Stop Using Enterprise Software Practices on Generative AI Projects
Generative AI development requires a shift from traditional software practices to a collaborative, business-led approach like Evaluation Driven Design (EDD), enabling rapid iteration and aligning with real-world needs to achieve success.
Harrison Kirby
1/21/20253 min read
A dangerous trend is emerging: attempting to apply enterprise software development best practices to GenAI agent development. While traditional software and Generative AI agents share some similarities, their core differences make this approach risky and counterproductive. In this article, we’ll explore these differences, explain why the business unit—not IT—must take ownership, and introduce the concept of Evaluation Driven Design (EDD) as the critical methodology for success.
The Problem with Applying Enterprise Software Practices to GenAI
In traditional software development, the goal is to create a program that delivers predictable, repeatable outputs. Developers work toward a defined end state, and success is measured against a fixed set of requirements. Generative AI, however, operates fundamentally differently. Here’s why:
Probabilistic Nature: GenAI doesn’t follow deterministic rules. Its outputs are probabilistic, meaning certainty is impossible. A well-designed agent doesn’t need to be perfect; it needs to be good enough to meet business needs.
Evolving State: Unlike traditional software, agents are dynamic, capable of improving through training and iteration. This makes them more akin to human team members than static software tools.
No Final Endpoint: There is no “final version” of a GenAI agent. It’s an evolving system that requires ongoing tuning and adaptation to stay effective.
These distinctions demand a new approach to development—one that recognizes the unique nature of GenAI and prioritizes speed, flexibility, and business alignment.
Enter Evaluation Driven Design (EDD)
Evaluation Driven Design (EDD) is a methodology tailored to GenAI development. It shifts the focus from achieving perfection to reaching a mutually agreed-upon “good enough” standard. Here’s how it works:
Set Clear Evaluation Metrics: Before development begins, stakeholders—including business users—must define what success looks like. These metrics could include accuracy, response time, or user satisfaction scores.
Iterate Rapidly: Achieving a “good enough” level requires continuous iteration. Teams must gather feedback, refine the agent, and test improvements in real-world conditions at a fast pace.
Engage the Business Unit: Just as you wouldn’t hire a person, isolate them from their team, and provide only occasional feedback, you can’t develop a GenAI agent in isolation. The business unit that will use the agent must be actively involved in its design, testing, and evaluation.
Why Business Ownership is Non-Negotiable
GenAI agents aren’t just IT tools; they’re embedded in business processes. For this reason, the business unit—not IT—must take responsibility for testing, configuring, and signing off on the agent. Without this involvement, the agent is unlikely to meet the practical needs of its users.
To achieve this alignment, the organization must:
Foster Collaboration: IT provides the technical foundation, but the business unit ensures the agent is fit for purpose.
Adopt Flexible Operating Models: A rigid, waterfall-style approach will fail. Instead, embrace agile practices that allow for fast iterations and continuous improvement.
Empower Decision-Making: Business users must have the authority to define success metrics and make final decisions about deployment.
Control Is Essential for Speed and Success
Rapid iteration and business alignment are only possible with complete control over your GenAI estate. This means having the ability to:
Track and Evaluate: Monitor agent performance against agreed-upon metrics in real time.
Refine Quickly: Deploy updates, test new configurations, and incorporate user feedback without lengthy delays.
Ensure Trust and Compliance: Maintain transparency and control, especially in regulated industries.
Platforms like Great Wave AI are designed to provide this level of control, enabling businesses to innovate faster while maintaining trust and governance.
Closing the Operating Model Gap
While early platforms offer the technical foundation for GenAI success, many organizations struggle with the operating model required to implement EDD effectively. That’s where initiatives like The Centre for GenAIOps come in.
The Centre for GenAIOps is dedicated to helping organizations close this gap. By providing guidance, frameworks, and best practices, it enables businesses to:
Understand the unique demands of GenAI development.
Build operating models that support rapid iteration and business ownership.
Drive real value from their GenAI investments.
Conclusion
Generative AI represents a transformative opportunity for businesses, but realizing its potential requires a departure from traditional enterprise software practices. By adopting Evaluation Driven Design, empowering business units, and leveraging platforms, organizations can develop agents that deliver real-world value. And with the Centre for GenAIOps providing support and guidance, the journey to GenAI success becomes faster, smoother, and more impactful.