Daniel Meppiel, who advises enterprise engineering teams at Microsoft and GitHub, published a decision framework on October 1, 2026 for starting an Agentic SDLC. He distinguishes four entry points, with the right one depending on what already works, what is missing, and the organization's capacity to own the topic. The goal is to build durable capability while delivering a fast return on investment.
First entry point: activate the native foundation. Copilot Code Review, Coding Agent, and Enterprise Managed Plugins bring shared capabilities to the whole enterprise with little to no custom engineering. The infographic specifies that reviews are enabled centrally across repositories and that approved plugins and shared skills roll out without individual installation on supported clients.
Second entry point: build up practices. HVE Core, Microsoft's open-source toolkit, offers engineering practices, agents, and skills, with a Research, Plan, Implement, Review loop. The idea is to draw on it and adapt it rather than inventing every workflow from scratch.
Third entry point: treat skills as software. As expertise in AI-assisted SDLC spreads and skills proliferate, the author recommends packages rather than copied files, with rigor in writing, packaging, versioning, and the supply chain. APM, an open-source agent package manager, makes it possible to create, validate, and compose packages, control sources, pin versions, and audit dependencies.
Fourth entry point: govern a catalog. Shared skills and plugins get owners, approval and lifecycle rules, up to retirement, and the organization defines its "golden path." Microsoft Copilot plugin registry, which publishes and governs plugins for Microsoft 365, is one option; GitHub Copilot support is noted as planned.
Skills are presented as the central building block: reusable instructions and resources that encode engineering and organizational expertise. No frontier model can infer this way of working, short of committing a considerable budget each time. Meppiel sees this as part of the foundation of enterprise-scale "Agentic Software Factories," and of any agent beyond the SDLC. His rule of thumb comes down to three sentences: start from the gap you have and what your resources let you handle, keep what works, add what you can maintain.