Framing the Development Process in the AI Era
In the AI era of software development, the division of roles in which humans think and AI implements is rapidly becoming a reality.
What becomes crucial at this point is what the AI understands and what assumptions it makes when generating code.
Generative AI fills in ambiguous specifications and missing assumptions in ways that differ from human reasoning, often in a manner that merely appears plausible. As a result, unintended designs and even destructive implementations can easily emerge.
Over the next several installments, we will use the Unified Process (UP (Unified Process)) as a guiding framework to organize approaches to requirements, analysis, design, and implementation in the AI era, and to examine how project management and development processes should be restructured.
UP in the AI Era
The Unified Process (UP) is a culmination of software development process technologies as of around the year 2000, and can be regarded as one representative conclusion of that era.
Although there have been difficulties in applying UP directly in later software development practice, its technical framework and constituent techniques have been established as standard technologies, making it a solid foundation for discussing development processes.
In 📄 Rethinking the Development Process in the AI Era, we organized the key discussion points of software development processes in the AI era by contrasting UP with agile development.
Based on those results, from this point onward we will examine the constituent elements of software development in concrete terms, using UP as a guiding reference.
Redefining Iteration and Increment in the AI Era
Even in the AI era of software development, iteration and increment will continue to serve as fundamental principles.
However, as the primary agent writing code shifts from humans to AI, it is no longer sufficient to understand iteration and increment solely in terms of increases in code volume or functionality.
In traditional software development, iteration and increment have been understood as follows.
-
Iteration: a work cycle of build, observe, and revise.
-
Increment: the gradual increase of functionality and quality.
The primary artifacts produced through this cycle are models (requirements, analysis, and design) and program code.
In the AI era, an additional crucial element is introduced: the context shared with the AI.
In AI-era development, iteration and increment take on the following extended meanings.
-
Iteration: a loop in which the context given to the AI is updated and reinterpreted.
-
Increment: the stabilization and expansion of the “assumed context” referenced by the AI.
In AI-era development, the quality of generated programs is determined by how stable the assumptions and semantic structures referenced by the AI are.
It is therefore fundamentally important to establish a framework that improves the quality of AI-generated programs and prevents the need for backtracking.
Phases
In UP, phases are introduced as a framework to clarify the purpose and positioning of individual iterations. Development up to product release is managed by dividing it into the following four phases.
-
Inception PhaseA phase in which what the project aims to validate and which problem domains it addresses are defined, and the outline of the context shared by the AI and stakeholders is formed.
-
Elaboration PhaseA phase in which the context obtained in the inception phase is iteratively refined, removing ambiguities and contradictions, and establishing assumptions robust enough for high-speed generation by AI.
-
Construction PhaseA phase in which software is materialized by leveraging AI, based on the context and constraints established in the elaboration phase.
-
Transition PhaseA phase in which the developed software is transitioned into actual operation, insights from use and operation are collected, and feedback is provided to the next inception.
In each phase, functionality is actually developed and validated through iterations.
Phases define the objectives and allowable scope of iterations, while iterations are the units in which functionality is incrementally built within that framework.
This relationship does not change in the AI era; however, alongside code, the context and assumptions shared with the AI become critically important elements that are updated with each iteration.
Inception Phase
In UP, the inception phase is the phase in which the project’s objectives and scope are clarified, and its feasibility and value are evaluated.
At this stage, rather than fixing detailed specifications or designs, the emphasis is placed on
-
what is being built
-
why it is being built
-
what scope is being targeted
—sharing this overall picture among stakeholders is considered essential.
The inception phase has traditionally been positioned as an exploratory phase to gather the information needed to decide whether to proceed to subsequent phases.
Role of the Inception Phase in the AI Era
In the AI era, the inception phase takes on an additional meaning: defining the context to be shared with the AI.
In this phase, sharing context is prioritized over the completeness of specifications.
For the AI, this phase indicates “which world it is allowed to think about,” with the goal of outlining the problem domain, terminology, and underlying assumptions.
At this stage, ambiguity and unresolved elements are permitted, but it is necessary to clearly indicate which context they belong to.
As a phase that constrains the domain of thought so that the AI can reason and generate appropriately in later phases, the inception phase plays an even more critical role than before.
Elaboration Phase
In UP, the elaboration phase is the phase in which the system’s structural backbone is solidified by advancing analysis and design based on the requirements and directions organized during the inception phase.
At this stage, functionality is partially implemented and validated through iterations, while ambiguous requirements and inconsistencies are identified and corrected.
One of the key deliverables of the elaboration phase is the architecture baseline.
Once the architecture baseline is established, subsequent construction phases can proceed with routine implementation work that no longer requires design decisions.
As a result, large-scale, volume-driven development by programmers becomes possible in the construction phase.
Role of the Elaboration Phase in the AI Era
In the AI era, this role of the elaboration phase becomes even more critical.
During the elaboration phase, the context obtained in the inception phase is iteratively revisited and refined, eliminating ambiguities and contradictions while sharpening context, boundaries, and assumptions.
The central deliverable of this phase is the architecture baseline that the AI will reference.
Once the architecture baseline is fixed, the AI no longer needs to make design decisions and can function as an agent that generates implementations according to the given structures and constraints.
This can be seen as transferring the same structure that enabled volume-driven development by programmers in the UP era into the AI era.
The elaboration phase is positioned as the final groundwork phase for safely utilizing AI as a high-speed implementation workforce.
Construction Phase
In UP, the construction phase is the phase in which system development proceeds in a volume-driven manner based on the architecture baseline established during the elaboration phase.
At this stage, the architecture and major design decisions have already been fixed, and functionality is incrementally added through iterations in accordance with those frameworks.
In the construction phase, correct application of the established framework is prioritized over creative design ingenuity.
As a result, this phase is accessible to general engineers, and development speed can be more easily increased by scaling the size of the team.
Role of the Construction Phase in the AI Era
In the AI era, the construction phase retains this structure while replacing the primary implementation agent with AI.
By following the architecture baseline established in the elaboration phase, the AI generates large volumes of implementations and tests, advancing system construction in a volume-driven manner.
In this phase, the introduction of new concepts or design decisions is minimized, and generation faithfully reflects existing structures and constraints.
The AI functions as an implementation agent that does not deviate from the given framework, while humans focus on reviewing and fine-tuning the generated results.
In this way, the construction phase is centered on the architecture baseline and driven by volume, with the responsibility for that volume shifting from humans to AI.
Transition Phase
In UP, the transition phase is the phase in which the developed software is moved into actual operation, and delivery to users and operational launch are carried out.
At this stage, the product is refined into a production-ready state through bug fixes and adjustments.
The transition phase has traditionally been positioned as the final phase for completing the handover from development to operations.
Role of the Transition Phase in the AI Era
In the AI era, the transition phase plays an important role not only in moving the developed software into production, but also in capturing insights from use and operation as context for the next development cycle.
Constraints and assumption mismatches revealed during operation, as well as implicitly applied decision criteria, constitute new contextual information that the AI should reference.
The context obtained here is reused as input for the next inception phase, and development re-enters an iterative loop.
The transition phase is thus positioned not as the endpoint of development, but as a context-updating phase that initiates the next iteration.
Documenting various forms of contextual information during the transition phase will become far more critical than in traditional development.
Phase Structure in AI Development and the Significance of the Elaboration Phase
As we have seen so far, in AI-era development processes, the role of AI changes step by step across the phases of inception, elaboration, construction, and transition.
-
In the inception phase, AI acts as a dialogue partner for exploring the problem domain and sharing the outline of the context.
-
In the elaboration phase, AI supports the validation of contextual and model consistency, while humans establish the architecture baseline.
-
In the construction phase, AI functions as the primary agent that generates large volumes of implementation based on the established context and framework.
-
In the transition phase, AI organizes insights obtained from real-world operation and extracts context to be passed on to the next inception phase.
Among these phases, the elaboration phase occupies the most critical position.
In AI-era development, the elaboration phase is the core phase that determines the quality and stability of all subsequent phases.
While the construction phase is where AI delivers the most visible value, that value only materializes by harvesting the results produced in the elaboration phase.
The significance of the construction phase does not lie merely in replacing human implementation work.
When context and assumptions are thoroughly prepared during the elaboration phase, AI can generate large volumes of code and tests with consistent quality.
As a result, humans are freed from implementation tasks and can focus on context design and validation in the inception and elaboration phases.
In other words, the role of the construction phase is to invest volume on the groundwork built during the elaboration phase and rapidly reap its results.
Groundwork Required for a Viable Construction Phase
A construction phase of this nature cannot stand on its own.
For the construction phase to be viable, groundwork carried out during the elaboration phase is indispensable.
Above all, the establishment of the architecture baseline is the most critical factor.
The architecture baseline defines the fundamental framework of system structure, responsibility allocation, and dependencies, serving as the cognitive map that the AI references.
When this framework is fixed during the elaboration phase, the construction phase can proceed by accumulating implementation at scale, without making new design decisions, and strictly following the given structures and constraints.
This mirrors the structure in the UP era, where establishing an architecture baseline enabled volume-driven development by general engineers, now transposed into the AI era.
The architecture baseline is itself the boundary condition for AI-driven generation, and as the outcome of the elaboration phase, it represents the most essential prerequisite for advancing the construction phase safely and at high speed.
Summary
Using UP as a guiding framework, we examined project management and development processes in the AI era.
In the AI era, iteration and increment are not about piling up code, but about stabilizing and expanding context.
The phase-based framework becomes an effective means of adjusting the context of interaction with AI and enabling a high-speed construction phase.
References
Glossary
- UP (Unified Process)
-
A process model based on UML, characterized by iterative, use-case–driven, and architecture-centric development. It has derivatives such as the Rational Unified Process (RUP) and provides a foundation for practicing Component-Based Development (CBD).
- Development Process
-
A development process is the entire set of activities involved in building, deploying, and maintaining software systems. It encompasses planning, modeling, implementation, testing, and delivery as a systematic flow of development activities.
- Inception Phase
-
The phase in which the project’s goals, problem domain, and scope are defined, and the value and feasibility to be validated are clarified. In the AI era, its central role shifts from fixing detailed specifications to defining the outline of the context shared between humans and AI.
- Elaboration Phase
-
The phase in which the system’s structural backbone is established by advancing analysis and design based on the requirements and directions defined in the inception phase. In the AI era, its most critical role is to refine context, boundaries, and assumptions, eliminating ambiguity and contradiction, and to establish the architecture baseline referenced by the AI.
- Construction Phase
-
The phase in which the system is implemented at scale through iterations, based on the architecture baseline established during the elaboration phase. In the AI era, this phase refers to the stage where AI becomes the primary implementation agent, generating large volumes of code and tests with consistent quality under given structures and constraints.
- Transition Phase
-
The phase in which the developed software is transitioned into actual operation and delivered to users. In the AI era, it is positioned as a context-update phase that captures insights from usage and operation and feeds them into the next inception phase.
- deviation
-
The state in which a computed or observed value differs from a baseline or true value. Represents a measurable quantitative discrepancy.
- bug
-
A colloquial term referring to software problems. It has no strict technical definition and is often used broadly to cover Defects, Faults, or Failures.
- validation
-
Validation is the activity of confirming that a system or product fulfills its intended use and stakeholder requirements.