Integrating AI into the Software Development Process

Modern coding agents are increasingly proficient, handling tasks from minor adjustments to significant engineering challenges, including writing code, creating multi-module tests, and preparing pull requests. Despite these advancements, the overarching software development process remains heavily reliant on human intervention at each stage. For instance, analysts collaborate with agents to define requirements, which are then passed to developers for refinement and the assignment of new tasks to coding agents. This scenario highlights a critical point: while agents can perform substantial work within a specific phase, the overall process continuity and integrity are still maintained by humans.

The international settlements team for Sber’s corporate clients identified this disconnect as a key area for the next phase of AI development. On a broader scale, the concept of transitioning to a unified, managed process—where humans set objectives and make decisions, agents execute tasks, and the environment provides context, rules, checks, and execution history—is extensively detailed in the AI-Disrupt PDLC guidance.

HG SDLC Orchestrator: The Connecting Link Between Development Stages

In response to these challenges, the HG SDLC orchestrator (Human Guided SDLC) was developed. The orchestrator aims to overcome the limitations of fragmented agent sessions that hinder scalability in development, establishing a unified connecting link between various software creation stages. This approach allows for shifting systemic constraints during the full automation of individual stages, ensuring deeper integration of AI into the process.

The practical implementation of the HG SDLC orchestrator focuses on creating an environment where automated agents can interact effectively throughout the entire development lifecycle, reducing dependency on manual transitions between stages. This tool enables executives and PMs to integrate AI more efficiently into the development process, concentrating on workflow and system constraints rather than the intricacies of coding itself.