Artificial Intelligence is evolving rapidly, changing not only how we interact with technology but also how we design, develop and deliver software solutions.
Over recent years, Generative AI has transformed the way we work with information, generate content, write code and solve problems. Large Language Models (LLMs) have made AI more accessible, allowing applications to understand natural language, generate responses and assist with increasingly complex tasks.
However, we are now seeing another important development: Agentic AI.
Rather than simply responding to prompts or generating content, AI agents introduce the ability to reason through tasks, interact with external tools, access information and take actions towards achieving specific goals.
This represents an interesting shift in software engineering and solution architecture.
What is Agentic AI?
At its core, Agentic AI refers to AI-powered systems that can work towards a defined objective with a degree of autonomy.
Unlike traditional AI applications, where a user provides an input and receives an output, an AI agent can take a more active role in completing a task.
An agent can:
- Understand a user’s objective.
- Break a complex task into smaller steps.
- Retrieve relevant information from knowledge sources.
- Select and use appropriate tools.
- Interact with external APIs and services.
- Evaluate results and determine the next action.
For example, consider an IT service management application.
In a traditional application, a user might submit a service request, and predefined business logic would determine how that request is processed.
With an AI agent, the system could potentially analyse the request, identify the appropriate service, retrieve relevant knowledge articles, check information through external systems and recommend or initiate the next steps.
The important distinction is that the agent can make decisions about how to approach a task rather than relying entirely on a fixed sequence of operations.
Of course, actions involving sensitive data, financial transactions or important business decisions should still operate within clearly defined permissions and human approval processes.
From AI Assistants to Intelligent Agents
Traditional AI assistants are primarily designed to respond to questions and provide information.
Agentic AI takes this further by introducing planning, tool usage and task execution.
Consider a simple example:
A user asks an AI assistant to investigate why a particular application is unavailable.
A conventional AI assistant might explain possible causes and suggest troubleshooting steps.
An AI agent, with appropriate access and permissions, could potentially:
- Check application health through monitoring APIs.
- Retrieve recent application logs.
- Identify relevant errors.
- Consult technical documentation.
- Summarise its findings and recommend corrective actions.
This ability to interact with different systems and combine information from multiple sources makes AI agents particularly interesting for enterprise applications.
It also introduces new architectural considerations around security, permissions, reliability and system integration.
The Role of Tools, APIs and MCP
One of the most important aspects of Agentic AI is the ability to interact with external systems.
An AI model on its own has limitations. It does not automatically have access to enterprise databases, business applications or external services.
Tools provide a way to extend an agent’s capabilities.
For example, an agent might use:
- REST APIs to communicate with enterprise applications.
- Database tools to retrieve business information.
- Search services to find relevant documents.
- Monitoring APIs to investigate application health.
- Custom functions to execute specific business operations.
Another interesting development is the Model Context Protocol (MCP).
MCP provides a standardised way for AI applications to connect with external tools and data sources. Rather than building separate integration mechanisms for every AI application, MCP provides a common protocol through which compatible systems can expose their capabilities.
From a software architecture perspective, this is particularly interesting because integration has always been an important part of enterprise software development.
MCP introduces another approach to connecting AI systems with the wider application ecosystem.
Knowledge-Driven AI and Foundry IQ
For enterprise AI solutions, having access to the right information is just as important as having an intelligent model.
An AI agent needs relevant context to provide useful and accurate results.
This is where knowledge-driven AI becomes important.
Technologies such as Microsoft Foundry IQ help connect AI applications and agents with organisational knowledge.
This allows agents to work with information from different sources rather than relying exclusively on the knowledge contained within the underlying language model.
For example, an enterprise agent could retrieve information from internal documentation, technical knowledge bases and business content to provide more relevant responses.
This is closely related to Retrieval-Augmented Generation (RAG), where relevant information is retrieved and supplied to a language model as additional context.
The combination of knowledge retrieval and agent capabilities creates opportunities for more context-aware enterprise applications.
Multi-Agent Systems and Orchestration
As tasks become more complex, a single AI agent may not always be the best approach.
This is where multi-agent architectures become interesting.
Instead of relying on one agent to handle everything, responsibilities can be distributed across multiple specialised agents.
For example, consider an enterprise software support solution consisting of:
- Coordinator Agent: Understands the user’s request and delegates tasks.
- Knowledge Agent: Searches technical documentation and knowledge bases.
- Diagnostic Agent: Investigates application logs and monitoring information.
- Resolution Agent: Suggests possible solutions based on the findings.
These agents can work together through an orchestration mechanism to complete a larger task.
Frameworks such as the Microsoft Agent Framework provide capabilities for building and coordinating these types of agent-driven workflows.
However, introducing multiple agents also brings additional complexity. Communication, state management, error handling, observability and execution control all need to be considered during the design.
The goal should not be to introduce multiple agents simply because the technology makes it possible. The architecture should be driven by the problem being solved.
Rethinking Enterprise Software Architecture
Perhaps the most interesting aspect of Agentic AI is what it means for the future of software architecture.
Traditional enterprise applications are generally built around well-defined business processes, APIs, services and predefined execution paths.
These architectural principles remain important. However, AI agents introduce another layer of intelligence that can interact with existing systems and coordinate activities across them.
This raises several important architectural considerations:
Integration: How should AI agents interact with existing applications, APIs and enterprise services?
Security: How do we manage agent permissions, access controls and sensitive information?
Reliability: How do we ensure that agents produce consistent results and handle unexpected situations?
Governance: How do we monitor agent activities, control decisions and maintain appropriate human oversight?
Scalability: How do we design agent-driven solutions that can operate reliably within enterprise environments?
These are not simply AI development questions. They are fundamental software architecture and engineering considerations.
In my view, successful enterprise AI solutions will depend not only on the capabilities of AI models but also on the quality of the surrounding architecture, integration patterns and engineering practices.
The Role of Software Engineers and Technical Leaders
For software engineers, Agentic AI introduces new opportunities to expand beyond traditional application development.
Understanding how language models work is useful, but building practical AI solutions also requires knowledge of APIs, integration patterns, data architecture, security, distributed systems and application design.
For solution architects, the challenge is to determine where agent-based approaches genuinely add value and how they can be incorporated into existing enterprise architectures.
For technical leaders, there is also an opportunity to guide engineering teams through this transition, encourage experimentation and establish sound engineering practices around AI adoption.
I don’t believe Agentic AI replaces the need for good software engineering. If anything, it makes solid engineering principles even more important.
AI agents still need reliable services, well-designed APIs, secure integrations, appropriate monitoring and clear operational boundaries.
Looking Ahead
We are still at an early stage of understanding the full potential of Agentic AI.
The technology is developing quickly, and there are still challenges around reliability, security, cost, governance and practical enterprise adoption.
However, one thing is becoming increasingly clear: AI is moving beyond being simply a feature within an application towards becoming an active participant in how software systems operate.
For those of us working in software engineering, architecture and technology leadership, this is an exciting time.
The opportunity is not simply to adopt another new technology, but to rethink how we design intelligent systems, integrate applications and solve business problems.
The future of software engineering will not just be about building applications that use AI. It will increasingly be about architecting systems where AI can reason, collaborate, interact and take meaningful action.
And that is a direction worth exploring.