AI agents, agent management, and agent harmonization

Fingers and the cakes in the west of Westeros

What are AI agents, agent management, and agent harmonization?

Nearly two years after ChatGPT’s release, people are becoming more accustomed to using AI for everyday tasks. AI agents are allowing more time to focus on strategic and creative tasks, by carrying out the tasks we delegate to them. AI agents can handle specific tasks like searching for information, controlling devices, providing suggestions, etc. without constant intervention, speeding up task execution. This is just the beginning; more is possible with agent management systems and agent harmonization. They sound similar but serve different purposes. AI agents and agent management solutions are helpful with repetitive tasks, while agent harmonization enables individuals and teams to harness unpredictability and foster innovation. Since our team enjoys analogies, I'd like to share my understanding using a finger analogy to compare the roles of an AI agent, agent management, and agent harmonization.

The analogy: harmonization hidden in everyday life

If you are not suffering from severe motor difficulties, you probably don’t calculate every single move that helps you hold your pen, move the razor on your face, or guide a piece of cake to your mouth. You probably don’t do the estimation of how to move your pen to form a letter, how many millimeters you need to move a razor on your skin, how many degrees each of your knuckles need to rotate to slide it properly to shave your face, or you probably don't think about how tightly or gently you should hold the cake (sometimes you might accidentally squish it if the texture is different unless you've had the exact same cake before). You don’t make these estimations; if you did, performing only one of these tasks would consume your entire day and probably would be a huge accomplishment. However, you've probably completed all three and many more within the first waking hour of today without paying much attention to them.

Would a knuckle be enough?

Now that businesses are pushing to accelerate their rate of innovation, creation, building, etc. Some companies are focused on designing their own agents, Let's take creating a knuckle as an example of creating one. Each knuckle serves a specific purpose. To create one that fulfills its purpose, it requires a clear understanding of details like appearance, texture, and the tasks they handle. However, the point is that you can only do so much with one knuckle. So, if you want to undertake more advanced tasks, you would require a hand, meaning more knuckles to form fingers, and even more to form all five with distinct characteristics.

The need for a connection and training

However, having fingers doesn’t guarantee they'll work effectively; they need to be connected and coordinated to perform tasks like holding a pen. This is where agent managers come into play. They can manage the calculations for the knuckles and fingers—the agents—by following programmed instructions or executing tasks that we fully understand and have practiced enough to replicate, similar to robotic hands. Although it would be life-changing to be equipped with a robotic hand for disabled people, what about more complexities? What if the goal isn't something repetitive like simply holding a pen? What if we aim to create something entirely new, like painting, playing an instrument never played before, or inventing a new dance move?

The unknown unknowns: the itch we need to and should scratch

That's why scientists haven't stopped there and are still working on improving the connection of robotic hands with the nervous system, hoping to allow for more intuitive, human-like movement. To move your fingers and interact with an object, you need fingers, of course, a connection between them, and most importantly, the ability to move them without having to consciously calculate and rehearse every step, angle, and pressure, or be aware of all that. You'll need to be equipped with fast feedback and iteration on each decision you make. Now we're discussing agent harmonization, and Momentum, in particular, is working to equip teams with seamless cooperation among millions of micro-decisions, operations, and feedback loops to improve building speed. It aims to empower teams to be more ambitious in setting original goals and to be more confident in achieving objectives far more complex than just holding a pen. By encouraging rapid goal-driven actions and taking the burden of coordination from the human side. It's similar to making connections between the brain and a robotic hand, or more accurately, like how you function as a healthy person.

Harmonization is the key: tasting cakes in the west of Westeros

Take the cake example: you can’t always be cautious enough while eating because you don’t always eat the same kind of cake from the same pastry shop. Every time you try a new one, there's a chance that some of it will squish or fall on your clothes because there are so many variables at play. Even the same type of sponge cake from different bakeries doesn’t necessarily behave the same. Just like cakes, no project or team is the same. Ignoring the randomness at play and relying on rehearsed actions and reactions, as Nassim Nicholas Taleb says, makes us more prone to fragility. This level of complexity typically falls outside the scope of what agent managers are capable of handling, as their main focus is on managing routine known tasks. So, the value they provide tends to decrease over time because it becomes challenging to recognize their significance. But what about a new cake from a new bakery in Tokyo, with no one to tell you how to eat it properly? We need a more advanced solution, much like a human hand, to help us deal with the unknowns, overcome the constant fear of them, and explore freely. We need to be fearless enough to be curious about what cakes are like in the west of Westeros!