Did I understand you right there? You would want to replace the voting by shareholders with the voting by autonomous agents?
You don't have me right. I want autonomous agents to direct the voting power of shareholders in accordance to their preferences. Voting agents would be similar to trading bots and would need to exist for similar reasons. Just as politicians are agents acting as political robots the autonomous agents are software agents acting as delegates. In situations where software agents can do the job then we should not burden human beings by making them act as political robots because honestly human beings aren't very good at being robots.
I would have two concerns then:
1) Either such voting decisions are very complex and need human assessment: Is the 100% pay justified? Should webdeveloper 1 or webdev 2 be voted in?
In this case the voting agent could simply send a text message to your phone, a message to your instant messaging app, an email, or some other alert which gets your attention. The point is that if a vote doesn't have to rise to your attention then it should be handled by your voting agent.
In the case where it's votes which you think need human assessment a lot of the time it doesn't. For example when people shop for food do you think they shop based on brand loyalty, based on the marketing, or do you think they read and analyze every ingredient on the ingredient list? Most humans due to attention scarcity do not know what they eat. Additionally humans don't really know what the laws are either because no one has the time to read thousand page bills passed in the middle of the night.
2) Or if decisions are rather easy (like for plain node delegates, no worker/paid delegates) and depend on metrics like reliability, trust (which again is harder to verify) etc. then an attacker could prepare lots of delegates that are modeled after the criteria of the autonomous voting agent.
The criteria of your voting agents would not have to be public. If it is public it does not mean attackers could model after it if the criteria takes reputation into account and or requires recommendation from members of your web of trust.
I think the purpose of autonomous agents is the fact that due to the sheer magnitude of the information which will eventually be available that the decision making becomes more difficult for humans as more information (and greater amount of choices) are offered. So for example if you take shopping again and you consider the fact that almost no human being has expertise in chemistry, nutrition and complete control of their attention to ignore all the advertisements, branding, then you'll realize most humans only show the illusion of free choice when in reality due to attention scarcity their choices are subconsciously dictated often by subliminal advertisements influencing their subconscious.
What I mean is that most people shop on gut feelings, ads, trends, and a lot of that is unconscious or subconscious decisions which are influenced by marketing. This means the typical shopper may not get very good deals or make very good choices because they simply cannot process all the information about every option available. If it's food almost no one is going to research every ingredient in every product and even if they tried they probably won't track the calories, the price, the amount of food compared to previous years, all which continuously changes to keep consumers off balance.
In the case of Bitshares then someone might offer a proposal and if Bitshares were to scale there could be thousands of people offering proposals every week. After a certain point no one is going to have the time to read through every proposal and also if we just rely on our experts then all voting power centralizes around these people who could someday collude. Autonomous agents would keep the power decentralized, allow the decision making capability to scale, and if it is a decision which no autonomous agent can make for you then these decisions would rise to your attention but if it's a decision you can train or teach an autonomous agent to make then you'd be able to save your attention for the more difficult decisions you have to make.