About

I'm Ryan. I build production enterprise AI systems for Fortune 100 companies, and I've built data and ML systems inside enterprises since 2010. The model is never the hard part. Production turns on two things the model can't fix: whether people can trust the output, and whether the organization can climb the learning curve to use it.

Once those systems are live, another problem shows up that the model can't fix either: seeing and controlling what the agents actually do. That is what EdgePlane is, the open-source control plane I'm developing for AI agent fleets. Run more than one agent in production and the questions stop being about prompts and start being about operations: persistent identity, lifecycle ownership, inter-agent coordination, and the observability to reconstruct what happened when something breaks. Documentation at edgeplane.ai.

I work as a strategic forward-deployed engineer, embedding with teams across complex operating environments, from financial services and marketing to supply- chain operations. Most organizations still treat putting AI into production as primarily a model problem. In practice, it is an operating-model problem: how systems, workflows, governance, and people adapt around the technology. That is the layer where I work.

I also build and operate Aria, my personal AI agent fleet. I run it the way a surgeon runs an operating room: I make the decisions, the agents handle prep, checks, and the work I should not be interrupted to do. They all work from aria-rs, the single CLI that every agent, cron job, and human command drives with the same verbs, so there is one command surface and one log of who did what. Aria is where I test some of the patterns I write about.

The same instinct keeps showing up in adjacent work. ayx-rs is an agentic CLI substrate that brings agent-style orchestration to a heavyweight enterprise data platform: reusable operator actions, explicit state, and repeatable, scriptable runs. Curio is git-native knowledge curation for material that scatters across Slack, docs, and files. It keeps that material in plain files you own. Different surfaces, same move as EdgePlane: give the work real structure instead of trusting it to memory and good intentions.

I'm PhD trained in Economics and Organizational Psychology from Claremont Graduate University, with a BS in Business Management from CU Boulder. Whether an organization actually adopts an AI system comes down to fundamental human decisions and behaviors, such as trust, incentives, and how people act under uncertainty, not model quality. That is the part most teams underestimate, and the part I was trained to read.

I write here for practitioners doing the actual work. Not tutorials. Field reports, research synthesis, and honest assessments of what works, what does not, and why the gap between AI demos and production systems persists. Recent pieces: Own the Operation, The AI Productivity Dip Is Longer, Deeper, and Diverging, and Why Agents Need a Control Plane, Not a Pipeline. New work goes up on LinkedIn.

I build in the open with EdgePlane. Where to find me is on connect.