From the Division of Labor to the Division of Agents

Just as mass production revolutionized manufacturing, Intelligent Process Automation (IPA) will transform service industries. This shift is driven by the digitization of processes and the advancement of artificial intelligence (AI). The impact of digitizing service processes parallels the effect of machinery on manufacturing, enhancing human productivity and precision. While robotics have enabled nearly fully automated production processes with minimal supervision, IPA will similarly automate standard service processes and significantly contribute to non-standard processes. This will be achieved through the integration of software applications, General AI, Web APIs, and Orchestration technologies.  

Agentic Processes will initially be implemented through the establishment of a Division of Agents. Analogous to how the Division of Labor standardized specialized tasks to achieve significant scale, the Division of Agents aims to automate service processes with similar success. Each specially trained Agent will execute a specific task with high predictability. The output from one Agent will then be handed off to the next specialized Agent (or Co-pilot) and so on, until the process is completed. An Orchestration Agent will plan and coordinate these Agents to ensure the smooth execution of tasks, while an Information Agent will gather data to generate process Key Performance Indicators (KPIs). Humans (Human In The Loop or HITL) will serve as escalation points, akin to how a floor manager oversees a production line in a factory.

The Division of Agents will be more profound with the establishment of the Small Language Models (SLMs). SLMs will enable Agents to possess specialized knowledge, allowing them to address specific tasks with greater accuracy. For instance, a Civil Engineering Agent could create initial drawings in a Building Information Model (BIM), which the user can then refine. This Civil Engineering Agent would collaborate with Electrical Engineering Agents, Mechanical Engineering Agents, and other specialists to produce a comprehensive construction design based on the user's input. The user can then make, or direct the Agent to make, any necessary adjustments to the drawings. Given its user interface capabilities, the Civil Engineering Agent should also be able to coordinate with other Agents to effect changes in response to modifications in the drawings—understanding and planning for the repercussions of each alteration.

The same Division of Agents will occur in software. Software agents will be trained on the interfaces and commands of applications such as SAP R/3, SAP S/4 HANA, Salesforce, among others. These pre-built application agents will then be provided to consultants to configure them according to the specific requirements of an organization. Eventually, an organization will have an application, such as ERP or CRM, tailored to their needs with agents that automatically perform tasks within the application. These agents will be orchestrated to create a seamless Intelligent Process Automation (IPA).

Manufacturing transitioned to mass production when engineers succeeded in standardizing the designs of products and constructing machines capable of producing these items in large volumes. Service industries, however, have traditionally been driven by human input due to the variability inherent in their processes. This variability arises because many service processes begin with unstructured information that must be processed into a structured format. For instance, if an organization seeks to create a Request for Proposal (RFP), it must gather unstructured data, structure it in a way that potential vendors can understand, receive responses in various formats, make comparisons, and ultimately create a purchase order (structured data) in an Enterprise Resource Planning (ERP) system. Agentic AI has the potential to automate much of this work because it can process unstructured data, perform actions within systems, and coordinate users and agents to achieve the desired outcome.

The Division of Agents refers to the existence of various types of Agents within an organization. These include Payroll Agents who enter payroll data into payroll systems, IT Agents who create and resolve tickets on behalf of users, Legal Agents who receive input from users and draft legal documents such as contracts, Accounting Agents who determine the appropriate postings for an ERP Agent upon receiving an invoice, and Customer Care Agents, among others. In this environment, not all processes will be driven by AI. Organizations will continue to utilize software for routine tasks or to execute clearly defined processes. Agents will be deterministic—executing specific workflows or actions—and stochastic, utilizing artificial intelligence. The integration of these technologies will result in Intelligent Processes (IPAs), where their combined use will drive innovation and efficiency.

Use Case:

User: Could you please inform me of my remaining vacation days? 

Agent checks the HR system for the user's remaining vacation days.

Agent: You have 5 days left until the end of the year.

U: Please schedule 3 vacation days for me starting from the 15th of March.

Agent reviews the calendar and verifies the schedule.

A: There is a meeting scheduled with Client Y and your Manager on the 16th of March. Would you like me to cancel this meeting and book your vacation?

U: Yes, please send an email to the client requesting to reschedule the meeting to one day later.

Agent books the vacation days, initiates a workflow to notify the User’s manager about the meeting change and requests approval for the vacation. The Calendar Agent sends emails to the client proposing a new meeting date.

Type of Agents

·        Action Agents: Perform specific tasks within one or more systems via APIs or GUIs. They are specialists, trained to execute particular functions and capable of taking targeted actions.

·        Information Agents: Can read both structured and unstructured data, retrieve relevant information, and provide well-reasoned summaries, comparisons, and library navigations.

·        Controlling Agents: Monitor key performance indicators (KPIs) related to agent and process performance, check for errors, collect user feedback, and make necessary adjustments to other agents. In a cybernetic (self-regulated) process, the controlling Agent compares outcomes with Agent or User (AoU) inputs and adjusts accordingly. In a human-regulated process, the Controller acts upon human input to take corrective measures. For instance, if an Agent is asked to compare payment terms between a contract and an invoice and approve them if they match, but later the User identifies an error, the Controlling Agent would initiate corrective action through the Orchestration Agent by flagging the invoice as rejected and starting the rejection process.

·        Orchestration Agents: Orchestrate the various Agents involved in the process. Orchestration Agents may also manage other Orchestration Agents responsible for smaller segments of the process or different processes, ensuring vertical or horizontal integration.