Three in a Box as the future delivery model?

The rise of GenAI has happened so quickly that it is easy to forget where we started. In software engineering, we have spent decades refining development processes to suit a clear relationship between humans and machines. We wrote code manually, and when we got it right, the system behaved exactly as intended. A straightforward transaction with well-defined roles and clear boundaries between instruction and execution.

This human-shaped approach to software engineering has, to put it mildly, changed. While no one yet has all the answers to the many questions GenAI raises, including ourselves, we do have some insights to share that can help companies better understand where we are today – and, just as importantly, prepare for a future where GenAI agents take on new and greater responsibilities in the engineering process.

A virtual team member

Right from the get-go, some of the most compelling use cases for GenAI have been in coding. GenAI can accelerate tasks like writing code, testing, and documentation with remarkable effectiveness. But what we have explored in recent months goes beyond simple assistance: GenAI is no longer just acting as a coding assistant; it is starting to function as a virtual team member.

Take this as an example: we recently used GenAI to facilitate the requirements elicitation and analysis phase of a new project, helping us identify, refine, and clearly define both functional and non-functional requirements. We ended up with a strong, detailed specification document. We then took that output and fed it back into the AI as the foundation for subsequent user story creation, with specific instructions to avoid overlaps. Once again, the results were surprisingly good. Finally, we handed the user stories to our preferred GenAI model, which produced the actual code, completing the flow from problem definition to working code.

New Box model

This brings us to our next point, and bear with us on this slightly philosophical note: The actual act of writing code is rapidly shrinking. If we accept that GenAI will continue to take on a growing number of responsibilities in software engineering, we must also accept that our current ways of working need to change. It is about rethinking the frameworks we rely on, how we educate and train developers, how we structure our teams, define user stories, and so on. In short, GenAI challenges the very foundation of how we approach the software development process.

The traditional ‘Two in a Box’ model, with a business lead and a tech lead working side by side, has long been the standard delivery setup in software engineering. But looking ahead, it may be timelier to consider a ‘Three in a Box’ model, where GenAI takes on a greater share of execution tasks across the Software Development Lifecycle.

Software Engineering Project Delivery Flow

As a starting point, we recommend that companies focus on integrating GenAI into their existing software delivery models. At Deloitte, we have developed a Software Engineering Project Delivery Flow framework that incorporates GenAI across the entire process from initial planning through to final release.

The framework illustrates how, during project planning, GenAI can improve how resources are allocated by analyzing needs and suggesting more effective ways to assign time and people. As development progresses, GenAI can help reduce manual effort by generating clear and consistent documentation, and it can assist product teams in identifying and writing well-structured user stories based on business goals.

When coding begins, GenAI can generate actual code, allowing developers to spend less time writing from scratch and more time reviewing and refining. It also plays a role in testing by generating test cases and catching potential bugs early. In parallel, GenAI can support risk management by highlighting technical concerns before they become issues. And toward the end of the process, it can help coordinate the product release, ensuring all components are aligned and ready for deployment.

Prepare for Agentic AI

Our next recommendation takes a longer view on how organisations should prepare for the future of software engineering.

Agentic AI is no longer a distant concept; it is a realistic development, where AI systems can act independently, make decisions, and interact with other systems or agents without constant human input. If you believe Agentic AI will play a role in the future, it makes sense to start building these capabilities within existing teams. It will help you understand where human input is still critical for problem-solving and creativity, and where AI agents can take over the execution.

Start working with the technologies already available today, so when Agentic AI starts scaling, you are part of the journey.

Build on existing setups

Our final recommendation is to make access to GenAI tools as simple and immediate as possible. Developers need practical, hands-on experience to truly understand how GenAI can support their work. When access is straightforward, teams naturally start to experiment and uncover new opportunities on their own. In the same spirit of speed and accessibility, focus on building on what has already been approved within your organisation, using existing compliant solutions to bypass time-consuming approval processes.

As always, however, flexibility must align with your company’s risk appetite. Organisations should carve out dedicated spaces where developers can explore and test ideas freely, while keeping clear guardrails in place. Striking the right balance between freedom and control allows creativity to flourish without compromising safety.