《印孚瑟斯（Infosys）：2025技术导航：代理型AI在企业中的应用Playbook（英文版）（70页）.pdf》由会员分享，可在线阅读，更多相关《印孚瑟斯（Infosys）：2025技术导航：代理型AI在企业中的应用Playbook（英文版）（70页）.pdf（70页珍藏版）》请在三个皮匠报告上搜索。

1、TECH NAVIGATOR:AGENTIC ENTERPRISE AI PLAYBOOKTech Navigator:Agentic Enterprise AI Playbook|3External Document 2025 Infosys Limited Knowledge InstituteCONTENTSExecutive summary 4Agentic AI systems 6Agentic AI architecture and blueprints 18AgentOps and agentic life cycle management 36Advanced agents 4

2、8Responsible agentic AI 58The transformative potential of artificial intelligence(AI)has captivated everyone for decades,evolving from a futuristic concept to the pinnacle of the hype cycle.Now,AI is on the verge of its next level of evolution:Agentic AI.Unlike traditional AI,which enhances human ta

3、sks through insights and automation,agentic AI redefines both expectations and capabilities.It goes beyond supporting existing workflows,instead reimagining and redesigning processes from the ground up to create truly AI-native systems.In its early stages,generative AI primarily served as a tool to

4、enhance efficiency and accuracy for individuals and teams.However,as AI models have matured and become increasingly commoditized,the focus is shifting from augmentation to reinvention transcending incremental improvements.This transformation echoes past technological revolutions,such as the shift to

5、 digital-first processes during the rise of digital transformation.Just as businesses once reengineered processes and business models,rather than merely digitizing analog workflows,agentic AI demands a similar rethinking of processes to unlock its full potential.Achieving this requires deep integrat

6、ion of cognitive capabilities into process engineering and experience design not just at the periphery of business but at its very core.The potential of agentic AI extends far beyond its early applications in customer service and IT operations.It has the power to revolutionize mission-critical domai

7、ns,such as customer onboarding and credit decisioning in banking;supply chain management in retail,consumer goods,Executive summaryKnowledge Institute4|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Tech Navigator:Agentic Enterprise AI Playbook|5External Document

8、 2025 Infosys Limited Knowledge Institutelogistics,and manufacturing;sales and marketing life cycles;and product design and development.By embedding intelligent agents across these functions,organizations can unlock higher levels of efficiency,adaptability,and innovation,fundamentally transforming h

9、ow they operate and compete in an AI-driven world.Yet this journey is not without its challenges the inevitable barriers,frustrations,and setbacks that mark all progress.The technology is still evolving,and enterprises face critical questions about how to strategically implement agentic AI at scale.

10、How should organizations architect systems that can accommodate hundreds or even thousands of intelligent agents?Which platforms and models should they adopt?How can they ensure interoperability among diverse AI systems while maintaining flexibility for future advancements?This report explores these

11、 critical questions through a pragmatic lens,cutting through the hype surrounding agentic AI to provide actionable insights for enterprise leaders.It offers practical guidance on navigating this complex landscape and implementing AI-driven strategies effectively.By adopting a poly-AI and poly-agent

12、architecture one that integrates the best models,providers and agents while still ensuring interoperability organizations can stay agile,future-proof their investments,and gain an early lead on competitors.AGENTIC AI SYSTEMSArtificial intelligence(AI)has steadily surpassed human cognition in fields

13、once thought to be uniquely ours from image recognition and speech processing to algorithm design.Yet,despite its astonishing power,computer scientists are still trying to make AI behave more like humans:Intuitive,adaptable,and independent.People are naturally skilled at recognizing patterns and mak

14、ing sense of chaos,even when information is disorganized or incomplete.However,comprehension doesnt always strike immediately:We rely on books,online searches,and the wisdom of others to make more informed decisions that lead to better results.Generative AI follows a similar trajectory by retrieving

15、 information,generating insights,and sometimes taking actions whether thats analyzing a customers purchase history to recommend tailored products or automating essential tasks,such as sending emails and processing transactions.The expectation that AI will deliver highly personalized experiences stem

16、s from the level of tailored care humans have come to expect across the services they consume.Businesses have refined their hyperpersonalization efforts,leveraging streamlined communication and interaction history to enhance customer engagement.Basic and routine inquiries were efficiently managed by

17、 chatbots that respond quickly and reduce human intervention.However,while early chatbot implementations served their purpose,their lack of context awareness and empathy soon became clear,exposing a gap in user experience.The shift from static 6|Tech Navigator:Agentic Enterprise AI PlaybookExternal

18、Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|7External Document 2025 Infosys Limited Knowledge Instituteacknowledgments to dynamic,adaptive responses capable of inheriting context from the industry domain and acting on insights represents a significa

19、nt opportunity for AI advancement.Developing systems that can understand,adapt,and respond intelligently in real time is the next frontier in AI-driven personalization.When AI gains the ability to reason,coordinate tasks,and act with purpose,it transcends being a mere tool and becomes an agent.And a

20、s companies seek more adaptable,self-sufficient solutions,these emerging AI agents will evolve from technology tools into indispensable business partners.The evolution of AI agentsAgentic AIs arrival is accelerating reminiscent of the rapid rise of generative AI just over two years ago.Gartner consi

21、ders agentic AI to be the top technology trend for 2025.And Deloitte forecasts that by 2025,one-quarter of companies that use generative AI will initiate agentic AI pilots or proof-of-concept projects,with adoption increasing to half by 2027.The consulting firm also projects that in certain industri

22、es and use cases,agentic AI applications could begin integrating into existing workflows in late2025.At their core,agents are autonomous software entities that use a simple yet potent operational loop:They observe their environment via sensors,process this input,and use either mechanical or digital

23、actuators to change their environments and achieve Knowledge Institute8|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited specific objectives.This traditional sense-plan-act cycle remains as relevant in the AI era as when it was initially conceptualized for robotics

24、.As agents evolve through each stage of enterprise expansion and adoption,their capabilities progressively advance.Traditional robotic process automation relies on rules-based or instruction-driven configurations,operating strictly within predefined parameters.These systems respond only to programme

25、d rules,lacking adaptability or contextual reasoning.In contrast,agentic AI systems introduce adaptive,autonomous decision-making,bridging the gap between rigid automation and human-like reasoning.With minimal to no human intervention,AI agents perform specific tasks using capabilities that are inte

26、grated across multiple layers in the overall agentic system.Now,the rise of large language models(LLMs)has transformed the agentic world by acting as increasingly sophisticated brains behind the agents.Agents and LLMs function as partners with agents orchestrating LLMs as needed.However,the agent re

27、tains control over when to call and execute an LLM,ensuring structured,task-driven interactions that align with the systems objectives and security requirements.From simple automation to the potential for artificial general intelligence(AGI),AI has steadily progressed from a world of rules-based act

28、ions to greater independence(Figure 1).Tier 0:Simple automation:Localized automation applies to a specific segment of a broader process.It lacks agentic behavior and operates through rules-based,deterministic automation.A common example is component test automation,where scripts execute predefined t

29、asks.Figure 1.The progress of AI agentsSource:InfosysSimpleautomationRobotic processautomationAI agent twinRefined AI agent with reflectionCompleteautonomy AGI or ASIElementary AIaugmentationTier 1:Robotic process automation:An advancement over basic automation,this approach extends beyond point sol

30、utions to cover entire process segments using predetermined rules-based logic.While AI or agentic behavior remains minimal to nonexistent,it enhances efficiency in tasks such as screen scraping or automated form completion.Tier 2:Elementary AI augmentation:This stage introduces the first meaningful

31、presence of agentic AI,offering an opportunity to replace tier 1 automation,particularly in areas requiring human oversight.By leveraging a language model,agents enable intelligent interactions while maintaining a limited,yet impactful,role in automation.Examples include sentiment analysis or ticket

32、 data labelling,where agents classify information into appropriate categories.Tier 3:AI agent twins:These systems function as digital twins to users,interpreting intent and autonomously taking action to achieve specific outcomes.Some of the best-known instances of AI agent twins include GitHub Copil

33、ot and Microsoft 365 Copilot,which assist users by generating code,automating tasks,and enhancing productivity through intelligent decision-making.Tier 4:Refined AI agents with reflection:These systems represent an advanced class of AI agents that many organizations are eager to implement.They can d

34、ecompose tasks from a given objective,formulate plans to achieve the intended outcome,and analyze results to adapt their approach in response Tech Navigator:Agentic Enterprise AI Playbook|9External Document 2025 Infosys Limited Knowledge Instituteto failures or unexpected events through complex reas

35、oning sequences.This tier can be seen in credit decision-making systems,where agents process loan applications,extract and analyze documents,and match them against stored information to ensure accuracy and compliance.Tier 5:Complete autonomy,AGI or artificial superintelligence:When this tier arrives

36、 in the future,these agents will possess the capability to conduct entirely original research,independently reason through complex problems,and develop innovative solutions beyond their initial training data.With advanced logical reasoning and adaptive learning,they will continuously acquire new ski

37、lls,refine their methodologies,and tackle previously unsolved challenges,pushing the boundaries of AI-driven discovery and problem-solving.Although tier 5 is still somewhere in the future with optimistic predictions ranging from 2026 to 2029 AI has already shifted possibilities and expectations.This

38、 technology has evolved from rules-based automation to intelligent,self-improving systems capable of operating independently,dynamically responding to their environments,and optimizing decision-making in real time.Organizations can now deploy these agentic solutions across domains such as software d

39、evelopment,IT operations,and customer care to drive unprecedented efficiency and adaptability.The blueprint for agentic AI Business leaders recognize the value AI has already delivered and see its potential 10|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowle

40、dge InstituteTech Navigator:Agentic Enterprise AI Playbook|11External Document 2025 Infosys Limited Knowledge Instituteto drive further transformation.However,fully capitalizing on AI-driven opportunities requires a strong foundation in both architecture and operational layers of advanced AI systems

41、.The blueprint for agentic AI defines the core components and processes that drive their functionality.By analyzing this framework,it becomes clear how user inputs,orchestration engines,planning modules,reasoning layers,utilities,memory,integration interfaces,human oversight,and pretrained language

42、models work together to enable intelligent and autonomous agent behavior.The following are the core layers of an agentic system and an examination of how it operates(Figure 2).User input or event trigger:A predefined instruction set or event combination that activates an agent,prompting it to initia

43、te Figure 2.AI agent frameworkUser input(Multimodal)Event trigger-schedulerGoal-based orchestration engineHuman-in-the-loopLLM Ops-cache-telemetryPre-trained models(GPT-Claude-Gemini)Fine-tuned SLM/LLM(Phi 3.5-fine-tuned Llama)Public infrastructure:Azure,AWS,Google CloudInfrastructure hardware layer

44、:NVIDIA,Groq,Cerebras,SambaNovaResponsible AIPlanning moduleEvaluationTool or skillsMemory moduleIntegrationReasoning layerSecurity andcomplianceSub-task deductionBenchmarkCustom skillsLong-term memoryAPI integrationLoad balanced orversion controlShort-term memoryEnterprise toolsTasks and instructio

45、nsRole-based access controlPrivacyRegulatory complianceSensory inputKnowledge graphFeedback Human evaluationValidation modulesReflection modulesRanking modulesSource:InfosysKnowledge Institute12|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited actions or processes

46、based on specified conditions.Goal-based orchestration engine:This serves as the foundation for decision-making and prioritization,orchestrating various actions to align with a systems goals.Planning module:-Reasoning layer:Algorithms designed to analyze inputs,develop strategies,and break tasks int

47、o subtasks to ensure efficient execution.-Tools and skills:Reusable software modules activated to achieve specific goals,such as document digitization,optical character recognition,and PDF generation.-Memory:Module maintains a record of current and past interactions,tool usage,and learned behavior,e

48、nabling context-aware planning.-Integration module:Module interfaces with the outside world,often through APIs.-Human in the loop:Humans approve certain critical decisions recommended by agents.Language model:Pretrained or fine-tuned models deployed responsibly on public or private cloud infrastruct

49、ure.Security and compliance:This layer ensures strict compliance with data security,privacy,and ethical standards,making sure the agent aligns with responsible AI principles and regulatory requirements.Evaluation layer:This layer measures the effectiveness and the efficiency of the agent.This evalua

50、tion layer also provides feedback to the planning module for continuous improvement.Planning moduleThe planning module,often considered to be the agents“brain,”is the central orchestrator of decision-making,task prioritization,and adaptive execution in an agentic system.This planning module also pla

51、ys a pivotal role in transforming reactive automation into proactive,intelligent decision-making.Functioning alongside the orchestration layer,the planning module interprets input triggers and breaks down objectives into achievable goals.The orchestration layer refines these objectives into a struct

52、ured sequence of subtasks.It also oversees execution,engages reflection modules for validation,and generates user responses.Together,these layers form a continuous cycle that guides agent behavior through the plan,act,and reflect stages.This process involves gathering data from internal and external

53、 sources,analyzing it using predefined logic,reasoning frameworks,or learned patterns,and determining the optimal next step based on the current state and objectives.By integrating these components,agentic AI systems achieve a high degree of autonomy,adaptability,and operational intelligence.The lay

54、ers that enable planning at scale include the following:Reasoning layerThis capability shapes interactions with the language model by refining prompts and evaluating or ranking responses.It plays a crucial role in identifying specific milestones within subtasks,processing environmental inputs,and ca

55、pturing sensory data.Additionally,this reasoning layer leverages knowledge bases and reference frameworks,such as knowledge graphs,to narrow down results and enhance accuracy.Memory moduleThe ability to retain context from both current and past interactions,while continuously learning from ongoing a

56、nd long-term activities,provides the reasoning layer with the necessary background and feedback for effective operations.Tools,skills,and integration layerThis layer compensates for the inherent limitations of language models in directly interacting with the real world.It achieves this by leveraging

57、 reusable software components,such as PDF generation and document digitization,to accomplish specific goals.Additionally,this layer enables integration with specialized systems by accessing external web APIs,including retrieval-augmented generation(RAG)frameworks.Tech Navigator:Agentic Enterprise AI

58、 Playbook|13External Document 2025 Infosys Limited Knowledge InstituteLanguage modelThe language model serves as the central decision-maker,consisting of one or more models that employ reasoning frameworks such as ReAct or chain-of-thought.These models can be general,multimodal,or fine-tuned for spe

59、cific business objectives.For optimal performance,the chosen model should align with target requirements and be trained on data relevant to the integrated tools.While the model is typically not trained on the agents specific configuration,its decision-making accuracy can be enhanced by providing con

60、textual examples that highlight the agents capabilities.This approach ensures more precise and context-aware outputs.The planning and reasoning cycle continues iteratively until the goal is achieved,or a stopping condition is met.The complexity of orchestration depends on the agent and task,varying

61、from simple calculations to advanced logic,such as chained reasoning or machine learning algorithms.Why do we need AI agents?The rise of agentic AI is fueled by the growing complexity of modern businesses and the demand for intelligent automation.Agentic AI overcomes the limitations of traditional a

62、utomation by integrating adaptability,contextual awareness,and autonomous decision-making.Unlike conventional systems,which struggle with unstructured challenges and evolving conditions,agentic AI processes ambiguous data through continuous learning and real-time analysis to optimize workflows.This

63、shift from rules-based automation to context-driven intelligence allows organizations to service increasingly complex demands.The following features make agentic AI particularly relevant for the modern enterprise.Operational efficiency:Integrates cross-functional tools,skills,and APIs to detect inef

64、ficiencies and autonomously implement improvements,ensuring process agility.14|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|15External Document 2025 Infosys Limited Knowledge Institute Ability to

65、scale rapidly:Unlike the resource-intensive traditional systems,agentic AI enhances capabilities without requiring proportional staffing increases by taking advantage of self-optimizing architectures.Dynamic problem-solving:The ability to manage unstructured tasks through contextual reasoning allows

66、 agents to generate intelligent responses to complex requests without relying on predefined rules.When is agentic AI the best way?Frameworks play a crucial role in the effective deployment and use of agentic AI,ensuring that agents operate efficiently,ethically,and in alignment with business objecti

67、ves.And with the limited number of real-world case studies,these frameworks serve as blueprints that can guide organizations that want to move quickly and decisively rather than waiting for competitors to define the way forward.At Infosys,we use the SCOPE framework to assess the suitability of agent

68、ic AI for addressing business challenges.Each parameter is then scored based on its impact.Strategic alignment Does the opportunity align with long-term organizational goals?Are the workflows we use today largely manual?Complexity of task Is the task planning intensive and multistep?Is the planning

69、largely deterministic,and if so,are the rules expected to increase exponentially to comprehensively cover the possibilities?Operational environment Is there a need for real-time action or decision-making,or a need to adapt to changing circumstances?Are there multiple data streams to integrate with a

70、nd will the need grow to integrate with more systems?Performance requirements Are there latency constraints or impacts that affect the cost?What are the ethical and security constraints Can the systems actions be adequately governed,controlled,and monitored?Are there ethical risks with autonomous de

71、cision-making or recommendations?In addition to insights from a robust framework,enterprises should closely examine the growing number of agentic AI use cases that are applicable to complex,real-world scenarios.Below are the most promising uses for agentic AI,although not all have been tested at ent

72、erprise scale.(Figure 3).Remember that agentic AI is an emerging technology that is still building a track record.While AI adoption is accelerating,its enterprisewide deployment remains Knowledge Institute16|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Figure 3

73、.Agentic AI use cases by industrylimited,with many organizations still in the experimental phase.Relatively few have successfully implemented and scaled agentic AI across specific operations.Below are examples of organizations that have made significant progress in integrating agentic AI into their

74、workflows.In one example,CommBank leverages agentic AI to process about 15,000 payment disputes every day.Customers can describe their issues through an AI-assisted channel,which autonomously verifies eligibility criteria and lodges disputes without requiring manual intervention.This automation enha

75、nces efficiency,reduces processing time,and improves customer satisfaction.Financial servicesPeriodic risk review and renewalsRisk and compliance reporting Watchlist reporting(adverse news,sanctions,PEP)HSBC has built NOLA2.0,a cloud native solution on GCP to modernize credit risk management in comp

76、liance with Basel III(automated regulatory adaptation with real-time capital allocation).High tech Driverless cars route adaptationWaymos driverless cars autonomously adapt routes for safety and efficiency by analyzing sensor data(LIDAR,cameras)to navigate and avoid obstacles in real time.Logistics

77、and supply chainDynamic fleet route managementPando.ai rerouted 5,000-plus containers during the 2024 Panama Canal drought,avoiding$12 million in delays.ManufacturingOptimizing engineering systems workflowsSiemens developed a multiagent system to make engineering workflows more efficient,with the ag

78、ents acting as system architects and requirements engineers.TelecomContinuous network monitoring and optimizationDynamically allocates bandwidth and resolves congestion by autonomously adjusting parameters.Nokia has announced agentic AI in autonomous network management.ServiceNow has introduced supp

79、ort for agentic AI for better network management.IndustryUse casesCapabilitySource:Infosys Knowledge InstituteTelecom company Telenor has deployed conversational agentic AI agents to autonomously handle customer queries,resolve issues,and facilitate sales.This implementation led to a 20%increase in

80、customer satisfaction and growth in revenue of up to 15%within the first year.Talkdesk introduced agentic AI-powered conversational agents for retail customer service,enabling autonomous management of complex tasks such as order updates,address modifications,and customer routing to in-store speciali

81、sts.These AI agents provide 24/7 support while delivering hyperpersonalized experiences,enhancing both efficiency and customer engagement.Levi Strauss has implemented agentic AI for granular demand forecasting across its supply chain.The system autonomously adjusts inventory levels based on real-tim

82、e demand signals,ensuring that the right products are available in the right locations while minimizing waste and inefficiency.Redefining AIThe emergence and maturation of agentic AI marks a defining moment in the evolution of intelligence one that transcends rigid algorithms and human-dependent dec

83、ision-making.Unlike its predecessors,which followed predefined rules,agentic AI is self-directed,adaptive,and deeply contextual,capable of understanding real-time data,reasoning dynamically,and acting with intent.This shift is more than an incremental improvement;it redefines how technology engages

84、with the world.No longer bound by static programming,AI is evolving into a thinking entity,a collaborator capable of navigating complexity,optimizing processes,and unlocking vast new possibilities.Ultimately,the rise of agentic AI represents a transformative leap in the field of AI.By enabling machi

85、nes to think,learn,and act independently,we are not only enhancing the capabilities of AI but also redefining the relationship between humans and technology.As businesses continue to explore and implement these advanced systems,the future of AI promises to be more dynamic,intelligent,and impactful t

86、han ever before.Tech Navigator:Agentic Enterprise AI Playbook|17External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute18|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited AGENTIC AI ARCHITECTURE AND BLUEPRINTSArtificial intelligence(AI)is evol

87、ving into more intuitive and independent systems with the advent of agentic AI,which allows for autonomous decision-making and real-time responsiveness.This year,many companies will begin exploring agentic AI pilots,with widespread adoption anticipated by 2027.This report delves into the transformat

88、ive potential of agentic AI,the roles of AI agents,and the importance of understanding agentic architecture for business leaders.Agentic AI is transforming how businesses automate decision-making and streamline complex workflows.However,simply deploying AI agents does not guarantee success.Agent cap

89、abilities vary and do not function in isolation;they operate alongside humans and are embedded within enterprise business processes across the value chain.As AI capabilities advance,business leaders will develop a deeper understanding of the intricacies at play:the distinct roles of AI agents,the la

90、yers of agentic architecture,and the tools and frameworks that facilitate integration into existing IT ecosystems.This will be a strategic imperative,not just a technical curiosity.This deeper knowledge allows business leaders to make more informed decisions about technology investments,resource all

91、ocation,and process optimization enhancing efficiency and strengthening their competitive advantages.A well-structured agentic system ensures that AI decision-making is transparent,adaptable,collaborative with human teams,and strategically aligned with business objectives.18|Tech Navigator:Agentic E

92、nterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|19External Document 2025 Infosys Limited Knowledge InstituteTypes of AI agentsAt the core of agentic AI architecture is the sense-plan-act cycle,where AI agents interpret data

93、 from their environment,formulate a plan,and autonomously execute tasks.This architecture establishes the fundamental building blocks required for AI agents to operate effectively,whether as a single agent or as a multiagent system.Both approaches leverage machine learning and computational reasonin

94、g,but the choice of agent type significantly affects whether the AI system can reach its full potential.Different agents are required for varying levels of task complexity,ranging from simple reactive systems to highly sophisticated learning agents.Selecting the appropriate agent type ensures it can

95、 handle the specific task,whether it involves basic automation or complex decision-making.Understanding an agents capabilities helps set realistic expectations about what it can achieve,preventing overreliance on its abilities and potential misuse.Knowing the type of agent allows developers and arch

96、itects to:Select the best algorithms and learning methods.Determine appropriate data collection and preparation processes.Choose architecture that enables optimal performance and integrates with existing systems.By aligning the agent type with its intended functionality,organizations can maximize Kn

97、owledge Institute20|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited AI performance,efficiency,and reliability while avoiding unnecessary complexity or operational risks.(Figure 1.)Reflex-based agentsReflex-based agents use a predefined set of rules,or reflexes,mak

98、ing them the most basic type.These agents are programmed to perform specific actions whenever certain conditions are met,without requiring advanced reasoning or learning.In manufacturing,reflex-based agents can play a critical role in quality control.These agents use cameras to capture images of com

99、ponents as they move through the production line.Using predefined criteria,the agents analyze images in real time and instantly flag defects.Their actions are binary,either accepting or rejecting components based solely on the visual input.For example,Intel uses reflex-based agents in its chip manuf

100、acturing plants to improve precision and efficiency in quality control.Model-based reflex agentsModel-based reflex agents are a specialized type of reflex agent that maintains an internal model to track environmental changes.This model is continuously updated as new information arrives,allowing the

101、agent to make decisions based not only on its reflexes but also on past experiences and current state.A smart irrigation system is a prime example of a model-based reflex agent.Rachio smart sprinklers integrate weather forecasting data and soil moisture models,learning from local weather patterns wh

102、ile considering soil type,plant type,sun exposure,and slope to create predictive,data-driven watering schedules.Goal-based agentsGoal-based agents integrate an internal model of the world with a defined goal or set of goals.Before acting,they plan and search for action sequences that will help them

103、achieve their objectives.This deliberate,strategic approach makes them more effective than reflex-based or model-based agents.These agents are widely used in warehouse automation to optimize order fulfillment.In automated fulfillment centers,goal-based agents operate with specific objectives,such as

104、 completing an order within a set timeframe.To achieve this,agents plan the most efficient picking routes through the warehouse while coordinating with other robots to prevent conflicts and ensure smooth operations.A notable example is Amazons Kiva robots,which efficiently manage inventory movement

105、and order fulfillment in the retailers automated warehouses.Utility-based agentsUtility-based agents select the optimal sequence of actions that allow them to reach their goals and maximize utility or reward.Utility is determined by a function that assigns a utility value a metric that measures the

106、usefulness of an action or how“satisfied”it will make the agent to each scenario based on fixed criteria.20|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|21External Document 2025 Infosys Limited Kn

107、owledge InstituteThese agents play a crucial role in portfolio management,particularly in automated trading and investment optimization.Investment companies deploy utility-based agents to navigate the complexities of financial markets,evaluating factors such as risk,return,market conditions,and clie

108、nt preferences.Using sophisticated utility functions,these agents balance competing objectives,such as growth and stability,to make optimized investment decisions.A prime example is BlackRocks Aladdin platform,which leverages utility-based agents to enhance investment management and performance.Obje

109、ct-centric or curious agentsAn object-centric,or“curious”,agent focuses on understanding and interacting with individual objects in its environment rather than simply perceiving the overall scene.It prioritizes detailed information about specific objects,their properties,and their relationships with

110、 one another to inform decision-making and actions.These agents are used frequently in the healthcare industry,particularly in radiology departments.Object-centric agents actively analyze medical images,detecting and learning new patterns across various imaging types,such as X-rays,MRIs,and CT scans

111、.By continuously learning from new cases,these agents improve diagnostic accuracy over time.A notable example is Arterys AI platform,which uses object-centric agents to enhance medical imaging analysis,ultimately supporting better clinical outcomes.Citizen-centric autonomous agentsCitizen-centric au

112、tonomous agents use advanced AI models to independently manage routine personal and professional tasks.These agents analyze user objectives,break them into actionable steps,and execute workflows across applications without manual intervention.For instance,OpenAIs Operator agent can autonomously mana

113、ge email sorting,Tech Navigator:Agentic Enterprise AI Playbook|21External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute22|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited appointment scheduling,and online purchases by interacting with web int

114、erfaces through screenshots and keystroke simulations.Similarly,Googles Agentspace enables users to automate document processing,such as extracting key details from PDF invoices and populating expense reports in real time.Anthropics Claude Computer Use is an AI agent that is good at autonomously per

115、forming desktop tasks while Operator is good at browser-based tasks.By integrating natural language understanding with cross-platform interoperability,these tools allow individuals to delegate repetitive digital tasks.Figure 1.Each agent type focuses on a different part of the sense-plan-act cycleAg

116、ent typeSensePlanActCharacteristicsand use casesReflex-basedModel-basedGoal-basedPre-defined rules triggered by conditionsMaintains internal model;updated with perceptionsGathers information relevant to specific goalsActions based on current model and historySearches for action sequences to achieve

117、goalsMinimal planning:immediate actions based on rulesBinary actions:accept or reject inputActions depend on model predictionsExecutes actions aligned with goalsBasic agents effective for simple tasks such as defect detection in manufacturingMore complex;deployed for smart grid optimizationDesigned

118、for efficiency;common in warehouse automationUtility-basedEvaluates input against a utility functionPlans actions to maximize utility based on criteriaChooses actions that optimize overall utilityMaximizes outcomes;used in finance for portfolio managementObject-centricor“curious”Understands individu

119、al objects and their propertiesPlans actions based on knowledge of specific objectsInteracts with objects for learning or analysisEmphasizes object relationships;used in medical imaging analysisCitizen agentsMultimodal inputs(text,screenshots,user goals)Decomposes objectives into steps with LLM-base

120、d task prioritizationInterfaces with web or app GUIs and document workflows via API integrationAutonomous execution of personal or civic tasks(scheduling,form processing,service access)with cross-platform interoperability and governance safeguardsSource:Infosys22|Tech Navigator:Agentic Enterprise AI

121、 PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|23External Document 2025 Infosys Limited Knowledge InstituteHow to build an agentic system advancements,developers can expand the capabilities of agentic systems,unlocking new possibiliti

122、es for enterprise applications.The complexity of a task determines whether a single-agent or multiagent architecture is the best approach.Single-agent architecture is ideal for tasks that follow clear,well-defined steps and have minimal tool requirements.In contrast,multiagent architecture is better

123、 suited for complex tasks that require collaboration,feedback,and the parallel execution of subtasks.This report focuses on multiagent architecture.Enterprise systems are inherently deterministic,designed to execute specific,rule-based tasks,even when those tasks involve complex business logic.Howev

124、er,integrating AI agents into these systems introduces an element of nondeterminism.Unlike traditional enterprise software,AI agents plan action sequences dynamically,adapting their behavior as inputs evolve through learning.Despite this adaptability,their actions must remain within the boundaries o

125、f standard operating procedures to ensure consistent and reliable outcomes.For developers and architects,understanding the interaction between deterministic enterprise systems and AI agents is critical when designing enterprise-grade applications.Organizations must carefully evaluate task complexity

126、,computational resources,and operational constraints to develop an effective and scalable AI architecture that balances flexibility with control while ensuring reliable performance.Agentic AI architectureWhen developing agentic AI architecture,organizations must determine the number of agents requir

127、ed and the roles they will play.Agentic systems follow two fundamental configurations:single-agent and multiagent architectures.Both leverage machine learning models and computational methods to execute the sense-plan-act cycle,enabling intelligent decision-making and automation.By combining establi

128、shed principles with AI Tech Navigator:Agentic Enterprise AI Playbook|23External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute24|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited After deciding that a multiagent system is needed,the agents mus

129、t then be classified.The role-based agents are labelled based on their jobs,such as advisor,coder,reviewer,or tester.The interaction model-based reflex agents are categorized by their cooperation with other agents.This could be vertical(leader or servant)or horizontal(peer-to-peer).Research indicate

130、s that multiagent systems are more effective when roles are clearly defined.A leader agent coordinates tasks,while servant agents communicate with both the leader and their peers.Teams with an organized leader complete tasks nearly 10%faster than teams operating without a designated leader.Agentic a

131、rchitecture layersThe foundation of any agent-based system consists of multiple interdependent layers that work in harmony to enable intelligent behavior.These layers function as a pipeline,transforming raw input into actions while maintaining the agents internal state and reasoning capabilities.At

132、the core,the sense-plan-act cycle drives decision-making,while the standard architecture layers ensure the system meets critical enterprise architecture requirements(Figure 2).Figure 2.Agentic architecture layers and supporting technologyPerceptionCognitiveActionIntegrationOperationsInfrastructureUs

133、er interfacePlanningTools libraryEnterprise APIsPerformanceand evaluation Runtime(GPUsand constrainers)Vector and agentmemory databasesLLM APIgatewaysData engineeringpipelinesExternalservicesResource andtoken utilizationAgent tracesand trackingGuardrails forresponsive AIMonitoringand alerts Human-in

134、-the-loopinterfacesLLM Hub and enterpriseLLM gatewaysEnterprise data andknowledgebaseRAG embeddingsPromptlibraryHITL controllerand coordinatorMultimodalitycontrollerAgent0rchestrationState ormemorymanagementContextualreasoningContinuouslearningDecision-makingAct withhelp of toolsAPIsSensorsOther dat

135、aacquisition systemsDatatransformationLangChain or LangGraphFiddlerAzureAWSNVIDIAGoogle CloudOpenTelemetryDeep Eval RAGASAgentOpsArize AI PhoenixAutoGenCrewAICoreStandardSource:Infosys24|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navig

136、ator:Agentic Enterprise AI Playbook|25External Document 2025 Infosys Limited Knowledge InstituteCore layersPerception:Functions as the agents sensory interface with its environment,responsible for processing raw inputs,reducing noise,and extracting relevant features.It converts unstructured data int

137、o a structured format,allowing the cognitive layer to process information efficiently.In a robotic system,this layer analyzes camera feeds,sensor data,and environmental readings transforming them into a structured state representation for further analysis and action.Cognitive:Serves as the agents br

138、ain,handling decision-making,planning,and learning.It manages the agents internal state,processes information from the perception layer,and determines the most appropriate actions based on its goals and current understanding.Depending on the agents complexity,this layer can range from a simple rules

139、-based system to an advanced neural network.Action:Translates decisions from the cognitive layer into activity.It is responsible for validating actions,performing safety checks,and monitoring implementation.This layer ensures that planned actions are feasible and safe before implementation,while als

140、o monitoring feedback to evaluate outcomes.Standard layersIntegration:Serves as the connective tissue of the agentic architecture,facilitating seamless communication and data flow between the systems layers and external systems.It enables real-time data access,manages interoperability,and ensures sm

141、ooth interactions,enhancing the systems overall functionality and responsiveness.Operations:Oversees and manages the real-time performance of agentic systems.It monitors system activities,provides feedback mechanisms,and facilitates continuous improvement by optimizing processes based on operational

142、 data.Infrastructure:Delivers the computational power and storage required for agents to execute complex tasks efficiently while ensuring high availability,scalability,and reliability.Tech Navigator:Agentic Enterprise AI Playbook|25External Document 2025 Infosys Limited Knowledge InstituteKnowledge

143、Institute26|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Agent vs.human:Audit use caseAgentic AI is a versatile and powerful technology capable of handling a wide range of tasks.This use case explores how it can be applied to the standard audit process for expe

144、nse vouching.For a human auditor,the vouching process involves verifying recorded transactions against supporting documents to ensure accuracy and authenticity.This method plays a crucial role in detecting and preventing errors and fraud,ultimately ensuring the reliability of financial statements(Fi

145、gure 3).Figure 3.Traditional workflow steps for the auditing process0102030405060708UnderstandSelectCollectExamineUnderstand the clients system,internal controls,and risks related to transaction processing.Select a sample of transactions to vouch typically based on risk assessment conducted during t

146、he planning phase.Collect relevant documentary evidence for each transaction selected for vouching.Examine documents to ensure they are authentic and complete;check for integrity,time stamp,counts,amounts,and compliance.CompareVerifyComplyCompileCompare document details with corresponding entries re

147、corded in the books of accounts;check for accuracy in amounts,counts,transaction details,and recording dates.Verify that each transaction has been properly authorized and approved based on organizations internal control procedures;helps confirm that all transactions are legitimate and approved at th

148、e correct levels of authority.Check whether transaction has been followed up properly.For example,in the case of purchase transactions,its verified whether goods received notes match the purchase orders and invoices.Compile findings and report discrepancies or anomalies to the audit committee,along

149、with recommendations for resolving discrepancies and strengthening controls.Source:Infosys26|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|27External Document 2025 Infosys Limited Knowledge Institu

150、teSource:InfosysTo automate the expense vouching process,an agentic AI solution can deploy multiple agents that collaborate within a structured workflow using a human-in-the-loop approach(Figure 4).AI agents interpret audit instructions,categorize collected files based on type,and extract relevant d

151、ata elements from each document.They then populate a tailored vouching template,performing calculations as specified in the audit procedures.Throughout the process,agents meticulously review outputs to ensure compliance with regulatory guidelines,flag discrepancies or exceptions,and facilitate human

152、 oversight where necessary.Figure 4.Agentic workflow steps for the auditing process0102030405060708UnderstandCollectSelect skillsExamineAgents understand the clients system and internal controls related to transaction processing and select the transactions based on instructions.Document collection a

153、gent gathers relevant documentary evidence for each transaction selected for vouching.Planning or coordination agent selects skills needed for tasks typically based on instructions,available agents,and skills required to carry out audit tasks.Validation agent examines documents to ensure they are au

154、thentic and complete;extract the contents of documents and check for document compliance with relevant data.CompareHuman-in-the-loopReportCompileInvestigation agent compares document details with corresponding entries recorded in the systems;check for accuracy in amounts,counts,transaction details,a

155、nd recording dates.Audit instructions guide the agents.Investigation or reviewer agent interacts with humans to perform reasonableness assessments and review audit procedure;verifies that each transaction has been properly authorized and follows the audit procedure guidelines.Reporting agent enables

156、 the system to generate reports or export data in the user provided template,facilitating easy sharing,analysis,and further data manipulation.Agents maintain an audit trail(logs)through process,which allows human analysts to examine agents function and review what went well and what procedures needs

157、 to be changed.This information informs future improvement.Tech Navigator:Agentic Enterprise AI Playbook|27External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute28|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited The following view shows the

158、agentic workflow in action,where each agent is equipped with the necessary skills to execute its assigned tasks.Agents interact with inputs through a user interface(UI),while a cognitive layer manages planning,reasoning,and learning using large language models(LLMs)and small language models(SLMs).Th

159、ese models work alongside a knowledge graph to provide business context,such as the vouching process in this example,and maintain workflow execution state using shared memory,such as a Redis cache (Figure 5).To complete their tasks,agents utilize specialized tools,such as enterprise APIs or utility

160、functions.When human intervention or feedback is required to complete a specific operation,the agent engages the human in the loop through the UI,ensuring accuracy and compliance while enhancing overall efficiency.Figure 5.Mapping the agentic workflow for expense vouchingSharedmemoryToolsSkill:Under

161、stand instruction,planning and coordinationDigital Business ContextHuman supervisor(user interface)LLM hubSkill:Collect transaction documentsGoalPlanning or coordination agentDocument collection agentCoordinatewith other agentsPlanSenseReasonAudit procedures(unstructured data)Vector storeDefine expe

162、cted outputMake correctionsReview reportsHuman-in-loop investigationsAssess agent performanceUpdate audit proceduresAudit taxonomy(knowledge graph)Structured data,APIsSupportdocsInvoicereceiptsSenseActSkill:Validate transaction against evidenceValidation agentSenseActSkill:Investigate against audit

163、procedures;human involvement when neededInvestigation agentSenseActSkill:Generate audit report;review with humanReporting agentSenseActSource:Infosys28|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook

164、|29External Document 2025 Infosys Limited Knowledge InstituteHow to implement agentic AIThe successful implementation of agentic AI systems requires careful selection of agent types,tools,and frameworks that align with their specific capabilities and overall business objectives.As a relatively new a

165、nd rapidly evolving technology,many companies remain uncertain about how to develop and deploy agentic AI effectively.Organizations must navigate significant technical complexity to ensure successful execution.Selecting the right tools and frameworks is crucial for an agentic systems operational suc

166、cess,ensuring scalability,efficiency,and integration with enterprise environments.There are four broad categories of agentic platforms and frameworks available to enterprises:1.Vertically integrated agentic platforms,with domain and services specific agents delivering services-as-software(Salesforce

167、 Agentforce,ServiceNow AI Agents,SAP Joule).2.Cloud provider agentic platforms and frameworks for agent development and management(Amazon Web Services Bedrock,Google Vertex,Microsoft Copilot Studio and AI Foundry SDK).3.Open-source AI agents,frameworks,and tools(LangGraph,CrewAI,AutoGen,MetaGPT).4.P

168、roprietary agentic platforms focused on agentic automation for specific processes and features(Beam AI,Cognition AI,Kore.ai).The following section outlines the key capabilities of three widely used agentic AI frameworks that can be used to create custom-built agentic solutions.Tech Navigator:Agentic

169、 Enterprise AI Playbook|29External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute30|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Compatibility with agent typesEnterprises should ensure that the selected tools and frameworks align with the

170、specific type of agents being developed,whether reflex-based,model-based,or goal-based.Each agent type has distinct requirements for example,reflex-based agents require real-time processing,while goal-based agents depend on advanced planning algorithms.ScalabilityPreferred frameworks scale efficient

171、ly as task complexity increases or as the number of agents grows.Multiagent systems often require a robust infrastructure to handle higher workloads and ensure seamless communication among agents.LangGraphGraph-based workflows,real-time processing,structured data handling,and memory management;can i

172、mplement cycles and branching logic,enabling agents to manage dynamic tasks effectively.Reflex-based,model-based,utility-basedCrewAIRole-based design,task delegation,multiagent collaboration,advanced memory;integrates well with LangGraph for combinations of different agent types.Goal-basedAutoGenCon

173、versational approach;quick responses;modular design;supports planning.Reflex-based,goal-basedFrameworkFeaturesBest fit for agent typeLangGraphScales effectively with graph nodes and transitions;supports complex workflows and memory systems for handling increased loads;enterprise-ready platform with

174、LangSmith.CrewAIProduction grade scalability with NVIDIA NIM integration;ideal for tasks requiring multiple agents to work in parallel.AutoGenWell suited for conversational agents and modular components;tailored for simpler multiagent scenarios;still at experimental stage and not fully production re

175、ady.FrameworkFeatures30|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|31External Document 2025 Infosys Limited Knowledge InstituteIntegration and interoperabilityTools should enable seamless integr

176、ation with existing enterprise systems and technologies.They must support standard protocols and APIs,ensuring effective communication between agents and external data sources and services.Flexibility and customizationFrameworks should offer flexibility for customization to meet specific business ne

177、eds and operational workflows.The ability to modify agent behaviors and integrate new functionalities is essential for adapting to evolving requirements.LangGraphSupports seamless integration with other frameworks by wrapping agents in nodes,allowing for multiagent systems and effective communicatio

178、n with external data sources;access to comprehensive LangChain ecosystem leading LLM app development platform.CrewAIBuilt on LangChain,it facilitates easy integration with various tools and supports standard protocols for effective agent communication,including collaboration with LangGraph.AutoGenMo

179、dular design allows for integration with multiple tools and services,ensuring agents can communicate through APIs;limited to Microsoft ecosystem.FrameworkFeaturesLangGraphCan tailor solutions to specific needs and manage complex business interactions.CrewAIProvides flexibility through its role-based

180、 architecture where developers can create custom agents with defined roles,skills,and behaviors.Its customization capabilities are currently limited to sequential and hierarchical task workflows rather than complex collaborative patterns to solve the problem(which is under development as of early 20

181、25).AutoGenMore suited for conversational model(group chat,reflection)among agents with limited support for complex workflow models.FrameworkFeaturesTech Navigator:Agentic Enterprise AI Playbook|31External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute32|Tech Navigator:Agentic

182、Enterprise AI PlaybookExternal Document 2025 Infosys Limited Support for learning and adaptationFrameworks with built-in machine learning capabilities are needed to support continuous improvement.This is especially critical for agents that must learn from their environments,adapt their decision-maki

183、ng strategies,and refine their performance as time progresses.Observability and monitoringTools should include observability features to monitor agents and the performance of underlying LLMs or SLMs in real time.Real-time monitoring enables developers to track agent effectiveness,identify bottleneck

184、s,and optimize system performance.LangGraphGraph execution details(node performance,traversal,errors)LangSmith,Arize Phoenix,and AgentOpsCrewAIAgent task execution,communication,and tool usageAgentOps and Arize PhoenixAutoGenConversation flow,LLM calls,tool usage,agent behaviorAgentOps and Arize Pho

185、enixFrameworkFocusTools or supportLangGraphSupports stateful,multiagent applications with automatic state management and coordination,enabling agents to learn from interactions and adapt strategies dynamically.CrewAISupports memory functionalities,enabling agents to retain context from previous inte

186、ractions;allows agents to build on experiences,enhancing adaptability and effectiveness in dynamic environments;can train agents using command line interface(during design time)to learn from human feedback.AutoGenAI agents can collaborate,share information,and solve complex tasks together while lear

187、ning from each other.Flexible architecture,allowing developers to create specialized agents(e.g.AssistantAgent and UserProxyAgent)that can engage in multiturn conversations,work together,and learn as a result with features like code execution,customizable conversation patterns,and human input integr

188、ation.FrameworkFeatures32|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|33External Document 2025 Infosys Limited Knowledge InstituteWhile not all these frameworks provide built-in observability and

189、 monitoring capabilities with LangSmith being an exception they integrate well with enterprise-ready platforms that support open protocols such as OpenTelemetry.For complex ecosystems utilizing quantized models or SLMs,additional production-grade tools,such as Fiddler AI or Galileo,can further enhan

190、ce model and data quality monitoring.AgentOpsAgent testing and debugging LLM prompt,completion logging Error tracingCost trackingArize PhoenixRetrieval-augmented generation system monitoringEmbedding analysis Vector store insights Semantic search evaluationsRelevance metricsLatency tracking and cost

191、 optimizationLangSmithLangChain development and monitoringEnd-to-end tracing Performance monitoringComprehensive LLM metrics Testing toolsFiddler AILLM performance monitoringModel metrics(real-time LLM analysis)Bias detection Explainability toolsProduction monitoringMonitoring alertsGalileoAI model

192、training performance and data qualityData quality metricsModel behavior analysisTest suite monitoringPerformance trackingToolPurposeMonitoring featuresTech Navigator:Agentic Enterprise AI Playbook|33External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute34|Tech Navigator:Agenti

193、c Enterprise AI PlaybookExternal Document 2025 Infosys Limited Development community and supportTools with a strong development community and robust support resources contribute to the successful implementation of complex agentic AI systems.A well-established community provides access to troubleshoo

194、ting assistance,knowledge sharing,and best practices.LangGraphStrong development community supported by active GitHub repositories;comprehensive documentation;integration with LangSmith for performance monitoring and LangSmith hub for collaboration on prompts.CrewAIBenefits from growing community wi

195、thin the agentic development community,along with partnerships with big names like IBM and NVIDIA;offers forums and collaborative resources.AutoGenNot as popular as competitors;Microsoft releasing similar frameworks;some community engagement available through GitHub and Reddit.FrameworkFeatures34|Te

196、ch Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|35External Document 2025 Infosys Limited Knowledge InstituteCost-effectiveness(FinOps)Assessing the total cost of ownership,including licensing fees,main

197、tenance costs,and resource requirements is a critical step in the deployment of agents.A cost-effective solution should provide substantial value without compromising functionality or scalability.While many agentic frameworks are open-source,some offer enterprise versions that may be better suited f

198、or large-scale deployments beyond the proof-of-concept stage.Since many of these frameworks especially those based on LLMs are relatively new,a thorough evaluation is essential to ensure seamless integration with enterprise FinOps capabilities and long-term sustainability.Foundation for AI success I

199、ndustry leaders such as NVIDIA and IBM have hailed agentic AI as the“next frontier of AI”and the“next big thing in AI research,”emphasizing its transformative potential.Salesforce CEO Marc Benioff describes it as a“new labor model,new productivity model,and a new economic model.”As with any emerging

200、 technology particularly those that automate increasingly complex tasks risks are inherent.When integrated into enterprise systems,these agents introduce an element of nondeterminism,enabling adaptive and intelligent behavior while maintaining consistent and reliable outcomes.When implemented effect

201、ively,agentic AI applications ensure scalability,seamless integration,flexibility,and continuous learning and adaptation.Selecting the right tools,frameworks,and a robust agentic AI architecture is crucial,especially for early adopters.The goal is not to“move fast and break things”but to move fast w

202、hile building a strong foundation one that provides the enterprise with a sustainable,long-term competitive advantage.LangGraphOffers tiered pricing structure(Developer,Plus,Enterprise)with predictable costs for smaller teams;enterprise solutions may involve variable costs based on deployment needs

203、and infrastructure investments.CrewAIFlexible pricing models(Basic,Premium);costs vary significantly depending on customization and integration requirements(API calls and data processing tasks billed at volume of usage).AutoGenStill at experimental stage;not fully suitable for production use cases;w

204、ould require management of cloud resources by frameworks users;not fully integrated with Microsoft Azure AI Foundry.FrameworkFeaturesTech Navigator:Agentic Enterprise AI Playbook|35External Document 2025 Infosys Limited Knowledge Institute36|Tech Navigator:Agentic Enterprise AI PlaybookExternal Docu

205、ment 2025 Infosys Limited AGENTOPS AND AGENTIC LIFE CYCLE MANAGEMENTArtificial intelligence(AI)must continually evolve to unlock its full potential in automating business and organizational processes.While technologists and business leaders anticipate groundbreaking advancements,these ambitions will

206、 remain unrealized without clear direction,structured execution,and strategic alignment.Agentic systems address this challenge by orchestrating dynamic planning,decision-making,and complex interactions essential for enterprise operations.However,to maximize their effectiveness,organizations must ado

207、pt systematic approaches to management of these systems.This is where AgentOps plays a crucial role.As an end-to-end life cycle management framework,AgentOps ensures enterprise scalability,reliability,transparency,and efficiency.It provides a clear pathway for AI integration by streamlining the desi

208、gn,evaluation,deployment,monitoring,and continuous improvement of agentic systems.Yet,despite its benefits,AgentOps remains underutilized in generative AI deployments an oversight that could limit AIs transformative impact.The value of AgentOpsAgentOps builds on the principles of DevSecOps,MLOps,and

209、 LLMOps while addressing the unique challenges of agentic systems.These systems incorporate planning,reasoning,and autonomous decision-making,leveraging memory and contextual knowledge to navigate complex interactions.By integrating tools and governance measures,AgentOps ensures seamless management,

210、enabling agents to 36|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|37External Document 2025 Infosys Limited Knowledge Instituteoperate efficiently,adapt dynamically,and stay aligned with enterpris

211、e goals while maintaining operational integrity.A crucial aspect of AgentOps is the establishment of guardrails constraints and safety mechanisms that prevent AI agents from taking unintended actions.These safeguards ensure autonomous systems operate within defined boundaries,enhancing scalability a

212、nd transparency.By mitigating risks and optimizing performance,AgentOps enables organizations to harness more of agentic AIs potential.Mature adoption of AgentOps practices and patterns will achieve the following key objectives:Ensure behavioral consistency by implementing a comprehensive evaluation

213、 framework that guides agents in both normal and unexpected situations.Enhance system reliability by reducing mean time between failures through anomaly detection and predictive issue identification.Accelerate issue resolution with robust observability and debugging tools that minimize mean time to

214、resolution.Maintain compliance by enforcing auditability through consistent audit logs and explainable decision-making.Enable scalable operations with centralized management and governance controls.Optimize costs by improving resource efficiency and minimizing operational overhead.Knowledge Institut

215、e38|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited AgentOps life cycle phasescycle should incorporate a rigorous design review phase to verify reliability,security,and safety.Once the design is approved,the process transitions to workflow and task mapping,outlini

216、ng the agents steps to achieve its objectives and goals.In the development stage,small or large language models(LLMs)are integrated into the agents reasoning and communication processes.Data sources,connectors,tools,and plugins are incorporated to enhance the agents capabilities.Agentic systems shou

217、ld maintain a dynamic registry of available tools,APIs,and their capabilities.Frameworks such as LangChain and LlamaIndex facilitate seamless tool integration and efficient functionality management.A critical design consideration is implementing restrictions or strict validations on user-provided pr

218、ompts to prevent unintended behaviors.Agents must be trained with specialized skills and techniques tailored to their environment.This process involves acquiring and structuring high-quality training data,accounting for potential edge cases and biases,and iteratively refining the agents decision-mak

219、ing through real-world interactions.The reflection design pattern enables language models to evaluate their own outputs,creating an iterative cycle of self-improvement.During the design and build stage,secure and safe development practices must be followed to mitigate vulnerabilities and safety risk

220、s.The AI bill of materials(AIBOM),which catalogs software,hardware,datasets,and tools,enhances transparency by AgentOps provides significant value but also requires substantial investment in understanding its phases and their effects on agentic deployments.A well-defined framework,from initial devel

221、opment to real-time observability,helps manage challenges such as drift,security vulnerabilities,and decision-making accountability.This systematic approach ensures that AI agents operate as intended while continuously evolving to adapt to changing conditions.The life cycle phases of AgentOps play a

222、 critical role in ensuring scalability,transparency,and the long-term success of agentic systems,with each stage contributing to their effective management and continuous improvement.Define and designThis initial phase focuses on developing agents and tools that align with an organizations needs.The

223、 process begins with defining clear objectives,specifying what the agent must achieve,and the context in which it will operate.These objectives should be comprehensive,encompassing both functional and nonfunctional requirements to ensure the agent meets performance,security,and compliance standards.

224、A well-defined design must explicitly address safe operations,transparency,and accountability in every decision and action.It should include mechanisms for human-in-the-loop interventions and a“big red button”(human override)to stop the agentic system if necessary.Like the traditional software devel

225、opment life cycle,the agentic AI life Tech Navigator:Agentic Enterprise AI Playbook|39External Document 2025 Infosys Limited Knowledge Institutedetailing the components used in the AI system,along with their dependencies and interactions.A mature design practice should prioritize generating an AIBOM

226、 for agentic AI systems while conducting continuous risk assessments,security incident response planning,compliance checks,supply chain security evaluations,and AI system audits.Testing and evaluationThis process begins with defining important success metrics for agentic AI systems and developing ri

227、gorous evaluation strategies and methodologies to ensure reliability,effectiveness,efficiency,adaptability,ethical adherence,and regulatory compliance.Quality engineering plays a crucial role in this phase by designing comprehensive test plans and creating a virtual environment that simulates real-w

228、orld scenarios to assess agent behavior.Evaluation typically follows a dual testing approach,incorporating both vertical testing of individual agents and horizontal testing of the end-to-end agentic process.An important consideration in the vertical testing approach is assessing the performance of i

229、ndividual agents.This involves capturing critical metrics,such as the number of attempts with successful task completions,the accuracy of tool selection,mean time to complete tasks,service level objective adherence,and the frequency of human intervention.End-to-end testing is essential to ensure the

230、 accurate functioning of agentic systems.The evaluation process incorporates automated assessments,agent-as-judge evaluations,LLM-assisted evaluations,and human oversight,creating a robust testing framework.Key performance metrics include the following:Tool utilization efficacy:Measures the agents a

231、bility to select and use appropriate tools effectively.Memory coherence and retrieval:Evaluates the agents ability to store,retrieve,and apply information efficiently.Tech Navigator:Agentic Enterprise AI Playbook|39External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute40|Tech

232、Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Strategic planning index:Assesses the agents capability to formulate and execute plans successfully.Component synergy score:Determines how well different components of the agentic system interact and function together.Bey

233、ond performance characteristics,security testing is a critical focus area,particularly in mitigating risks associated with the OWASP Foundations top threats for LLMs and agentic AI.Agentic systems expand the systems attack surface,increasing the risk of security vulnerabilities.For example,a comprom

234、ised agent could serve as a gateway for attackers to exploit the underlying database,leading to data breaches,unauthorized access,or system manipulation.Additionally,attacks on multistep reasoning processes target weaknesses in AI logic,affecting input interpretation,intermediate steps,and final out

235、puts.To mitigate these risks,organizations must implement end-to-end observability,tracing,and anomaly detection to identify malicious activities such as prompt injections,data leakage,model poisoning,and excessive agency.DeploymentOnce the agentic AI system meets the required evaluation criteria an

236、d resolves all outstanding issues or defects,it is ready for production release.During deployment,the agent is introduced into the production environment and integrated with relevant tools and APIs to enable real-world interactions.40|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 20

237、25 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|41External Document 2025 Infosys Limited Knowledge InstituteAs part of deployment,an identity management solution such as HashiCorp Vault,AWS Secrets Manager,or Azure Key Vault is implemented to enable agents to secu

238、rely store and manage credentials for tools and external APIs.The next step involves properly sizing the infrastructure to support auto-scaling,high data volume handling,low latency,reliability,high availability,security,data privacy,and cost optimization.Agentic components are typically deployed as

239、 container workloads,with a container orchestrator such as Kubernetes providing built-in resiliency and auto-scaling capabilities.A pivotal decision is whether to deploy on a hyperscaler or a private cloud,depending on security and regulatory requirements.Based on the target deployment architecture,

240、an automated provisioning pipeline is established for the agentic AI system.This process can follow a traditional infrastructure-as-code approach or an infrastructure-from-code model.In the latter,the agentic system determines its infrastructure requirements and directly orchestrates provisioning an

241、d configuration using cloud APIs or tools such as Terraform,OpenTofu,and Ansible.After infrastructure provisioning,a continuous delivery pipeline is established to automate the deployment of agentic components,utilizing release strategies such as blue-green or canary deployments.Like any digital app

242、lication,deployment performance is measured using DORA metrics,including deployment frequency,change lead time,change failure rate,mean time to recovery,and service level objective adherence.Observe and improveBefore a production release,it is essential to establish a robust framework and toolchain

243、for end-to-end observability,traceability,and debuggability.These ensure transparency,enhance credibility,and provide a comprehensive understanding of the agents internal state,behavior,tool invocation,interactions,knowledge retrieval,and decision-making throughout its life cycle.Observability is es

244、sential to gain insights into how an AI agent or a system of agents works internally and interacts with the environment.Capabilities include:Input tracking:Monitors the inputs an agent collects from users,systems,or the environment.Output monitoring:Tracks responses to ensure they align with expecte

245、d outcomes.Reasoning logs:Captures intermediate steps in an agents decision-making process.Model insights:Analyzes how contextual knowledge is acquired and how LLM agents process prompts.Anomaly detection:Flags unexpected behavior or outputs.System integration tracking:Monitors how the agent interac

246、ts with tools and other software or hardware components.Traceability ensures transparency by tracking an agents decision-making processes,interactions,and outcomes,which are critical for regulatory compliance and actionable insights.Capabilities include:Knowledge Institute42|Tech Navigator:Agentic E

247、nterprise AI PlaybookExternal Document 2025 Infosys Limited Decision logs:Captures what the agent decided and why.Version control:Tracks dynamic updates to code,configurations,workflows,or prompts.Reproducibility:Preserves the agentic systems state,including all metadata,to demonstrate how a decisio

248、n or outcome was reached.Debuggability focuses on rapidly diagnosing and resolving production issues to minimize mean time to resolve.Capabilities include:Prompt refinement:Enables iterative adjustments to enhance agent responses.Workflow debugger:Tracks inputs,outputs,execution time,and flags faile

249、d steps for troubleshooting.Snapshot of relevant logs:Captures logs to facilitate easier problem diagnosis and resolution.Scenario simulation:Provides a structured framework to test and assess agent performance,distinguishing between ill-defined user requests and system malfunctions.As agentic appli

250、cations scale in production,a continuous learning model is essential for maintaining long-term effectiveness.This requires frequent updates to knowledge bases,ensuring the agent has access to accurate and up-to-date information.Regular performance audits are critical,with decision logs and outcomes

251、reviewed by experts or other agents to assess and improve performance.Additionally,behavior refinement involves adjusting processes or cues based on observed behaviors,enhancing the agents adaptability and efficiency over time.By integrating observability,traceability,and continuous learning mechani

252、sms,enterprises can develop agentic AI systems that are reliable,accountable,and adaptable to evolving real-world conditions.Evolution to AgentOpsThe shift from LLMOps to AgentOps expands the scope,complexity,and life cycle imperatives.The figure below outlines the most significant differences and T

253、ech Navigator:Agentic Enterprise AI Playbook|43External Document 2025 Infosys Limited Knowledge Instituteillustrates how AgentOps builds upon the foundation established by LLMOps (Figure 1).AgentOps tools and frameworksThe AgentOps tools landscape is rapidly evolving to support the full life cycle o

254、f Figure 1.Comparisons between LLMOps and AgentOpsScopeLLMOps focuses on preparing,deploying,and maintaining LLMs,including tasks such as prompt management,fine-tuning,and versioning;primarily focused on LLM inference,model-specific APIs,and integration.Oversees full life cycle of agentic systems,wh

255、ere LLMs and other models or tools function within a broader decision-making loop;must orchestrate complex interactions and tasks using data from external systems,tools,sensors,and dynamic environments.Integration complexity Focuses on single model or a few models;primarily monitors inference calls

256、and prompt templates rather than real-time external actions performed by AI agents.Manages fleets of interacting agents,introducing challenges such as concurrency,role-based collaboration,and conflict resolution;must track action lineage,manage resource locks,and implement rollback mechanisms to mit

257、igate undesired changes since agents operate within environments and connect to external tools.EvaluationEnsures accurate and reliable outputs from language models.Ensures agents are dependable,traceable,and auditable across operations.Observability and debuggingTracks model performance metrics such

258、 as accuracy,latency,and drift while monitoring prompt usage and output.Incorporates tools to capture multistep interactions,including agent goals,chain-of-thought,tool usage,memory,triggered subagents or tools,and real-world decision-making;extra dimension of observability is needed to diagnose and

259、 debug more autonomous agent decisions.Logging and audit trailDocuments the model training,datasets,and outputs.Expands documentation to include agents decisions,workflows,and interactions;deals with agent memory persistence(audit trail capability required to show how agents internal memory store is

260、 updated and used over multiple sessions).ParametersLLMOps AgentOps Tech Navigator:Agentic Enterprise AI Playbook|43External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute44|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Life cycle managemen

261、tLimited to model selection,deployment,fine-tuning,and retraining;includes versioning of models and prompts.Covers agent design,orchestration,updates,performance evaluation,cost optimization,continuous improvement,and agent retirement;requires AIBOM and versioning for entire agentic system(not just

262、for model or prompt).Tools and frameworksRelies on model evaluation,performance monitoring,retraining,and prompt management tools.Incorporates tools for orchestration,evaluation,observability,decision tracking,security scanning,logging and auditing;agentic system cost management.Feedback loopsCollec

263、ts feedback on model outputs for fine-tuning.Includes feedback on agent behavior and outcomes for continuous improvement.Source:Infosysagentic system development.However,it is still in its early stages compared to DevSecOps and LLMOps.The table below highlights some of the tools and options.As a new

264、 technology with limited tools,the implementation of a comprehensive and effective agentic AI life cycle management solution presents significant challenges.Design frameworks Workflow design Memory management Multiagent interaction management Integration with tools and external APIs Tool registryLan

265、gChain LangGraph AutoGen Crew AI Evaluation Planning,decision-making,and output quality evaluation Tool usage analysis Agent collaboration and system interaction assessment Automation Test reporting and analyticsRagaAI Braintrust Databricks Mosaic AI Agent Evaluation Okareo Giskard Agent-as-a-Judge

266、AgentOps life cycleKey capabilitiesIndustry tools or works in progress 44|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|45External Document 2025 Infosys Limited Knowledge InstituteSecurity and comp

267、liance testing OWASP AI top 10 check Build and run time scan AI security posture management control Compliance testing regulatory requirementsCalypsoAI Wiz Zenity Giskard AIBOM generation Components of agentic AI applications Version and other metadata Source and authorWiz OWASP Manifest Observabili

268、ty and debugging Monitoring Distributed tracing Debugging Cost managementAgentOps.AI LangSmith Arize Langfuse Braintrust End to end agentic AI life cycle management Project management Integration with design framework Evaluation and release management End-to-end observabilityAgentOps.AI Arize Braint

269、rustSource:InfosysOne major hurdle is the lack of a standardized evaluation and testing framework for agentic systems,making it difficult to benchmark performance and reliability consistently.Additionally,no widely adopted platform exists for managing the entire life cycle of agentic AI,requiring or

270、ganizations to integrate disparate tools and processes to achieve full functionality.Another critical challenge is the generation of AIBOM and compliance testing,both essential for regulatory adherence and transparency but lacking mature,automated solutions.Real-time monitoring further complicates m

271、atters,as observability agents can be costly and resource-intensive,especially in large-scale systems where managing vast data volumes demands substantial effort.Traceability is another critical concern,particularly with black-box AI systems like LLMs.The opaque nature of these models makes it diffi

272、cult to understand and document their decision-making processes.Likewise,maintaining an auditable and consistent agent memory is essential,as uncontrolled memory growth can lead to inconsistencies,inefficiencies,and compliance risks.Overcoming these challenges requires robust frameworks,advanced obs

273、ervability tools,and Knowledge Institute46|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited industrywide standards to support the evolving landscape of agentic AI.Future of AgentOps As agentic AI systems gain autonomy and integrate more deeply into critical infrast

274、ructure,AgentOps will evolve to introduce new capabilities that enhance scalability,reliability,and self-regulation.seamless communication and coordination among multiple agents handling complex tasks.As these innovations advance,AgentOps will not only streamline the management of agentic systems bu

275、t also cultivate a more resilient,adaptable,and intelligent AI infrastructure capable of sustaining enterprise-scale automation and decision-making.From experimentation to scaleAgentOps encompasses the entire life cycle of autonomous agents,from design to One significant advancement on the horizon i

276、s the automated design of agentic systems(ADAS),where AI-driven meta-agents autonomously generate and refine new agents.This self-referential approach allows AI to design and optimize its own successors,continuously improving agentic systems by discovering novel building blocks and more advanced arc

277、hitectures.Self-provisioning and deployment are also transforming how agents manage infrastructure,allowing them to autonomously configure resources and optimize deployment strategies based on workload demands.At the same time,the rise of self-observing agents will introduce self-regulating mechanis

278、ms,enabling them to monitor and supervise their own actions to maintain alignment with predefined objectives and ethical considerations.To support these advancements,industrywide standardized protocols will establish best practices for event tracing,system visibility,and operational control monitori

279、ng enhancing transparency and interoperability across AI-driven ecosystems.Additionally,interagent collaboration frameworks will be crucial for facilitating Tech Navigator:Agentic Enterprise AI Playbook|47External Document 2025 Infosys Limited Knowledge Instituteorchestration to performance evaluati

280、on,and ultimately,agent retirement.This will become essential to AI initiatives,ensuring that intelligent agents operate efficiently,ethically,and in alignment with enterprise goals.By establishing structured processes for managing AI agents,organizations can maintain control,compliance,and continuo

281、us improvement,enabling intelligent systems to operate effectively in dynamic environments.However,while the potential is clear,the path to full-scale adoption requires patience.The ecosystem supporting AI agents including tools,infrastructure,and governance frameworks is still maturing.Standardizat

282、ion efforts are underway,but businesses must navigate a period of iteration and refinement before these agents can function seamlessly across industries.As AgentOps evolves,organizations will need to balance experimentation with responsible deployment.Early adopters will face challenges in defining

283、best practices,integrating agents into existing workflows,and maintaining compliance.Yet,as standards solidify and AI governance improves,AgentOps will shift from an emerging concept to an essential function,much like DevOps transformed software development.Those who invest in measured,strategic ado

284、ption will be well-positioned to reap the long-term benefits of intelligent agents that are not only powerful but also trustworthy,adaptable,and enterprise-ready.Tech Navigator:Agentic Enterprise AI Playbook|47External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute48|Tech Navig

285、ator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited 48|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited ADVANCED AGENTSAs organizations race to adopt agentic artificial intelligence(AI),successful implementation demands a comprehensive,strateg

286、ic approach.Rather than focusing solely on the technologys capabilities,businesses must also understand how AI agents develop and execute those capabilities effectively.AI agents employ diverse reasoning models to accomplish tasks across various domains,including IT operations,customer service,and c

287、ontent management.These approaches range from a single agent accessing multiple information sources via APIs to multiagent systems collaborating in hierarchical or conversational structures.However,these approaches come with inherent challenges,including memory management,efficient information retri

288、eval,and defining execution limits to ensure timely responses.Despite these obstacles,they serve as steppingstones toward the next generation of AI agents,which will further enhance reasoning capabilities and deliver more accurate,context-aware outputs.For agents to successfully automate a process o

289、r workflow,they must be able to break down a users request into a clear,step-by-step plan.For example,solving an algebraic equation requires an agent to follow these five steps:1.Identify the variable that needs to be solved.2.Simplify the expression.3.Isolate the variable by adding or subtracting t

290、erms,multiplying or dividing coefficient to isolate the variable.Knowledge Institute48|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Tech Navigator:Agentic Enterprise AI Playbook|49External Document 2025 Infosys Limited Knowledge Institute4.Check the answer by s

291、ubstituting the calculated value of the variable.5.Present the final answer.This is a simple example of a plan or workflow that an agent or a large language model(LLM)needs to come up with to solve a problem and then execute each step using an appropriate tool to come up with the final answer.In bus

292、iness scenarios,the plans could be how to resolve a customer ticket by predicting the resolution,planning the steps to execute like calling enterprise API and summarizing the outcomes as a final answer,Challenges lie when agents and LLMs need to understand enterprise process or product specific nuan

293、ces,policies and tools.Due to these,specific agentic patterns be it a single agent,multiagent,external aided,or fully autonomous planning agents are required.Below are some common agentic patterns.Single agent with task decomposition These were made popular with the paper“ReAct:Synergizing Reasoning

294、 and Acting in Language Models,”and are based on task decomposition,or breaking down the task into a series of subtasks.This is done by using LLMs ability to generate a reasoning chain and act,or calling on or more tools,and finally processing the outcome of the tool to accomplish the task.Lets take

295、 an example in which a user asks about the weather in New York for the next four days.An LLM is given a prompt where Knowledge Institute50|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited it is asked to think,to determine a tool such as an API or code,wait for tool

296、 output,and then present its observation.The typical prompt looks like this:In the example prompt above,suggesting the tools and giving examples such as few-shot learning provides a guide to an agent to perform its tasks.Agents can You run in a loop of thought interleaving thought,action,pause,obser

297、vations steps.Use thoughts to describe your thoughts about the question you have been askedUse tool to run one of the tools available to you,then pause.Make sure you perform one step at a time and wait for next step to be called.Your available tools are:Weather tool to perform weather prediction for

298、 expressions like weather for next four days,weather on a given day of the week,or date or month.Weather tool accepts inputs like 1,3,4,5 and up to 7 days,a valid date in month/day/year format and a city code like LN and country code such as UK.Examples:Question:What is the weather for next three da

299、ys in London?Thought:I should look up weather API to find the weather Tool:weather:inputs LN,UK,3d:PauseObservation:use more than one tool by calling them in a sequential manner.For example,if a user wants to plan a holiday based on weather,an LLM can be supplied with a tool such as Transport for Lo

300、ndons API to plan a mode of transport to reach a desired location.Developers can use frameworks like LangChain,and LlamaIndex to implement react agent-based solutions quickly while supplying their preferred choice of an LLM.Agents with external planner Engineers in ITOps need detailed and specific k

301、nowledge to find a root cause for a given Tech Navigator:Agentic Enterprise AI Playbook|51External Document 2025 Infosys Limited Knowledge Instituteproblem or incident,and they will execute several steps to resolve an incident.To come up with a plan to find a root cause of an issue,an agent needs to

302、 understand an incident,query appropriate monitoring systems,understand logs and metrics associated with the application in question,and understand and query observability systems like user experience monitoring,including tools such as Dynatrace and Newrelic,in order to arrive at various data points

303、 and make an appropriate suggestions on root cause analysis.The challenge here is with LLMs like Open AI or Claude,which lack understanding and workflow to resolve a specific task for a given enterprise to determine an appropriate flow.To manage these variances in sources of information,and the sequ

304、ential tasks an agent needs to perform,it might not be prudent to manage these tasks with one single prompt as it would be difficult to introduce new plans based on debugging information.In such cases based on complexity and scale of the problem various patterns can be introduced:Static-or graph-bas

305、ed agentsThese are semi-autonomous agents that take dynamic decisions.As an example,when an invoice order tracking query is sent by the customer,after comprehending the email,based on the order number,product type and status of the order,the ticket is assigned to appropriate team,who then responds t

306、o the customer or the contact center agent.Here,an autonomous agent has to understand the ask from the email,use a tool to retrieve the order details,based on the product,inventory and shipping status,then make an appropriate decision to either inform the product inventory manager or send an email t

307、o the logistics company to check the on status of the order to get back to the customer.For problems that require concrete steps and LLM-based decisioning(understanding emails,sending or assigning tickets to either the product manager or the logistics Tech Navigator:Agentic Enterprise AI Playbook|51

308、External Document 2025 Infosys Limited Knowledge InstituteKnowledge Institute52|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited company in above example),or one that has to implement specific workflow patterns,graph-based agents really shine.A graph structure wher

309、e a node can be defined as an agent or task,and edges that define the relationship or how these agents can depend on each other,helps with parallel execution and reallocation of the tasks.In the above example,the inventory agent,shipment agent and customer support agent can orchestrate when for exam

310、ple the inventory agent comes back with an ETA which is longer than usual,the customer service agent can then either advise the customer on the delay or escalate to a human supervisor.A complex retrieval-augmented generation(RAG)system,which augments LLMs by retrieving relevant knowledge,and rewrite

311、s queries based on the history of the conversation,or dynamically routes to the various tools and agents,is where this graph-based pattern approach shines.Additionally,self-RAG patterns,where LLMs are required to critique their own output in order to refine the answer,are some examples where complex

312、 yet well-defined patterns can also be implemented with graph-based agents.Dynamic external plansThis is a simple set of external instructions in form of prompts or structured input that provides a plan to guide various sequences of steps based on the task at hand where the LLM can choose a specific

313、 prompt or planning sequence based on the problem statement.To generate a plan,agents can use a static prompt such as one without variables that can either be configured using an external source like a database or a prompt DB.Using intelligent search,the LLM can then retrieve an appropriate plan.The

314、 external source could also be a configuration management database,where system dependencies and their observability platform,which monitors the health of the whole environment,can be queried and used to create a troubleshooting plan for an agent.Multiagent or collaborative agentsSeveral frameworks,

315、including MetaGPT,AutoGen,and CrewAI,enable different patterns for collaborating multiple agents to 52|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited Knowledge InstituteTech Navigator:Agentic Enterprise AI Playbook|53External Document 2025 Infosys Limited Knowled

316、ge Instituteperform various tasks.These can all work in parallel,with or without a team leader sitting on top of them.Although the agent can use a range of models,each has a specific,static role and can only do one thing.These patterns enable automating and optimizing either part of a process or an

317、entire process to improve performance.For example,a multiagent coder proposes an approach to develop better code comprising three agents:A programmer,a test designer,and test executor agents.During the code development phase,the programmer and test designer agents can execute in parallel to develop

318、the code and test cases.In later phases,the programmer agent refines the code based on feedback from the test executor agents.The test designer agent can re-generate or refine the test cases with code generated and tested in previous phases.To develop these applications,where agents need to interact

319、 with each other,the following patterns are used:Hierarchical patternThese are applications that require a hierarchical pattern where a supervisor agent routes the inquiry to lower-level agents.In the example discussed above where a customer is querying the status of their order,in this model a cust

320、omer service agent acts as a supervisor agent and directs a customer query or issue to either the product inventory manager,the logistics company or back to the customer,based on current state of the order.Here the customer service agent is responsible for breaking down the tasks:Finding the order,c

321、hecking the inventory,checking with the logistics company,and forwarding these tasks to appropriate agents.Based on inputs from each agent,it makes an appropriate decision to either respond or escalate to a human manager.The graph-based agents described above are a special case of orchestration,wher

322、e there are no supervisor agents,and the entire agent interaction flow is defined using DAGs.Conversation patternThis approach is used for tasks which require more dynamic nature of interactions.An example of this approach is a knowledge publishing platform producing an article where the author,revi

323、ewer,copy writer,or social media influencer agents can act simultaneously along with human authors and reviewers to produce,critique and do plagiarism checks on the article.In this pattern,individual agents are allowed to send messages to each other without having a supervisor present to direct the

324、conversation.Although the systems require policies to define in which order the agents must execute for example,a research agent could produce a summary by browsing the internet a writer agent could use resources found by the researcher agent and then produce the content.At the same time,the social

325、media influencer agent can generate the posts for various social platforms.Knowledge Institute54|Tech Navigator:Agentic Enterprise AI PlaybookExternal Document 2025 Infosys Limited A reviewer agent then carries out an overall review of the content,including social media posts which can align to the

326、content.The reviewer agent then either provides a rewrite or goes ahead to approve and post the content.Technical challenges Enterprises face several challenges implementing agentic architecture and platforms,including:Memory management:Memory or conversation context among the agents needs to be man

327、aged in the form of chat history,specific facts and outcomes of the agent.Storing conversations in vector database or caches and the ability to recall this history either in present conversation or in the longer term requires a special approach that needs to be designed.Context management:Multiagent

328、 chats can be very long and relying on the content of these long chats can either meet with LLMs token size limitation or increased cost/response time due to context lengths.To manage the context,various techniques from summarization,overwriting context and memory recall from long-term memory is imp

329、lemented.Deployment:Running multiagent applications can turn into long-running,higher-order messaging processes and hence require distributed execution of agents as well as messaging capabilities that can help communicate with different agents in a low latency and scalable manner.Replay and replan b

330、ased on user inputs(human in the loop):It is important for agents to restart or replan based on user interaction,where an end user might ask the agent to run the entire workflow based on new facts,or ask a specific agent execution to repeat from a specific step.Policy management:To reduce the chatte

331、r between agents in cases where the outcome Tech Navigator:Agentic Enterprise AI Playbook|55External Document 2025 Infosys Limited Knowledge Instituteis not clear,agents can run into loops or take a long time to resolve the tasks.This in turn could result in a poor customer experience,or cost overru

332、ns due to multiple LLM calls.To tackle this,limits on agent execution time,order of orchestrations,and interdependency should be set and managed via clear and visible policies.Foundation of future applicationsThe reasoning capability of LLMs is at the core of agent frameworks.There are various resea

333、rch initiatives in progress to improve generation of reasoning chains,including these approaches:Reasoning or search algorithmsLLMs fundamentally generate chains using either greedy search,which tries to make the locally optimal choice at each stage,or the more complex beam search,which systematically expands the most promising nodes.These are effective algorithms for predicting the next token,but