AI can write code. What does it take to ship?Join AI DevCon, Nov 2-4
Book a Demo
CareersPricingDocs
Log inBook a Demo
Daniel Meppiel

SPEAKER

Daniel Meppiel

Software Global Black Belt @Microsoft

Hi, I'm Daniel. I build the tooling and the playbooks for what I call the Agentic SDLC: what actually changes in software engineering when agents do the work. I wrote the Agentic SDLC Handbook, vendor-neutral and free at danielmeppiel.github.io/agentic-sdlc-handbook. I created and maintain APM, an open-source package manager for agents that global organizations have adopted for their new Agent Supply Chains. At Microsoft I lead this work across EMEA, mostly inside very large, very regulated engineering orgs where the fun meets business reality. Before that, I worked at GitHub and SonarQube, plus co-founding my own startup as CTO.

LinkedIn

Upcoming sessions

See complete program

4 November

10:05 AM · 30 min

52AConference Session

Version Your Factory, Not Just Your Code

Most agentic software factories begin as a collection of prompts, scripts, and agent skills. That works until a model changes, the team adopts another harness, or a security rule requires changes across the entire factory.

What if we treated the factory like the software it produces?

Its capabilities would become composable modules. They would have versions, dependencies, tests, owners, and release notes. Once factory capabilities become versioned components, they can be tested, reviewed, shared, replaced, and composed. Teams could upgrade one part without rewriting the rest. Factories could be improved through the same open-source practices that built the rest of our software ecosystem.

Agent Plugins 1.0 gives us a starting point by making Skills and MCP servers easier to move between harnesses. The factory sits above them: the process remains stable while Copilot, Claude Code, Codex, and other harnesses can be swapped underneath it.

The same principle applies to governance. A factory can accept reusable and swappable rules about tools, risk, and required checks. Governance determines which versions and capabilities may run.

I’ll demonstrate a prototype factory built from versioned modules, run it through two harnesses, replace one capability without changing the workflow, and show how guardrails as code allow agents to ship autonomously without crossing their limits.

Agent Plugins 1.0 is not a complete factory specification. That is exactly why this conversation and more open standards are needed.