Agentic AI: Promising Yet Challenging Future for Enterprise Automation

The Rise of Agentic AI in Enterprise Software

Agentic AI, a new breed of artificial intelligence capable of autonomous action and decision-making, is gaining significant attention in the enterprise software landscape. Unlike traditional generative AI, agentic AI can independently set and pursue goals, plan, reason, and execute tasks across multiple steps, adapting to changing environments 1

1

. This technology is seen as a potential game-changer for businesses, with Gartner predicting that agents will be embedded in one-third of enterprise software applications by 2028 1

1

.

Practical Applications and Industry Adoption

Companies are already beginning to integrate agentic AI into their products and services. Thomson Reuters, for instance, acquired Materia to leverage its agentic AI capabilities in breaking down complex queries and consulting numerous information sources to deliver compound answers 1

1

. This approach could be particularly beneficial in fields like accounting, where there's a projected shortage of professionals.

Other major players entering the agentic AI space include:

• Salesforce, with its Agentforce tool set for building customized agents 1

  1

• Workato, adding agentic capabilities to its data integration platform 1

  1

• Amazon Web Services, expected to highlight AI agents at its re:Invent conference 1

  1

• A new startup, /dev/agents, raising over $50 million to build an operating system for agents 1

  1

Technical Challenges and Reliability Concerns

Despite the enthusiasm, agentic AI faces significant technical challenges that limit its current applications. Niloufar Salehi, an assistant professor at UC Berkeley, highlights the unpredictable nature of machine learning algorithms as a major hurdle 2

2

. Unlike traditional data processing systems, agentic AI systems are stochastic, meaning their outputs can vary based on context and other factors.

Key challenges include:

1. Reliability: The stochastic nature of these systems makes it difficult to ensure consistent results 2

   2

   .
2. Shared Memory: Creating persistent, shared memory for agents to learn from past actions is a complex engineering feat 2

   2

   .
3. Coordination: Enabling agents to work together reliably across services and companies is a significant challenge 2

   2

   .

Safety and Accountability Concerns

While agentic AI holds promise for automating complex tasks, it also raises important questions about accountability and safety. Unlike generative AI, which may produce misleading information, agentic models have the potential to cause more significant damage if given too much decision-making latitude 1

1

. This concern has led experts to suggest that AI agents will likely be limited to non-critical business decisions in clearly defined domains for the foreseeable future 1

1

.

Future Outlook

Despite the challenges, researchers and companies are working on solutions. Projects like MemGPT aim to extend the context length of AI models, potentially improving their predictability and repeatability 2

2

. Salehi is optimistic that agents capable of coordinating across services and companies will be practical within the next five years, though their applications may differ from current expectations 2

2

.

As the field progresses, the focus will be on solving the engineering problems necessary to make agents work together reliably. While the potential of agentic AI is significant, its widespread adoption in critical business processes will likely require overcoming substantial technological and governance challenges in the coming years 1

1

2

2

.