on a precise data-set to enable pre-defined outcomes. And as we have tens and maybe hundreds of things we need to do each day that now have digital components then we will need tens and maybe hundreds of action agents. Some we will use daily, some on a regular schedule and some ad hoc.
Non-fiduciary (to person) agents are a perfectly valid concept. That just means they are fiduciary to another person or entity.
I’ve tried to summarise that in the visual below.
Next up…., a day in the life of a Fiduciary Orchestrator Agent….