Q 762. Imagine a scenario in your domain where multiple AI agents — each with a different specialty — need to collaborate to complete a task (e.g., one handles customer interaction, another verifies documents, and a third updates records). Describe one such multi-agent collaboration scenario.

What challenges might arise in their coordination, and how would you design their interaction so the process remains smooth, accurate, and explainable?

 

🏆 The best answer will be selected on the basis of:

• Creativity & relevance of the multi-agent use case

• Clarity in defining roles and handoffs

• Insight into coordination or conflict-resolution mechanisms

 

Note for website visitors -

• This platform hosts two weekly questions, one on Monday and the other on Thursday. 
• All previous questions can be found here: https://www.benchmarksixsigma.com/forum/lean-six-sigma-business-excellence-questions/. 
• To participate in the current question, please visit the forum homepage at https://www.benchmarksixsigma.com/forum/.
• The question will be open until Monday or Thursday at 5 PM Indian Standard Time, depending on the launch day. 
• Responses will not be visible until they are reviewed, and only non-plagiarised answers with less than 5-10% plagiarism will be considered for winner selection. 
• If you are unsure about plagiarism, please check your answer using a plagiarism checker tool such as https://smallseotools.com/plagiarism-checker/ before submitting. 
• All correct answers shall be published, and the top-rated answer will be displayed first. The author will receive an honourable mention in our Business Excellence dictionary at https://www.benchmarksixsigma.com/forum/business-excellence-dictionary-glossary/ along with the related term.

• Some people seem to be using AI platforms to find forum answers. This is a risky approach as AI responses are error-prone because our questions are application-oriented (they are never straightforward). Have a look at this funny example - https://www.benchmarksixsigma.com/forum/topic/39458-using-ai-to-respond-to-forum-questions/

• We also use an AI content detector at https://quillbot.com/ai-content-detector. Only answers with less than 45-50% AI-generated content will be considered for winner selection.

 

Every software company/organization has client facing support teams(different levels of support). These teams handle the support issues that are raised by the clients. If needed they act as the channel to interact with different set of development teams. Even in case of Emergency or IRC these teams are the first to be engaged. When the support teams decides/suspects that there is a bug/mistake/flaw in the workflow/design the support team involves the development team. Considering the support teams as different agents in this process there can be multiple factors that affect the co-ordination  :  1) communication, 2) Lack of conceptual awareness, 3) conflicts due to prompts, 4) complexity of  issue and engagement, 5) Security , 6)Understanding of user issue/context, 7) Quality of data collected, 8)integration, 9) inter-operability, 

10) emergency of the scenario/situation, 11)User experience/interaction with the process and so on. 

 

I think the first/best option is to implement a hand shake procedure between the AI agents to enable the smooth communication process. 

 

  Ex : This would be similar to the three handshake process that our Internet (TCP/IP) protocol uses. First the agent that wants to communicate sends out an open synchronize/connection request, which then will be synchronized-acknowledged by the second agent. The first agent then responds back with an acknowledgement. This makes sure that the communication channel is established. 

 

The handshake procedure allows the agents to first start a reliable communication. Based on the handshake, we then start the communication by sending the issue, context of the issue, processed data, negotiating parameters or the expectations, state of the scenario/situation and then log the necessary data on both sides. We can do this for every communication aspect. We can also use the log data to further study/research/analyze when there are hinderances. As the support team is assumed/considered as an agent we can leverage the same  for each and every role or across the agents. This way we can have a proper hand off between the agents.

 

Once the agents are successfully engaged, while there can be different options/solutions/patterns or algorithms that can be used, based on the necessity of the organization we can use a centralized approach or a de-centralized approach as the agents might need to concurrently work on different issues/tools/applications. Based on the cost/budget we can analyze the necessary factors/features that we can first address/implement and later update the same agents with different features to avoid conflicts/hinderances. Deployment/Maintenance/Monitoring of these agents would be the next step. 

 

Other features/factors we might need to account for : size/quantity of the incoming issues, resolution times, latency in engagement, integration complexity, reliability, bias. We might also need to focus on several integrity issues, source of truth for the agents, standardization methodologies, scalability and performance issues, the number of agents needed (based on the issue queue),  re-direction of unused agents.

 

In an example of fulfilment Ops in my current organization, we have 3 AI products (1. Invoicing; 2. Payment support; 3. Dispute).

• The invoicing AI enables seamless invoice submission with the necessary, inputs from the core system like the Customer ID, product, payment terms etc.
• The expectation from Payment support AI is to classify payment query emails, extract information from the source systems and generate email responses.
• The Dispute AI looks into the query and extracts the required info and recommends solutions basis the disputes raised by customer and pulling data from source

In the above case, the primary requirement is to ensure that there is a common source of information enabled for all the above 3 products without which it would be challenging for the AIs to generate the right level of information to the customer query. e.g. - If the invoice details and payment terms are not aligned the, the support AIs will pull incorrect information to respond to queries or resolve disputes.

 

Ensuring the source is common, real-time updates enabled for the source and a standardized communication template is designed to ensure standardized response and query management.

 

Scenario: AI-Driven Pharma R&D Product Development

Use Case Overview

In R&D, multiple activities are required to design a new drug product (e.g., a tablet, capsules, injectable, or others), which requires expertise in various domains. Here, below example is provided where three AI agents collaborate:

1. Drug-Excipient Compatibility Agent – Analyzes drug-excipient compatibility concerning interaction, stability, and physical behaviors.
2. Drug Product Formulator Agent – This agent suggests the most suitable formulation composition based on the best pharmacokinetic profiles. This module also suggests suitable alternatives, with the best alternatives that can be scaled.
3. Regulatory Compliance Agent – This module ensures that the suggested formulation meets all regulatory guidelines.

Product Development Process Flow

1. Drug-Excipient Compatibility Agent screens the best excipients (e.g., binder, filler) and highlights incompatible combinations (e.g., a drug that deteriorates in UV light).
2. Drug Product Formulator Agent provides suitable formulations for which the release profiles match with reference products.
3. Regulatory Compliance Agent evaluates formulations as per regulatory guidance and immediately highlights any discrepancies before conclusion.

Challenges in Coordination

1. Conflicting Priorities – Drug-Excipient Compatibility Agent may suggest an excipient that is uncommon or not listed as per the regulatory guideline. In case of Drug Product Formulator Agent suggest a complex formulation that has the highest chances to match, but Regulatory Compliance Agent flags that this seems to be difficult to approve.
2. Data Ambiguity – Regulatory Compliance Agent may not be updated with real-time updates from regulatory agencies.
3. Regulatory Gaps – A formulation developed by Drug Product Formulator Agent could fail due to a lack of toxicity data for one of the novel excipients.

Designing their interactions

1. Interactive Feedback Loops

• Agents share continuous awareness proposals (e.g., formulator agent suggests a formulation for a delayed release coated tablet and Regulatory Compliance Agent checks if the proposed formulation has any impact as per regulatory guidelines.
• If Regulatory Compliance Agent rejects a proposal, Drug Product Formulator Agent generates alternatives (e.g., matrix tablets instead of coated tablets).

2. Conflict resolution via priority scoring

• A rank-based decision system prioritizes ranking criteria (e.g., Regulatory compliance > compatibility material > release profile).
• Example: If Regulatory Compliance Agent and Drug-Excipient Compatibility Agent disagree on a selected excipient, the system defaults to the option with the highest compatibility + compliance score.

3. Rationale for decision

• Each agent must record rationale (e.g., "Drug-Excipient Compatibility Agent rejected Lactose monohydrate due to Maillard reaction").

Case Study: 3-Agent AI Collaboration for Statutory Audit Onboarding

Client Overview

Client: ABC Pvt. Ltd.
Industry: FMCG
Engagement: Statutory Audit
Timeline: 3 Days Total

Workflow & Timeline

Day 1: Client Interaction & Document Collection
- Agent A (Interaction Bot) initiates onboarding and collects basic details.
- Shares secure link for uploading PAN, GST, CIN, and financials.
Time Taken: 2 hours
Outcome: Documents received.

Day 2: Document Verification & Risk Flagging
- Agent B (Verifier) checks document validity via OCR and APIs.
- Identifies minor discrepancies in GST and financial data.
- Flags a pending litigation.
Time Taken: 4 hours
Outcome: Risk rating assigned: Moderate.

Day 3: Approval & ERP Update
- Agent C (Workflow Manager) routes case to Partner for review.
- Partner approves with remarks, ERP updated, and audit team notified.
Time Taken: 1 day
Outcome: Engagement formally initiated.

Benefits Achieved

Aspect	Manual Process	AI-Driven Process
Total Time	7–10 days	3 days
Risk Detection	Post-review	Early-stage detection
Human Involvement	High	Moderate
Process Transparency	Limited	Logged & Explainable

This lean approach demonstrates how even a 3-agent AI system can transform audit onboarding through automation and explainability.

Imagine a BPO company handling customer onboarding for a bank. Three AI agents work together:

1. Customer Interaction Agent – Talks to the customer via chat or voice, answers questions, and collects required info.

2. Document Verification Agent – Checks uploaded ID proofs, utility bills, etc., to make sure they’re real and match the customer’s info.

3. Record Update Agent – Once everything’s good, this one updates the internal system with the new customer’s details.

They need to work like a team. The first one collects the info, passes it to the second to verify, and when that’s done, the third updates the database.

---

Challenges in Coordination:

• Miscommunication: If one agent misunderstands or mislabels the customer data, the next one may mess up.

• Timing issues: If one agent finishes too soon or takes too long, it could mess up the flow.

• Error handling: If something goes wrong (e.g., a blurry document), which agent should take charge?

• Explainability: If something fails, it should be easy for a human to understand why.

---

Designing for Smooth Collaboration:

• Clear Handoff Rules: Set specific points where one agent “hands off” to another — like a relay race.

• Shared Workspace: Use a common data platform where all agents store and read info in a consistent way.

• Fallbacks and Alerts: If one agent gets stuck, send an alert or ask for human help instead of just failing.

• Logs and Audit Trails: Record every step each agent takes so we can trace what happened and explain it clearly if needed.

• Central Orchestrator: Think of it like a conductor in an orchestra — it makes sure all agents are in sync and doing the right thing at the right time.

As a BPET team, we conduct comprehensive audit of the company's operations, ensuring that all client contractual obligations, ISO standards, Statutory and regulatory regulations, and company internal controls are being followed.

 

We can involve the following AI Agents:

• Agent C – Contractual Compliance Agent
• Agent I – ISO Standards Agent
• Agent S – Statutory & Regulatory Compliance Agent
• Agent R – Records and Reporting Agent
• Agent H – Human Interaction Agent

 

Collaboration Flow between these AI agents would be:

• Agent R initiates the audit, triggering workflows for Agents C, I, and S.
• Agent C parses active client contracts → flags missed SLAs.
• Agent I scans ISO-related SOPs, audits access logs → detects missing training logs.
• Agent S analyzes statutory compliance → flags 2 outdated safety certifications.
• Agent H contacts HR to upload updated training and safety records.
• Agent R aggregates the domain agent outputs.
• Agent R creates an audit summary categorized by compliance domain.

Most likely challenges in coordination would be:

• Semantic Misalignment
• Timing Conflicts
• Contradictory Judgments
• Traceability & Explainability
• Regulatory Drift

What would be the System Design for Smooth, Accurate, Explainable Interaction:

• Compliance Knowledge Graph
• Orchestration Layer (Compliance Task Manager)
• Compliance Ledger with Provenance Logging
• Shared Messaging Protocol
• Conflict Resolver Module
• Human Auditor Interface

In a BPO environment where utility bills are generated and payments are received, below are various AI agents that can be a part of . 1) First AI agent calls the customer, guides them to take a picture of their electricity meter and submit it on the website 2) Second AI agent will process this and updates the records 3) 3rd AI agent should generate a bill 4) 4th  agent will request for a payment from customers bank details 5) 5th agent will obtain the amount and assign the payment to the correct customer's profile 6) 6th agent should share reminders in case the payments are not received

 

Below are a couple of scenarios where there could be challenges in their co-ordination

 

Scenario 1

There may be issues with the picture being shared by the customer which the 2nd  agent may not be able to read clearly. It should identify in case of any inconsistent readings and highlight this to the customer instead of generating a bill. To avoid this we should train the agent to read the previous values and compare the current one against it.  If the current reading is lower than the previous one then the agent should throw an exception and stop the next AI agent from generating a bill

 

Scenario 2

 

Once the bill is generated then the 3rd AI agent should take the payment from the correct bank details. In case there is more than  one bank detail the agent should know how to select the correct bank account. In case the payments are not received then 6th AI agent should start sending reminders highlighting what went wrong while attempting to get the payment

 

For this discussion, let us consider the automated process of maintenance request management in a Facilities Management (FM) company which caters to B2B and B2C clients. This process is automated involving multi-agent AI collaboration. Described below, the AI agents that shall be used and their roles.

 

I.            Agents & Roles:

 

1.       Agent A (Conversational AI) - This agent interacts with the client (through omni channel platforms including, web portal, mobile app, chatbots, WhatsApp or voice), gathers all required (defined) variables, understands and categorizes the request.

2.       Agent B (Classification & Prioritization AI) - The role of this agent is to analyze the request considering all input variables and classify the urgency level (Critical / High / Moderate / Low) of the request.

3.       Agent C (Scheduling & Optimizing AI) - Based on the urgency level classified by Agent B, this agent optimizes, and schedules technicians based on their availability, skill and location and communicates available slots back to the client through Agent A.

4.       Agent D (Analyzer AI) - This agent checks, if the asset mentioned in the client request has IoT sensors, gathers log data, fetches historical maintenance records and analyzes them to validate the fault described by the client and possibly identify its root cause(s). The agent also provides fault & potential remedy insights to the technician prior to the site visit.

5.       Agent E (SLA Compliance AI) - The role of this agent is to monitor and track the workflow and escalate risks and potential SLA non-compliances proactively.

6.       Agent F (Feedback AI) - This agent captures client / technician feedback, collates overall workflow performance and feedback insights for other agents to learn and improve their performance continuously.

II.            High-level Workflow:

 

 

 

III.            Potential challenges in coordination between agents:

a.       Conflicts:

Agent C (Scheduling & Optimizing AI) could schedule over / underestimated duration prior to Agent D (Analyzer AI) validating the complaint and finding the root causes. Likewise, Agent B (Classification & Prioritization AI) could misclassify the priority prior to validation by Agent D (Analyzer AI).

b.       Time Delays:

If all technicians are busy and the company doesn't have adequate resources, Agent C (Scheduling & Optimizing AI) could fail to schedule allocation of technician for a critical job leading in delays to addressing the priority.

c.        Data Consistency:

Formats of varied input data used across the Agents must be normalized, else might lead to misinterpretation leading to incorrect agent outputs.

d.       Error Dissemination:

Logical error caused by the agent at any stage in the workflow could have a cascading effect on subsequent decisions and actions.

e.       Explainability:

Both the agents D (Analyzer AI) and E (SLA Compliance AI) must have the capability to explain the rationale behind their findings about the root cause(s) and non-compliance(s) respectively.

 

IV.            Strategies for smooth AI agents' collaboration:

a.       Central Orchestrator AI

Introduce a central workflow manager agent to ensure the workflow progresses in the right sequence with adequate information to resolve conflicts. This will help avoid time delays and avoid conflicts.

b.       Shared Memory

Build a central repository that stores real-time data along the workflow. This helps break data silos.

c.        Explainability

Agents must have the ability to record the rationale behind each action / decision. Based on the flow of work the agent must have the ability to provide real-time alerts such as "the work is delayed due to the complexity of the problem" etc.

d.       Fallback Protocol

Define clear fallback protocols such as escalation mechanisms to alert delays, disputes, SLA noncompliance and unresolved issues.

e.       Secure design:

Firmly controls the exchange of various information from knowledge base(s) and between agents. Map exchange of required information across agents. Doing this shall eliminate conflicting decisions.

 

There could be more strategies applied depending on the type of applications, architecture and technology used, considering their limitations and the application purpose.

 

 

Will take a scenario where multiple AI agents collaborate to onboard new employees in a company. Here's how it could work:

1. Customer Interaction Agent: This agent handles initial communication with the new employee, providing information about the onboarding process, answering questions, and collecting necessary personal details.
2. Document Verification Agent: Once the personal details are collected, this agent verifies the authenticity of submitted documents such as identification, qualifications, and previous employment records.
3. Record Update Agent: After verification, this agent updates the company's internal systems with the new employee's information, ensuring all records are accurate and up-to-date.

Challenges in Coordination:

• Data Consistency: Ensuring that all agents have access to the same up-to-date information to avoid discrepancies.
• Task Sequencing: Properly sequencing tasks so that each agent performs its role at the correct time without causing delays.
• Error Handling: Managing errors effectively, such as handling incomplete or incorrect information provided by the employee.

Designing Interaction:

• Centralized Database: Implement a centralized database that all agents can access and update in real-time to maintain data consistency.
• Workflow Management System: Use a workflow management system to sequence tasks and ensure each agent knows when to perform its role.
• Clear Communication Protocols: Establish clear communication protocols between agents to handle errors and exceptions smoothly. For example, if the Document Verification Agent finds an issue, it should notify the Customer Interaction Agent to request additional information from the employee.

By designing their interaction with these elements, the process remains smooth, accurate, and explainable, ensuring a seamless onboarding experience for new employees.

Scenario: loan application process in a bank. This process employs three AI agents with distinct specialties:

1. Customer Interaction Agent (CIA): Engages with the applicant via chat, collects initial details, answers queries, and gathers necessary documents.
2. Document Verification Agent (DVA): Analyzes submitted documents, checks for authenticity, and flags discrepancies.
3. Record Update Agent (RUA): Updates the bank's database with verified details and generates the loan approval/rejection notification.

Here’s how these AI agents could collaborate:

• The CIA initiates the process by collecting the applicant’s information (e.g., personal details, income proof, and identification documents) and shares the data securely with the DVA.
• The DVA verifies the authenticity of the documents, cross-references them with trusted databases (e.g., government IDs or financial institutions), and categorizes them as verified, requires human review, or rejected. The results are then passed to the RUA.
• The RUA updates the applicant's profile in the bank's database with the verified information, processes loan eligibility, and sends an appropriate response (approval, additional requirements, or rejection) to the applicant through the CIA.

---

Challenges in Coordination

1. Data Handoffs: Ensuring seamless transfer of accurate and complete data between agents without loss or duplication.
2. Conflict Resolution: Managing discrepancies, such as conflicting outputs (e.g., the CIA accepts a document that the DVA flags as suspicious).
3. Traceability: Maintaining a clear, explainable audit trail for decisions made by each agent.
4. Concurrency: Avoiding bottlenecks, especially if one agent processes data slower than the others.
5. Security & Privacy: Handling sensitive customer data securely across all interactions.

---

Designing Smooth and Accurate Interaction

1. Centralized Communication Protocol: Use a message queue system or middleware where agents exchange data in a standardized format (e.g., JSON). This ensures structured and organized communication.
2. Priority-Based Workflow: Implement a priority queue where tasks requiring immediate human intervention (e.g., flagged documents) are escalated to minimize delays.
3. Shared Knowledge Repository: Build a centralized knowledge base accessible to all agents to ensure consistent and uniform decision-making (e.g., approved document formats, regulatory compliance guidelines).
4. Error Handling Mechanisms: Design fallback protocols where ambiguous cases are escalated to human reviewers with complete logs for context.
5. Explainable AI (XAI): Integrate explain ability mechanisms that log every decision with justifications (e.g., "Document flagged as invalid due to mismatched name").
6. Parallel Processing: Allow agents to work on different parts of the same application concurrently to improve speed.
7. Data Encryption: Use robust encryption techniques for data storage and exchange to ensure privacy and security compliance.

---

By enabling modularity and embedding checks and balances, the multi-agent system can work cohesively, minimizing errors and improving customer satisfaction. 

Use-case: Enrolment of a new person for a Clinical Trial

In the pharmaceutical industry, enrolling individuals in clinical trials involves multiple steps that can be efficiently managed by specialized AI agents. Here’s a proposed implementation of AI agents for this use-case:

• User Agent: This agent interacts with users who wish to enroll in a clinical trial. It collects consent and gathers information such as medical history and current medications. Additionally, it continues to interact with users during the trial to collect ongoing information.
• Eligibility Agent: This agent cross-checks the collected information against the trial’s eligibility criteria to determine if the user qualifies.
• Documentation Agent: This agent manages the necessary documentation, ensuring all user records are complete and compliant with regulatory requirements.
• Trial Management Agent: This agent updates the clinical trial management system with the user's information and ensures all relevant databases are synchronized.

 

Challenges:

• Coordination and Communication: Ensuring seamless and secure data flow between agents without compromising user confidentiality. Manage conflicts between the agent during the process.
• Compliance: Adhering to strict regulatory requirements and maintaining data integrity.
• Error Handling: Efficiently managing discrepancies in clinical trial user information or eligibility criteria.
• Explain-ability: Providing clear and transparent justification for eligibility decisions and data processing steps.

 

Best way to Design the Process:

• Supervising Agent: Coordinates the actions of each agent, ensuring that data is routed correctly and efficiently. It controls the Definition of Done (DoD) for each agent's task. Any conflicts between Agents work or decision, Supervising agent will be responsible to make the final decision.
• Explain-ability and Auditability: Ensures each agent logs its actions with detailed explanations, providing a clear rationale for each decision and making the process transparent.
• Feedback Loops: Outcomes of individual agent tasks and feedback are passed to the Supervising Agent, allowing the system to continuously improve the enrolment process. A trial user feedback mechanism gathers input on the enrolment experience, which can be analysed to enhance patient interaction protocols.

 

Sample Flow:

1. User Agent
   • Engages with the user.
   • Collects consent and preliminary medical information.
   • Sends the data to the Supervising Agent.
2. Supervising Agent
   • Validates the information.
   • Forwards it to the Eligibility Agent.
3. Eligibility Agent
   • Checks the data against trial eligibility criteria.
   • Returns eligibility status to the Supervising Agent.
4. Supervising Agent (continued)
   • If the user is eligible, sends info to the Documentation Agent.
5. Documentation Agent
   • Manages and verifies documentation.
   • Ensures compliance with regulatory requirements.
   • Updates the Supervising Agent when complete.
6. Supervising Agent (continued)
   • Sends verified data to the Trial Management Agent.
7. Trial Management Agent
   • Updates the CTMS and other databases.
   • Confirms updates back to the Supervising Agent.
8. Supervising Agent (final step)
   • Completes the user’s enrolment.
   • Provides a summary report to:
     • Clinical Trial Coordinators
     • The User

Patient Screening in a Hospital can be one scenario where multiple AI agents may be deployed

1. Patient Interaction AI agent: Answers basic questions and gives generic info on how they may be screened. Interacts with patients and collects symptoms, medical history, and personal details.  Message - "Welcome to ABC hospital! Please provide your personal details and describe your symptoms."
2. Symptom Analysis Agent: Agent to analyse patient symptoms and suggests further tests based on referencing medical KB. Message - "Thank you for providing your information. Based on your symptoms, we suggest further tests for possible conditions XXX."
3. Medical Record Verification Agent: Verification step for existing patient records for accuracy and updates. Message - “Thank you for sharing your details. Allow me a few moments to verify the same.”
4. Appointment Scheduling Agent: Post initial analysis and verification of records, this agent schedules appointments based on doctor availability.
5. Record Update Agent: This AI agent updates the hospital's health record system with all details on a regular basis so that doctors will have the most up to date info

Challenges in Coordination

1. Data mismatch between the 5 agents deployed with accurate linkage
2. Delays in communication that could lead to low accuracy of screening process
3. Any errors in symptom analysis or record verification need to be recorded and communicated real time
4. Status updates to patients may be delayed due to gaps in agent coordination

Design corrections to create a smooth workflow

·         There should be a centralized data repository where all agents can access and update information real-time.

·         Agent interactions to be defined through standardized APIs & data formats

·         System alerts and monitoring to track the progress of the screening process.

·         If an error occurs, the system should immediately notify the relevant agent and provide a detailed explanation of the issue.

·         Empathetic agent and user-friendly interface for patients to track their screening status. This interface should provide real-time updates and explanations for any delays or issues to the patient with justification. 

 

Imagining multiple AI agents interacting for an insurance underwriting process:

 

Multiple interacting Agents:

• Sales coordination Agent: AI gathers client information post closure of sale and guides the customer through the application process
• Verification Agent: Agent that analyses the submitted application as per the underwriting norms
• Financial underwriting agent: Agent that specifically analyses the income & credit related information shared by the application and assesses the risks
• Decision making & documentation agent: On the basis of information from Verification & financial underwriting gent, approval of the application is processed and the required information is moved into the backend for data storage

 

Challenges:

• Communication/interaction challenges between agents
• Data inconsistencies in terms of formats & interpretations
• Waiting time for one agent to start post completion of previous agent's tasks
• Explanation of exceptions & decisions taken on exceptions may require human intervention

 

Design of interaction:

• Define a consistent data format and schema which is common across all agents
• Create a messaging queue to enable harmonious interaction between agents
• Implement a workflow engine to handle task sequencing & running of parallel agents
• Data maintenance: Log each agent action and decisions taken
• Defining clear error handling procedures

 

Vidhya R

 

There are multiple examples of multi-step AI or reasoning agents - such agents break complicated tasks into smaller, manageable chunks which are carried out in a logical sequence. The tasks must happen in a sequence - example - you cannot start cooking until you have the right ingredients. And while you can make adjustments while cooking if you realize that you are missing an ingredient, AI needs to have the path clearly outlined to complete the process end to end. To mention a few examples:

1. AutoQA in Call Centers: The first AI agent will convert voice into transcription using ASR, the second agent will analyze the data collected to identify key quality elements using LLM, yet another agent will utilize the identified quality elements for scoring against a yardstick, and there could be another that analyzes deviation from expected norms

2. Self Driving Cars: The car sends out LIDAR signals and the data captured from its environment is compared against the mapped data, this acts as an input for the next step in decision making of the action the car should take e.g continue driving, slowing down or coming to a complete halt, plotting a map of the trajectory for the car to follow, and sending that information back to the cars actuators

 

Such agents operate using 3 different methods:

Chain of Thought (COT)  - Problems are broken into steps, with calculations for each part of the sequence outlined

ReAct (Reasoning + Acting) - Here the AI agent uses reasoning to take real time actions, allowing complete interaction with environment and making adjustments to the process based on feedback. Each reasoning step is followed by an action step

Reflexion - Feedback loop iterations are used to enhance the model by using past output as learning inputs

 

Some of the challenges could be resource usage, unclear rules for coordination and conflict resolution. 

 

In order for a multi step agent to work in a coordinated manner, the goals must be clear, agent design is usually modular, communication protocols must be well defined. Accuracy and speed of each modular element must be assessed in order for the logical sequence to work well. Communication protocols must outline how the agent shares information, coordinates actions and resolves conflicts. There must be governance system for identifying and resolving issues within the system. 

 

One of the other ways of ensuring coordination is making the flow sequential to avoid conflict. 

 

 

Scenario: Multi-agent collaboration in Vendor Registration in Procurement Process

 

In Procurement process, vendor registration is a key activity that includes multiple steps requiring accuracy, compliance checks and system integration. In the proposed solution, 3 specialized AI agents will collaborate to accomplish the required task.

 

Agent 1: Vendor Integration Agent

Responsibility: Collect initial vendor data through a secure web portal or a chatbot.

Tasks:

• Gathers vendor profile, contact info, product categories and banking details.

• Guides vendor on required documents (e.g., GST, PAN, ISO certification, etc)

• Ensures data is captured in a structured format and confirms completion.

 

Agent 2: Compliance & Document Verifier Agent

Responsibility: Validates submitted documents for authenticity and compliance.

Tasks:

• It will use OCR and ML to extract data from uploaded document.

• Verifies legal/tax compliance (e.g., GST registration, etc.)

• Checks documents validity, expiry dates and flag discrepancies.

 

Agent 3: ERP Integrator & Workflow Trigger Agent

Responsibility: Will be responsible for updating the internal procurement system and initiates approval workflows.

Tasks:

• Creates and updates Approved Vendor List (AVL) in ERP.

• Notifies Procurement, Finance and Legal teams for approvals.

• Triggers automatic onboarding emails and vendor codes.

 

Challenges in Coordination

• Standardized & Consistent Data: Ensuring clean hand-offs between agents without data loss.

• Status updates: Keeping all agents aware of process status to avoid rework.

• Error handling: Managing incomplete/invalid documents and re-engaging the vendor smoothly.

 

Design for Smooth, Accurate, Explainable Interaction

• Workflow Design: Workflow to be designed in such a way that all 3 agents work sequentially (1-->2-->3)

• Central Database: Central vendor registration databased shared across agents.

• Explainability: Each Agent will maintain its logs for traceability, audit and compliance.

• Fallback Mechanism: If AI agent is unable to handle the query, it should flag it for manual intervention.

 

The above design will ensure a smooth, accurate and explainable interaction in this agentic approach.

If AI agents worked as one team it would lead to smarter planning, data analysis and reporting. Multiple AI agents handling different departments like Operations, Finance, Quality, Risk leading to lesser human intervention. However, there could be some negatives outcomes like coordination issues between AI agents if not programmed correctly ,conflicting goals.

Multiple AI agents are networks of AI agents that collaborate with each other and have specialized roles to solve more complex problems, adapt in real time, and coordinate actions across systems and workflows. Each agent performs a distinct role and adapts its behavior based on evolving inputs. In a Contact Centre environment we could incorporate multiple agents in relation to Energy Company, were one agent maybe a Chat Bot to record a process Client chats and information, another agent maybe introduced for the live calls. A third agent can be introduced to capture details from the other agents and extract information around customer interactions, account information and customer sentiment etc. The challenge may arise when customers may use short text language and voice agent may struggle with accent and different languages.

Going along with my previous example of healthcare, there could be a scenario of three different AI agents helping doctors with their patients. These agents separately may take of different things (test results history, medical records, possible treatments etc) but if one fails, this hampers the work of the others as well. A proper BRD must be designed to properly outline the objectives and for utmost clarity. Ethical considerations must also be considered with proper resource management. Potential risks must also be assessed beforehand. 

What If AI Agents Worked as a Team?

Use Case: AI-Driven Procure-to-Pay Process in SAP

Scenario Overview:

A top notch FMCG company, NutriChufa Group uses an AI-powered SAP Procure to Pay process system, where different AI agents, each with a distinct specialization, collaborate to manage the end-to-end procurement lifecycle of the company.

Roles of the AI Agents and Handoff.

1. Agent A – RequestBot (Procurement Request Assistant)
•    Role: Handles employee procurement requests.
•    Capabilities:
i.    Converses with end-users to capture purchase needs.
ii.    Classifies the request (e.g., materials vs services).
iii.    Checks budgets and preferred suppliers in SAP.
•    Handoff: Sends structured purchase requisition (PR) data to the next agent.

2. Agent B – VendorCheckAI (Supplier Verification, Negotiator & Onboarding)
•    Role: Verifies vendors and checks compliance and risk.
•    Capabilities:
i.    Flags high-risk suppliers, according to the company’s audit requirements.
ii.    Checks for duplicates, if the vendor already exists in the system.
iii.    Selects the best fitting vendor from the requests for quotation (RFQ) documents submitted over emails/chats with suppliers.
iv.    Initiates onboarding if it is a completely new vendor and passes audit check.
•    Handoff: Updates vendor master data and passes verified PR to BuyerBot.

3. Agent C – BuyerBot (Purchase Order Creator)
•    Role: Creates POs and communicates with suppliers.
•    Capabilities:
i.    Converts PRs into POs.
ii.    Logs decisions with audit trails in SAP.
•    Handoff: Sends final PO to DocuVerifyBot and notifies suppliers.

4. Agent D – DocuVerifyBot (Document Compliance)
•    Role: Validates invoices and delivery notes.
•    Capabilities:
i.    Uses AI to read delivery documents and check if they match the purchase order and invoice
ii.    Detects anomalies (e.g., quantity mismatch, price variance, duplicate invoices).
•    Handoff: Approves matched invoices and sends them to FinanceBot.

5. Agent E – FinanceBot (Payment & Record Update)
•    Role: Handles payment execution and SAP ledger updates.
•    Capabilities:
i.    Triggers payment runs based on invoice approval.
ii.    Reconciles accounts.
iii.    Generates reports for audit and compliance.

Challenges in Coordination + Conflict Resolution Mechanisms

 1. Challenge: Timing Conflicts or Deadlocks
•    Example: BuyerBot can’t create a PO because VendorCheckAI hasn't completed its compliance check.
•    Solution: Introduce a central coordinator that ensures tasks run in the correct order and handles delays by triggering backups or involving a human when needed.

2. Challenge: Conflicting Data or Decisions
•    Example: DocuVerifyBot detects invoice inconsistency that RequestBot missed (e.g., price variance).
•    Solution: Set up a shared system where agents record their actions and reasons. If there's a conflict, they can review the logs and either resolve it together or pass it to a human for review.

 3. Challenge: Difficulty in understanding why some AI decisions are made.
•    Solution: Use explainable AI tools that show clear reasons for each action, with details shown directly in SAP (e.g., “Price changed due to contract clause with supplier”).

This can work in multiple domains. In Healthcare, we can create such team of AI agents who will do the following:

• Patient Interaction Agent: Which will interact with patients initially. This should be enough to answer basic queries and collect basic info.
• Document Verification Agent: This will verify patients' documents.
• Record Update Agent: It will update hospital's EHR, ensuring data accuracy and storage.

Communication can be a challenge, which can be eradicated by implementing a centralized communication protocol using APIs or message queues.

 

Maintaining data accuracy and consistency can be another challenge. Using a shared database and implementing validation checks in each step should be sufficient.

For any successful task, Mutiple units need to be performing as a one to achieve the goal. 

 

Taking scenario of Hospital setup:

 

Different AI agent would be required to perform individual task

• To make repository for the patients: OPD/IPD/ ICU etc.
• Supply management of medicines/surgical items/ manpower attendance/ Doctor availability.
• payment record
• Insurance record
• inward/ outward record
• Medical record

These are few areas where automation can help to bring efficient and smooth routine operations to run the hospital.

Collaborations of these individual unit activities will bring the best of out come. Any missed activity can halt the routine operations. To achieve completeness of the activity, feed back mechanism, real time solution activity, other plan 
(plan B: which may not be as active) should be available. 

 

Interesting examples of usage of multi-agent setups.

 

Vinod has provided the most comprehensive answer and hence has been selected as the best. Well Done.

 

Answers from Divya Iyer, Hardik Joshi, Rupinder Narang are also an interesting read.