The blog post by a software development and AI coach from Cheezy World chronicles the journey of integrating Artificial Intelligence (AI) into the software development process at a multi-billion dollar company. The narrative focuses on a practical case study involving one of the client’s product groups, which comprises three delivery teams. Each team is manned by developers, testers, and a product owner, responsible for managing an Angular UI and a Spring Boot services layer.
The exploration into AI-driven development began after extensive discussions with both management and the technical teams, highlighting an openness to investing in advanced, paid AI tools rather than relying on free versions. To facilitate this exploration, two trusted developers (one specializing in Angular and the other in Spring) were selected to spearhead a pilot experiment under the guidance of the blog author. This setup, including necessary tool installations, took about two months of planning and was followed by a two-month experimentation period, with a midpoint evaluation scheduled after the first month.
The selected tool for this experiment, Claude Code by Anthropic, was chosen for its suitability for software development tasks. Before diving into AI-assisted coding, the team established an “AGENTS file”— a configuration guide that helps the AI understand and adhere to the company’s coding standards. The experimental process involved the AI (referred to as ‘the Agent’) suggesting code modifications, which the team meticulously reviewed in bi-weekly sessions. The developers would reject or accept changes, instruct the AI to make adjustments, and continuously refine the AGENTS file to improve future outputs.
Over the course of the experiment, the AI progressively handled more complex tasks and demonstrated a high level of code quality, including superior organization and 90% test coverage. These advancements were not integrated into the main codebase right away; instead, pull requests were created merely to showcase the potential of AI-driven outputs to other company developers, which sparked broader interest and discussions within the company.
Following the success of the initial exploration, one of the three teams underwent a focused workshop involving AI tools to prepare them for a two-week sprint using AI for actual feature development. The team retained its standard workflow, the only change being the integration of the AI in coding tasks. Remarkably, the workload usually set for two weeks was completed in just three days, though this rapid progress also exposed some workflow bottlenecks, particularly in testing, due to the accelerated development pace.
As more teams adopted the AI tools, they realized significant improvements in code and test quality. Traditional testing roles shifted towards focusing on critical updates and non-functional requirements, foreseeing a potential future where AI covers a broader range of testing tasks. Product owners also found AI helpful in refining tasks, a revelation that led to a new AI-assisted process for backlog management. This involved working closely with AI to produce a vast array of mostly accurate tasks, then fine-tuning specifics in tandem with ongoing development.
Throughout this transformative process, two primary constraints emerged: the high pace of changes overwhelming the teams, and the challenge of managing code merges effectively. The teams learned that rigorous standards and frequent adjustments, made feasible by AI’s efficiency, were instrumental in mitigating these constraints. Importantly, the AI’s ability to predict necessary file changes provided a strategic advantage in planning tasks to avoid merge conflicts.
The blog post wraps up by returning to the initial thought—why do experiences with AI in software development vary so significantly? While the author doesn’t provide a definitive answer, the documented success of this case study stresses the importance of structured, iterative engagement with AI tools, proper training, and an openness to reshaping workflows around the capabilities and efficiencies offered by AI. The coach’s insights serve as anecdotal evidence that with the right approach, AI can significantly boost software development productivity and quality.
Read the full post on cheezyworld.ca
Do You Want To Learn Scrum & Agile?
Learn Scrum & Agile at lost cost and online. Our 5-star rated Ultimate Scrum & Agile eLearning Courses can take you from beginner to advanced at your own pace.
Prepare and practice for the assessments from the major Scrum & Agile providers. Our 5-star rated Ultimate Scrum & Agile Practice Assessments will help you gain certification.
About TheScrumMaster.co.uk
