Engineering the agentic enterprise

A new era of enterprise software

Enterprise software is at an inflection point. Organizations are moving from traditional, standard packaged applications and configurable software-as-a-service (SaaS) to “SaaS+” models augmented by agent-enabled customization. The next horizon most certainly will be agentic architectures, where autonomous software agents can coordinate workflows, make decisions, and act with human guidance and oversight.

This evolution isn’t an “either/or” proposition between packaged and custom, or human-led and agent-led. It’s an “and” approach to help enterprises determine the right models for their needs. Executives must understand how to best integrate these models for an effective balance between innovation, security, and governance. 

The shift from packaged to SaaS+ to agentic only

Historically, organizations relied on packaged or custom, off-the-shelf (COTS) software with minimal customization to avoid unnecessary complexity. They also avoided customizing solutions due to the restrictive nature and cost of customization. But the rise of SaaS enabled faster deployment, regular updates, and configurable workflows while keeping complexity in check. Today, there’s a shift toward “SaaS+” which inserts agentic capabilities into standard platforms, thereby embedding intelligence, automation, and customization within core systems.

We believe the next stage and the future of software engineering is “agentic only.” This is the process of building systems with autonomous agents that can analyze, act, and collaborate across environments. AI agents can address the unique processes and needs of individual organizations, giving them flexibility over COTS solutions. For example, agents can autonomously interpret business goals, invoke tools and processes, and execute workflows across systems, often with more efficiency than traditional automation. Agents can also provide persona-based “digital twins” to execute business functions without a corresponding growth in headcount. 

Three enablers for packaged, SaaS+, and agentic AI operations

For organizations to operate effectively across a continuum of packaged, SaaS+, and agentic AI, three enablers are crucial: modern infrastructure, advanced observability, and robust cybersecurity.

Infrastructure


A cloud-native, edge, and automated infrastructure is the foundation for the agility that enables choice. Agentic applications require dynamic compute capacity, near real-time data access, and flexible scalability. Techniques such as containerization, orchestration, and infrastructure-as-code approaches allow organizations to efficiently deploy, scale, and integrate agents across hybrid environments. Where latency or local-processing capabilities matter, edge computing becomes essential. Without this agility, agentic systems could remain as proofs of concept rather than valuable, scalable realities.

Observability


Observability enables organizations to monitor how agents operate, make decisions, and impact broader systems. Observability is more than simple monitoring. It can provide insight into agent behavior, workflow efficiency, and operational anomalies. This enables teams to spot issues and remediate them before they become critical. Further, metrics on decision quality, tool use, and escalation patterns can enable continuous improvement and reveal areas that could benefit from human oversight. Finally, observability fosters trust in systems that are partially autonomous “black boxes,” which is a non-negotiable condition for scaling agentic operations to the enterprise.

Security


Because agentic architectures redefine the security environment, modern approaches such as zero trust, identity-aware networking, and application programming interface (API)-level protection are mandatory. Fundamentally, each autonomous agent represents a new identity within the enterprise environment that requires authentication, authorization, and auditing akin to that of a human user. Complexity is increased with autonomous agents working in a multi-agent system to achieve a business goal. Agents can spawn ephemeral agents and control, auditing, discovery, and life cycle management of child agents. Critical steps to protect these environments involve:

• Zero trust. Each interaction, whether human or agent-initiated, must be verified. Techniques such as least-privilege access and micro-segmentation can reduce risk.

• Agent and API governance. Just like humans, each agent must adhere to defined access parameters, credentials, and life cycle management policies such as registration, auditing, and retirement.

• Data security and governance. Agents often require access to large, disparate volumes of enterprise data. It’s essential to practice effective data handling, lineage tracking, and encryption.

• Life cycle control. Agents learn and iterate as they evolve. Their decisions and outputs must be completely traceable, explainable, and constantly subject to review.
  
  

These factors are fundamental to an enterprise’s ability to innovate securely. Infrastructure enables performance, observability fosters trust, and security maintains control. Taken together, they can determine whether agentic AI introduces more vulnerability or delivers 
sustainable value.

Engineering and cyber: Securing the future

The success of agentic AI depends immensely on the creation of a trusting, collaborative partnership between software engineering and cybersecurity. Both may need to shed their histories as independent, sometimes rival, disciplines to become partners that are jointly responsible for design, delivery, and safeguarding intelligent enterprise systems. Trust is the cornerstone of successful AI adoption.

Benefits of adopting across packaged, SaaS+, and agentic architectures

Progress is process of managed, secure evolution

Agentic AI’s emergence heralds an evolution in how enterprises design, deploy, and manage software. Success in this new era will require thoughtful planning—deciding where to maintain packaged solutions, where to move to SaaS+, and where to implement fully agentic architectures—and whether and when to deploy all three.

By approaching agentic adoption as a process that’s grounded in engineering discipline, cybersecurity consistency, and strategic vision, leaders can turn potential into better performance and gain a competitive edge in the era of the intelligent enterprise. Learn how we can help in our full report.