I’m Mitch, making AI a first-class participant in how we ship software.
2026 – present
The AI-Era SDLC
My current chapter is a three-person tiger team, chartered by senior leadership, with a deceptively large question: how does agentic AI reshape the software development lifecycle, and what do we build so the answer becomes practice instead of a slide? In practice that’s a harness, a context substrate, and the workflows that connect them, now serving a cohort of around 45 engineers and feeding the enterprise AI strategy.
Two convictions drive most of it. The first is that an agent is a model plus a harness (the guides, hooks, skills, tests, reviews, and judges around the model), and the harness is where the real engineering lives. Hooks always execute; instructions only get interpreted, so anything that has to hold gets enforced deterministically rather than left to a model’s discretion. The second is that context, not the model, is the load-bearing layer of agentic delivery. Karpathy’s writing on LLM wikis matched what I keep seeing in practice: AI output improves dramatically the moment an agent can reach the “pet knowledge” a team usually keeps in human heads.
And the throughline underneath all of it is a refusal to run on vibes. “AI made me 3x faster” is an anecdote, not a measurement. The interesting work is turning that into something falsifiable: replacing how it felt with evidence you could be proven wrong about.
2019 – 2025
Platform & Cloud Foundations
You don’t get to redesign the SDLC until you’ve lived in its hardest corners. Before the AI work, I spent most of six years on platform and developer-experience work at enterprise scale: PR-to-production governance that monitored 50,000+ repositories for separation-of-duties compliance and cut audit-prep time by roughly three-quarters; a network consolidation across about 2,000 VPCs that took seven figures a year out of cloud spend; a DNS migration that remediated 140,000+ security vulnerabilities at 100% uptime through the cutover.
Different decade of tooling, same instinct that runs through the AI work: make the hard, important thing also the easy, default thing. Compliance you don’t have to remember. Renewals that don’t need a human in the loop. Onboarding that goes from hours to minutes.
I’m not interested in pulling humans out of the loop. The target is complementarity: higher confidence, higher quality, lower cognitive load on the person doing the work.
That was true when the output was a pipeline nobody had to think about. It’s just as true now that the output is an agent.
Credentials
- B.S. Software Engineering, Iowa State University (Cum Laude, 2020). Cyber security minor.
- AWS Certified DevOps Engineer – Professional · May 2026
- AWS Certified Advanced Networking – Specialty · Aug 2025
- AWS Certified SysOps Administrator – Associate · Feb 2025
- AWS Certified Solutions Architect – Associate · Mar 2024
- AWS Certified Cloud Practitioner – Foundational · Nov 2021
Skills
Colophon
Built with Astro, Tailwind, and MDX; hosted on GitHub Pages. The source is public; see the link in the footer.