I was reading some things about agent governance. Mainly Vitalik Buterin's tweet that exposes a framework for having agents become representatives of ourselves in democratic decisions. Basically personal governance agents.
The core problem with democratic /
I'm fascinated by this type of thinking. To me, it opens up a number of topics, one being having a virtual entity represent us not only from a decision making perspective but essentially us, who we are or represent ourselves to be when we communicate and act virtually. "This is my AI" you could introduce to your network. Over time, is the AI gathers more and more information about you, could it represent you when interacting within that network?
There's been plenty of talk and literature about an eventual economy of agents talking to agents from a task or output perspective. One agent is tuned to purchase mortgages on behalf of individuals with instructions to negotiate a good rate. The other agent is tuned to sell the most to the first agent - extracting as much value as possible via their interactions and eventual resolution.
I'm talking about an agent (or AI) that represents YOU. Your values, principles, beliefs, morals. Your political affiliations, your dislikes and likes, your penchants. It knows all of these things and is able to navigate situations that require input from you. Whether that's a decision or feedback about something in a conversation, or your thoughts on a topic that might require research before responding. They have complete autonomy to make decisions on your behalf - or at least - notify you that a decision needs to be made, here's how the agent would respond, do you approve or deny. If we skip the philosophical dilemma with virtual entities representing us, the question that comes to mind is: can this be applied to enterprises?
Plenty of organizations slow due to an onslaught of ideas, requests, epiphanies, and new directions. You've experienced it and will continue to experience it. Sometimes it takes weeks for a decision to be taken. In the meantime, a few more decisions were brought up and left while the initial decision was itself being discussed. Then some leader had an idea and that was discussed at length. The project's on hold, the product is on hold. There might be a need to redo the roadmap and the timelines. In the not-too-distant future, we could imagine agents whose purpose is to bring forward decisions. The agents have been fed user-generated documentation including roadmaps, timelines, priorities, etc. Their purpose is to stay on top of all of work. They've worked extensively with you to understand you and represent you appropriately. Imagine a little inflection point is reached. The agents alert all stakeholders that they must make a decision and vote on what the project should tackle next. If someone fails to vote within the allotted time period then the agent votes on behalf of the individual it represents.
There is a shift here within what I've described. Agents would make the process democratic. Under this scenario, there is no one individual making the decision. That will pose challenges for organizations that are used to that but for others, this forces to show to go on. It could even be taken a step further - the agent in charge of the project/product (who talks to the agent in charge of the democratic process) talks to the agent that monitors code releases. The code release agent already has multiple designs for solutioning the need and presents those to the agent committee. Those agents then vote on the most optimal solution and the code is released. All without human intervention. The agents have specific roles. They represent those roles or individuals within the organization that hold those roles. Now there are a multitude of things that could impact this process - one being the perceived complexity of business solutions (they say it's not that simple but I can be and should be). Any tacit knowledge won't be something any AI can action on. That's another problem that needs to be solved and puts the emphasis on context storage.
When thinking of implicit or explicit knowledge, we can envision an environment owned entirely by agents. For every business process unique to an organization, there are twice as many software or data engineering processes that are agnostic. They can follow best practices and a superior structure. Those agents know the best practices, they know the structure. They can implement the structure and the foundation that enables engineers to focus on handling logic and processes that make up the business. That's the future that we're heading towards. Where, by-and-large, the data engineering function can be packaged and sold (next blog post hint). Within this spectrum, the argument can also be made that AI can elicit tacit knowledge. Think of all of the individuals asking a model about the work they do, what they need help with. That knowledge is retained and can be accessed for future use. Best use case for AI?
Now the road to agents being our representatives is long but worth thinking about. Does the outcome change if you or your AI rep voted for something? How would you know? If the project moves forward on time and as expected, does it really matter if you've had a chance to opine? Do we get lost in the quagmires of our organizations? There's something to be said if we remember that it's all about business.