Humans are the first and only species that possess reasoning capabilities beyond the most basic ones that are exhibited by other mammals. We build amazing infrastructures, put people on the moon, can talk to anyone on the planet at any point in time using phones and create advanced medication and procedures to treat people who are ill, extending average life span by 40 years over the last century, to name just a few of our accomplishments. Not a single species comes even close to what we’ve achieved as humankind.
Nevertheless, we still have limitations. The typical human can keep 7 ± 2 concepts in their head at any point in time. Our response times are measured in milliseconds or dozens of milliseconds. The number of computations we can process per second is rather low. The amount of knowledge and information a single human can cram into his or her brain is quite limited as well.
The secret to our success is divide and conquer. We take on large, complex problems, break them into smaller pieces and give the pieces to individuals who may divide them even further or solve the problem at that granularity. Once each chunk has been addressed, we reverse the process, integrate the results and deliver a large, complicated system that solves the large, complex problem.
Because humans have these limited information-processing capabilities, the importance of compartmentalization is paramount. People need to be able to solve their specific problem in isolation without having to consider a host of dependencies and constraints outside of their scope. Of course, no decomposition is perfect, but the more dependencies and constraints exist, the slower and lower-quality the resulting solution tends to be. This is why the topic of architecture is so prevalent in computer science and software engineering: the initial decomposition into subsystems and components has an outsized impact on the effectiveness and quality of the final system.
This decomposition is used everywhere in companies, especially around roles and processes. The product manager is concerned with what the product should entail in terms of functionality, but not how development should realize this functionality. Similarly, development shouldn’t be concerned with why certain functionality should be realized; it should just focus on building it. Sales shouldn’t be concerned with the evolution of the functionality in products; it should simply focus on selling. Finance shouldn’t care about anything besides the numbers.
As companies break their organization into departments and teams, these organizational units develop their own ways of doing things and define interfaces to other departments and teams in the organization. Preferably, these interfaces are as narrow as possible for the aforementioned reasons of effectiveness and quality. A company that finds a suitable decomposition and manages to execute well in each of the components tends to do well for itself as long as the industry is in a relatively stable state. The moment we enter times of disruption, though, any decomposition automatically turns into a disadvantage, as it often is no longer optimal in the face of the new and changed reality.
The interesting thing is that although AI agents have limitations as well, these tend to be different than the shortcomings of us humans. One of their abilities is to include a broader context in their operations than we often can. This brings us to the topic of super-agents.
Once the first processes have been reinvented from an AI-first perspective, the next stage is to explore the dependencies between them. Optimization can be achieved by continuous connection and mutual adjustment.
In most companies, for example, when the product development organization designs a new product, it decides which components to develop internally and which components to source from suppliers. The sourcing list is then given to the procurement team, who is responsible for acquiring the components at the right quality level and the right price level. However, there seldom is a bi-directional exploration where procurement offers suggestions on components to be acquired that, if selected, could improve the product in some way, like lowering the bill of materials, decreasing maintenance costs or increasing differentiation.
The input we received during our interview study is that some of the companies are thinking about and exploring ways in which they could use AI agents to connect business processes that were earlier disconnected and thus achieve optimization across business processes. If this can be accomplished, the potential is enormous, as in most cases, companies use static, default structures that are far from optimal for most contexts. In addition, whenever humans are involved in the process, there’s a tendency to increase the scale of each transaction or the batch size of each activity. For instance, most procurement managers seek to negotiate discounts by ordering larger quantities. With AI agents, the deal size can be reduced to much smaller amounts, potentially to a unit size of one, and marketplace bidding approaches can be used to achieve savings.
We call this stage the “super-agent” level, as we’re now connecting AI-driven processes that traditionally have been disconnected by introducing “super-level” dependencies, handled by AI agents that don’t manage or execute processes but only the dependencies between them. In this way, we can achieve efficiencies that can’t be attained cost-effectively when humans are expected to execute these steps.
The fourth level in our AI-driven organization maturity model is the super-agent phase. Here, we focus on connecting previously disconnected AI-first business processes to replace static, standardized connections with dynamic and continuously optimizing connections that improve efficiency and allow for lower cost, delivery of more value or other benefits. We can achieve this because AI agents have different limitations than humans and have less need for compartmentalization and separation than human organizations and teams. To end with a quote by Henry Ford that hammers home this point: coming together is a beginning, staying together is progress and working together is success. This is as true for our business processes as it is for our teams.
Want to read more like this? Sign up for my newsletter at jan@janbosch.com or follow me on janbosch.com/blog, LinkedIn (linkedin.com/in/janbosch) or X (@JanBosch).
