AI powered Software Development Life Cycle at Microsoft
| Barbara McElnea | Ryan McDonald | Tom Fanelli | Joseph Hindo |
Episode #294
Introduction
In episode 294 of our SAP on Azure video podcast we talk about Software Development Life Cycle at Microsoft with AI.
I have always been pretty impressed by the way how Microsoft pushes the way how to develop forward. Especially when we look at the SAP world Microsoft is actually quite a big customer with hundreds of internal ABAP developers. In the past we already talked about how we do Software Development Life Cycle at Microsoft, but obviously AI is also becoming more and more important.
So for today I am really happy to have Barb, Ryan, Tom and Joseph joining us today to tell us more how we do AI Powered SDLC at Microsoft!
Find all the links mentioned here: https://www.saponazurepodcast.de/episode294
Reach out to us for any feedback / questions:
- Goran Condric: https://www.linkedin.com/in/gorancondric/
- Holger Bruchelt: https://www.linkedin.com/in/holger-bruchelt/
#Microsoft #SAP #Azure #SAPonAzure #Copilot #ABAP #SDLC
Summary created by AI
- AI-Driven SDLC Transformation at Microsoft:
Barbara, Ryan, Tom, and Joseph described how their SAP engineering team at Microsoft is leveraging AI to compress traditional application build cycles, notably through Project Tempo, reducing a two-year build cycle to one year and accelerating development, testing, and migration processes.
- Project Tempo Overview: Barbara explained that Project Tempo is a suite of applications being built for Microsoft’s finance systems, with the team using AI to compress the typical two-year build cycle into a one-year cycle, including design, build, and migration phases.
- AI Integration in SDLC: The team utilizes AI to capture context from conversations, specs, and wireframes, enabling rapid unit testing and large-scale application delivery, while maintaining governance and managing risk, especially in brownfield implementations.
- Human-in-the-Loop Approach: Barbara emphasized that the current process is agentic with humans in the loop, ensuring oversight and quality as AI assists in automating and scaling SDLC tasks.
- Accelerated Testing and Migration: By month five and a half, the team was already in the testing phase, demonstrating the effectiveness of AI in speeding up the SDLC and migration work.
- AI-Powered Contextualization and Knowledge Management:
- Joseph and Ryan discussed how the team leverages meeting recordings and contextual artifacts to document key design decisions, which are then used to inform specs, code, and future development, creating a robust knowledge pool for ongoing and future projects.
- Recording Key Design Decisions: Joseph described the early adoption of documenting key design decisions from meeting recordings, storing them centrally to inform specs and code references, which proved invaluable for future tool versions and development cycles.
- Building a Knowledge Pool: Holger highlighted that any customer can build a similar knowledge pool by collecting meeting recordings, which can be used to enrich context and drive development efficiency.
- Contextual Artifacts in SDLC: Ryan explained that contextual artifacts, such as recorded team meetings and specs, are used to automate spec creation, build code, and iterate on development, ensuring that all relevant information is available for testing and triaging issues.
- Development Tools and AI Skills Integration:
- Ryan, Tom, and Barbara detailed the integration of tools like GitHub Copilot, Azure DevOps, and ABAP DevTools (ADT) with AI-driven skills, enabling local coding agents to automate and scale development, testing, and code analysis across Microsoft’s SAP landscape.
- GitHub Copilot and Local Coding Agents: Ryan described the use of GitHub Copilot accessed through various IDEs and local coding agents, which ground themselves in context and autonomously execute tasks, with user safeguards and approval steps.
- Azure DevOps and MCP Integration: The team uses Azure DevOps for planning, execution, and deployment, integrating it with MCP catalogs and work IQ for seamless access to the 365 ecosystem and modern authentication.
- ABAP DevTools and ADT APIs: Ryan explained how the team interacts with ADT APIs for reading, analyzing, and crawling ABAP code, enabling both traditional and modern AI-accelerated ABAP development.
- Repository of AI Skills: Barbara and Tom noted the growth of their repository to 22 AI-driven skills, which are shared among team members to ensure repeatable and deterministic outcomes in development and testing.
- Code Analysis, Refactoring, and Technical Debt Reduction:
- Barbara challenged the team to use the ADT crawler to analyze code packages, identify duplicate and orphaned CDS views, and reduce technical debt, with Ryan and Joseph describing how these audits are scheduled and drive quality improvements at scale.
- ADT Crawler for Code Analysis: Barbara tasked the team with using the ADT crawler to read all packages, analyze CDS views, and identify duplicates and orphans, resulting in the discovery of 26 duplicate CDS views and 67 orphaned views.
- Technical Debt Management: The crawler’s findings enabled the team to burn down technical debt by merging duplicates and addressing orphaned code, ensuring clean core principles and reducing future maintenance burdens.
- Scheduled Audits and Continuous Quality: Joseph explained that these code audits are scheduled to run independently, continuously exposing quality issues and prompting teams to resolve them, thus maintaining high standards across development streams.
- Impact Analysis and Dependency Graphs: Ryan described how dependency graphs and impact analysis are generated from crawler data, helping teams understand the effects of code changes and facilitating efficient refactoring and optimization.
- Visualization and Artifact Generation for SDLC:
- Ryan and Joseph showcased how AI-generated HTML dashboards and 3D visualizations provide granular insights into application architecture, skill impact, and development progress, supporting decision-making and storytelling within the SDLC.
- HTML Output Artifacts: Ryan explained the shift towards HTML output artifacts, which offer granular, visual storytelling of development results, making it easier to share and interpret outcomes across teams.
- 3D Tempo Stack Visualization: The team uses 3D modeling techniques to visualize application architecture, dependency graphs, and managed wrap models, aiding in architectural reviews and understanding inner dependencies.
- Dashboard Integration: Dashboards consume results from all AI skills, providing a global view of skill impact, repeatable activities, and deterministic outcomes, with artifacts pushed to SharePoint for broader accessibility.
- Governance, Risk Management, and Project Fidelity:
- Tom described how work IQ and AI agents automate governance tasks, risk analysis, dependency management, and project plan updates, ensuring fidelity between conversations, documentation, and code, and enabling 360-degree project oversight.
- Automated Risk Analysis: Tom explained that AI agents identify, update, and close risks by analyzing meeting transcripts and conversations, automatically documenting and assigning actions in Azure DevOps.
- Dependency Management: Technical dependencies are mapped from conversations and integrated with project management tools, allowing for real-time updates and improved oversight.
- Project Plan Updates: Agents update project plans based on meeting discussions, reducing manual effort and ensuring plans reflect current migration and conversion strategies.
- Fidelity Checks: Agents review transcripts, chats, and emails to ensure decisions are tracked and implemented, preventing missed actions and maintaining accurate, real-time artifacts.
- Future Directions: Data Conversion and Collaborative AI Environments:
- Joseph and Barbara outlined the next phase of the project, focusing on data conversion, OData agent interactions, and evolving towards a shared AI environment where local coding efforts are consolidated for greater organizational benefit.
- Data Conversion Challenges: Barbara noted the transition from ABAP crawler to OData agent interactions for data migration and reconciliation, highlighting ongoing work in unit testing and data conversion.
- Collaborative AI Environment: Barbara discussed the need to move from isolated local coding agents to a shared environment where token usage and AI skills are consolidated, enabling collective benefit and a unified ‘brain’ for the team.
- Scaling and Uniformity: Joseph emphasized the goal of building skills into every SDLC component, scaling the approach, and collaborating with other groups to drive uniformity and quality improvement across the organization.