Buckle Up – Agentic AI is here and accelerating rapidly. Agentic AI refers to artificial intelligence systems that can autonomously make decisions, plan actions, and execute tasks with a degree of independence, rather than simply responding to predefined inputs. It signifies a shift from AI that primarily analyzes and assists to AI that can proactively initiate actions and pursue goals. In essence, agentic AI systems are designed to operate more like human employees, capable of understanding context, reasoning through problems, and adapting to changing circumstances.
Building agentic AI into software systems across the enterprise will form the foundation of new knowledge economies and markets. Agentic AI is different because of its unique combination of capabilities that take it beyond existing approaches to automation and insights. What distinguishes agentic AI is that it can plan strategically, reason through complex scenarios, collaborate between different components, and leverage external tools to achieve objectives with remarkable autonomy. This is because agentic AI comprises systems of foundation models, rules, architectures, and tools that enable software to flexibly plan and adapt to resolve goals by taking action in their environment, with increasing levels of autonomy.
The agentic AI market is experiencing rapid growth, with projections indicating a significant increase in market size over the next decade. The market is expected to expand from approximately $2 billion in 2024 to over $94 billion by 2035, representing a compound annual growth rate (CAGR) of roughly 40%. This growth is fueled by the increasing demand for automation, the rise of AI copilots, and the need for intelligent workflow optimization.
What are the key attributes of Agentic AI?
- Autonomy: Agentic AI systems can operate independently and proactively, without constant human guidance.
- Goal-Oriented: They are designed to achieve specific objectives, breaking down complex tasks into smaller steps and adapting their approach as needed.
- Reasoning and Planning: Agentic AI can reason about situations, plan actions, and execute those actions to reach its goals.
- Learning and Adaptation: These systems can learn from their experiences and adapt their behavior to improve performance over time.
- Integration: Agentic AI often integrates with multiple systems and data sources, allowing it to leverage a wide range of information for decision-making.
How does Agentic AI differ from other forms of AI?
- From Predictive AI: Predictive AI focuses on analyzing data to forecast future outcomes, while agentic AI goes a step further by taking action based on those predictions.
- From Generative AI: Generative AI creates new content based on prompts, while agentic AI can execute complex workflows, make real-time decisions, and adapt to changing conditions.
- From Traditional AI: Traditional AI systems often rely on pre-defined rules and require explicit instructions, while agentic AI can operate more independently and dynamically.
What are examples of Agentic AI?
- Customer Service:Intelligent agents are being developed to handle multi-step queries and provide more efficient customer support.
- Sales: Providing real-time coaching to sales AEs based on their work steps, providing auto-sales reach outs, and engaging with customers to answer questions they may have.
- Healthcare: Managing patient appointments, analyzing medical records, and assisting with treatment plans.
- Finance: Agentic AI is being used to moderate risk, optimize workflows, and analyze data in financial institutions. Optimizing investment strategies, detecting fraud, and automating trading processes.
- Cybersecurity: Continuously monitoring networks, detecting anomalies, and responding to threats.
- Scientific Discovery:AI agents are being used to automate complex research workflows, including literature reviews and experiment design.
- Supply Chain Management: Automating order placement, optimizing inventory levels, and managing logistics.
Agentic AI represents a major advancement in the field of AI, enabling systems to act as intelligent agents that can autonomously solve highly routine problems and achieve key objectives, but in a clearly defined scope and set of circumstances.
Are technology vendors confusing users?
According to Forrester Research, technology vendors are claiming that they are agentic AI solution providers, which is more often not accurate, and they are conflating all kinds of AI agents with agentic AI agents.
They are, however, very different.
This has resulted in traditional process automation tools, predictive analytics systems, standalone large language models, and retrieval augmented generation systems being referred to as AI agents when they alone don’t possess the capabilities of planning, adapting, and acting that agentic AI delivers. While each of these technologies contributes valuable capabilities to the modern enterprise technology stack, agentic AI is distinct from them. Forrester provides helpful context to demystify how to interpret agentic AI and put it in the context of other AI technologies. Below is a summary they prepared that is very useful:
- Process automation tools lack adaptability and intelligence. Traditional process automation systems like robotic process automation (RPA) and digital process automation (DPA) operate within clearly defined parameters to execute repetitive tasks in structured environments. They excel at executing predetermined rules and procedures, making them highly efficient for standardized operations like data entry, file management, or transaction processing. However, they lack the adaptability and intelligence that characterizes agentic AI. For example, an RPA system might efficiently process invoices following a strict set of rules, but agentic AI can flexibly handle exceptions, learn from new scenarios, and dynamically adjust its approach based on changing circumstances. This distinction becomes particularly evident with unstructured data or variable business conditions, where agentic AI can demonstrate greater flexibility, unmatched by traditional automation systems.
- Predictive analytics and ML can anticipate but not execute. Predictive analytics and machine learning systems excel at identifying patterns and generating forecasts by processing vast amounts of data to predict future trends, detect anomalies, and classify information with remarkable accuracy. However, these capabilities are limited to the analytical domain and lack the crucial ability to autonomously act on or adapt based on changing objectives. Agentic AI, by contrast, integrates analytical capabilities with autonomous decision-making and action execution. In a retail context, while a predictive analytics system might forecast seasonal demand patterns, agentic AI could autonomously adjust inventory levels, modify pricing strategies, and coordinate with suppliers to optimize the entire supply chain ecosystem. This demonstrates a level of operational intelligence that goes far beyond mere prediction to act, plan, and adapt.
- LLMs can understand and plan, but not act. The distinction between agentic AI and standalone LLMs represents the most nuanced comparison, as LLMs often serve as fundamental components of agentic AI. LLMs demonstrate remarkable capabilities in natural language processing, generation, and understanding, but alone, they cannot act within their environment. Agentic AI integrates LLM capabilities within a broader framework that includes strategic planning, tool utilization, and autonomous action execution. This integration enables not only the understanding of tasks, planning, and communication, but also the capability to initiate actions, coordinate multiple tasks, and persistently pursue objectives over time. For example, while an LLM might excel at drafting an email or analyzing a document, agentic AI can manage entire projects, coordinating multiple stakeholders, adjusting plans based on feedback, and actively working toward defined goals with minimal human intervention.
- RAG-based architectures lack planning, reasoning, and adaptability. Retrieval-automated generation (RAG) is one of the most commonly adopted architectures to integrate the power of foundation models into enterprise applications, and it’s often conflated with agentic AI. RAG systems can perform sophisticated searches into existing knowledge repositories and vector databases and can use natural language to synthesize retrieved information into new formats for a user, but most stop there — they don’t include significant capabilities for planning or reasoning across information, nor do they include abilities to take adaptable action within enterprise environments. Some agentic applications may leverage RAG within agentic loops (sometimes described as agentic RAG), but RAG pipelines on their own are not agentic.
AI Agents & Agentic AI: The Applications & The Systems (Forrester Research, 2025).
Agentic AI Enables Adaptiveness And Flexibility, With Some Caveats
True agentic AI systems demonstrate a high degree of autonomy, actively orchestrating and managing the pathways necessary for executing programs or workflows with precision and intent. These systems can navigate and execute multistep processes with minimal human oversight by coordinating diverse models, data sources, and tools and applying these mechanisms holistically. Today’s agentic AI can’t achieve an enterprise-wide level of broad-based autonomy and agency because of limitations on integrations, security, and explainability — but it’s evolving along that path. Today we see agentic AI manifesting as (see Figure 3):
- Complex-flow agentic AI that enables use cases that need flexibility. This is the first type of agent to truly use agentic AI to support planning and task execution. Using the context and tools presented to them, these systems choose how to execute a control flow based on their goals. Importantly, traditional automation takes part in controlling and orchestrating the end process. For example, Taskade advertises itself as an AI-powered work management tool, offering users the ability to build a workflow to access agents as needed, such as research, project planning, or to-do list tracking. These agents support each of their discrete workflows but are managed by traditional automation. The agents can modify actions based on explicit feedback instrumented into the system, allowing it to handle more complex and varied tasks than traditional automation can.
- Multiflow agentic AI will coordinate multiple use cases for more strategic value. As systems mature, multiple systems using agentic AI will operate as specialized independent entities that communicate, negotiate, and collaborate in real time. For instance, an agentic AI could pull information from sales records, summarize it, and then assemble it into a prospecting list. It then reaches out to a marketing platform to select visual assets and generate copy using foundation models, then pushes that content back to the sales platform, where an agent decides on the best contact method and launches the campaign. The reality of this aspiration is underwhelming today, but we are seeing rapid improvement. We expect to see multiflow agentic AI with advanced reasoning over the next few years in production in enterprise deployments.
- Any-flow agentic AI will eventually offer holistic enterprise decisions. The vision of a mature agentic AI is multiple ecosystems of agents acting independently and with increasing autonomy on behalf of companies, employees, and customers. These systems will allow holistic communication across multiple data tenancies and enterprise environments and won’t be restricted to a single vendor’s garden or enterprise’s architecture. Achieving these interfaces will require overcoming major obstacles. The first is a lack of standardization: Different vendors build their agents on disparate technology stacks, and no universal protocol currently exists — beyond descriptive APIs — to facilitate communication among agents. The second is motivation: By offering pre-packaged agent workflows that leverage proprietary client data, application-specific workflows, and their own data stacks, vendors can create tighter lock-in within their ecosystems. Currently, this reduces the incentive to develop a common, industrywide framework for agent-to-agent interactions. We expect to see numerous attempts at this in the coming year, and no single standard in the short term.
Fortune Will Favor The Bold: Architect Your Agentic AI Future Across Roles
- The adoption of agentic AI is a competitive necessity. And while there are still challenges around testing and securing these systems, early adopters will establish and build a lead, while laggards will decelerate and risk obsolescence. Enterprises that fail to grasp the importance of agentic AI will waste investments on rigid non-agentic automation instead of transformative agentic AI. Leaders must develop clear AI strategies, progressively integrate an agentic AI, and continuously evolve their capabilities. All leaders must recognize that adopting agentic AI will not be a complete redesign of their applications and processes tomorrow, but the impacts of agentic AI will be felt by all, and each role’s place in this new landscape will be different and evolve over time. Every role needs to start acting now, but carefully.
- Chief executive officers: Architect the “autonomous enterprise.” For CEOs, agentic AI represents more than just an efficiency play — it signals a fundamental shift in how enterprises compete, operate, and deliver value. The winners of the AI age will not merely be the fastest adopters but those who redesign their business models to harness agentic AI as a strategic differentiator. They can drive step-change improvements in productivity, customer engagement, and decision-making, but their real power lies in enabling new revenue streams, reshaping industries, and unlocking exponential scale. CEOs must champion AI transformation from the top, ensuring alignment across technology, operations, and business strategy. This requires bold investments in AI capabilities, a workforce reskilling agenda, a culture that embraces AI-driven experimentation, and strong AI governance to manage issues of trust, bias, and compliance.
- Line of business leaders: Select business cases for metrics and modularity. In early stages, many tempting use cases seemingly provide significant value, but their scope is far beyond what has actually been built, deployed, and proven in the enterprise. Business leaders who want to get started with agentic AI shouldn’t invent new workflows or use cases, exciting as that may be. Focus on use cases with known and easily capturable metrics — both for ongoing operations and for outcomes. Additionally, because both the underlying technologies and their applications are still evolving, select use cases with evolution and modularity in mind. Build pragmatically for today, but be prepared for rapid and dramatic shifts in technology capability over the medium term.
- Technology leaders: Prepare for transitional stages and establish foundations. Recognize that today’s gap between the promise and the realization of agentic AI is temporary. Within the next three years, agents will be trusted to control and manage a significant portion of enterprise processes. Process orchestration is critical to reach the agentic milestone and must be defined in your architecture. Ensuring proper interaction will be a challenge, often leading to inefficiencies and conflicts. Agentic capabilities with higher levels of autonomy, where a model controls the orchestration, will take time to evolve. Therefore, prepare for a transitional stage of process control for the next several years.
- Data leaders: Build the foundations of agentic memory and security. High-quality data is the key ingredient for any AI, and agentic AI makes data even more essential. For agentic AI, it’s essential to use tools like metadata and knowledge graphs to understand and manage the memory context. Agentic AI will also rely on robust structures for data validation, lineage, and auditing to provide a basis for establishing and growing trust between agentic AI and humans, as well as between different agentic AI systems in the future. AI tools and platforms are making this task easier: Data teams today can leverage foundation models to help generate data definitions, documentation, synthetic datasets for privacy or testing purposes, and more.
- Security leaders and AI governance leaders: Be innovators yourselves. Agentic AI presents a tremendous opportunity to businesses but a tremendous challenge to leaders tasked with using AI responsibly and managing its security and risk because many of the supporting frameworks and tools aren’t in place yet. This challenge also presents an opportunity for security and governance leaders to drive AI innovation and business lift. Risk leaders must ground agentic AI with rigorous testing, simulation, and validation practices to make it more secure and accurate. Many of the challenges around interoperability extend to questions around how to properly manage access controls, sensitive data, and other essential enterprise interconnects. Additionally, AI governance leaders must ensure that reasoning and decision-making processes are transparent to maintain accuracy and avoid compounding errors. This is where much of the engineering effort will be, and it may be a big barrier to success. Like all aspects of AI, this space is moving quickly, and security and responsible AI leaders must also innovate to support the resilience needed.
Understanding Quality Agentic AI Research Sources
Multi-Agent Collaboration:
- “Talk Structurally, Act Hierarchically: A Collaborative Framework for LLM Multi-Agent Systems”:This paper explores a framework for improved collaboration and hierarchical refinement in multi-agent AI systems.
- “Mixture of Agents: Enhancing Large Language Model Capabilities”: This research investigates how combining specialized AI agents can improve performance on complex tasks, particularly in language understanding and generation.
Self-Improving Agents:
- “SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills”: This work focuses on how AI agents can autonomously learn and improve their abilities by discovering and honing new skills.
- “A Self-Improving Coding Agent”: This research explores the development of agents capable of self-improving their coding abilities.
Governing Agentic AI:
- “Practices for Governing Agentic AI Systems” by OpenAI:This practical paper outlines seven key practices for developing safe and accountable agentic AI systems.
- “AI Agents: Governing Autonomy in the Digital Age”: This report examines the challenges of governing AI systems that can act independently and proposes solutions.
Agentic AI for Scientific Discovery:
- “Agentic AI for Scientific Discovery: A Survey of Progress, Challenges, and Future Directions”:This survey paper explores the use of agentic AI in scientific research, covering areas like literature review and experiment design.
Agentic RAG Systems:
- “GFM-RAG: Graph Foundation Model for Retrieval Augmented Generation”:This paper focuses on using graph-based models for more effective retrieval-augmented generation. “DeepRAG: Thinking to Retrieval Step by Step for Large Language Models”: This research explores how to improve retrieval augmented generation by incorporating a step-by-step reasoning process within the agent.
What are the leading Agentic AI Frameworks?
- LangChain:A popular framework for building applications powered by language models, including agentic applications.
- CrewAI:A framework specializing in creating intelligent agents that can collaborate and share tasks.
- LangGraph:A graph-based framework for structuring relationships between data points, enhancing agentic decision-making.
- Microsoft Semantic Kernel:A framework effective for industries needing automation and advanced language comprehension.
What are the important considerations in advancing Agentic AI Wisely?
While agentic AI holds great promise, it’s crucial to manage expectations and avoid overhyping its capabilities. In addition, as AI agents become more powerful, it is crucial that executives develop robust governance and secure control systems to ensure their usage of Agentic AI methods is safe and ethical. The successful implementation of AI will require careful consideration of data quality and accessibility. In addition, Agentic AI is in its very early stages of experimentation, although its growth is staggering, the ROI use cases must be carefully thought through to ensure value is validated before investing in its implementation.
We always recommend to our customers to start small and learn to fail fast. However, every mid to large enterprise must start to develop a cohesive Agentic AI Strategy and architecture plan in 2025/2026. The market is moving at a speed I have never seen before so the learning is critical — paying on older technology upgrades may be a very prudent decision and refocus on understanding how to leapfrog ahead.
If you do not have an AGENTIC AI strategy for your enterprise, please reach out to us, and we can take you on a guided tour of what is possible.
Research Sources:
1.) Dr. Cindy Gordon AI Musings Research and Reflections/Writings
2.) Broda, Eric. Agentic Mesh (Book Released March, 2026)
3.) Agentic Mesh — More Details to Ponder
4.) Gemini – Defining Agentic AI
5.) OpenAI – Defining Agentic AI
6.) Forrester Research
To stay updated about the latest in AI and its impact, subscribe to our CEO’s weekly AI Insights newsletter
