Network Intelligent Agent Automation
Using intelligent agents to automate operations of communications, processing and storage systems has become a popular subject. The recent acquisition of OpenClaw by OpenAI and the announcement of
NemoClaw by Nvidia have added fuel to the fire. As the fundamental technology advances, those applying the technology are working hard to move up the learning curve. To help in that
process, what follows is a generic approach for creating a group of intelligent agents to automate operations.
Background
Early on, communications networks operations and computer operations were considered separate functions. While cybersecurity wasn’t considered much at all. Tools were developed to help manual operations staffs do their jobs. As the complexity and scale of systems grew, attempts were made to reduce the stress on operations staffs (e.g. by over provisioning) in an effort to eliminate systems changes. But, system complexity and scale outgrew this approach too. Then, attention turned to automated orchestration. Orchestration was rebranded as Digital Twins.When GenAI arrived, it brought dramatically advancing capabilities in the intelligent agent arena. The application learning curve, as in most fundamentally new technology introductions has lagged. In part because the old ways of thinking about automated operations don’t fit the new technology. OpenClaw and NemoClaw have turbo charged the move to intelligent agents. There are likely to be other new tools introduced to improve productivity, security and effectiveness.
Intelligent Agent Five Steps
The names of the five steps are not new. What is new is that ways of approaching them effectively have changed. The five steps are:
1. Requirements
2. Architecture
3. Development
4. Deployment
5. Operation
Requirements
Requirements definition is still very important. And, maybe more important than in the past. This is for two basic reasons. First, recent AI development attempts by organizations that have involved line units in the process have proved to be more successful. While projects that have not heavily involved line organization staff have had a very poor track record. This is likely to be a result of the line staff’s detailed knowledge of requirements.
Second, in the past, requirements have been captured in a somewhat narrative fashion. For effective intelligent agents, requirements need to be captured in terms of Objectives, Algorithms and
Constraints. The clearer and more precise the better. The Objectives need to be statements of what the system must do. Constraints need to be statements of what the system must not do. Algorithms
are processes and procedures that the system can use to meet the Objectives and Constraints. These have to be described in precise, concise, complete and unambiguous language. This is because the
resulting statements are not just guidelines that will be used by a staff of programmers. They will become specific language that LLMs (Large Language Models) will use to determine
behavior.
The requirements also need to be a set of parameters that can be used to test the behavior of the developed system. Tests that determine if the system behavior fully meets what the
requirements specify. Finally, the requirements need to explicitly describe the needed security.
Architecture
The architecture described below is a generic structure that fits most operations intelligent agent systems. The nomenclature is used to make it easier for the reader. Architecture decisions
revolve around the level of granularity that the automated operations system will operate at. Today’s systems can be decomposed in three ways as combinations of:
B. Services, applications, modules, files, and data blocks
C. Large units (such as data centers, points of presence and communications centers), racks, boards, disk drives, and chips
The overall system can be considered as built of subsystems composed of the three ways of decomposition. Granularity refers to how ‘small’ a unit is under separate control. Implications include how many separate intelligent agents are in the overlay network that automates design, development, operations and security functions in the underlying system.
