Day 1 – Breakout Session 3 – 3:30-4:00 PM
Room: Wildfire
Beyond the sandbox: when boring AI starts winning
Ryon Brewer · RizeCon 2026 Pocatello
73% of small businesses have tried AI. 61% don’t train their teams on it. The result is the same in most companies: every employee uses it differently, nothing standardizes, and the experiment quietly dies. Ryon Brewer — 20 years deep in enterprise tech, now founder of Vir Technology — walked the room through three real client deployments where the AI wasn’t a demo. It was boring. And it was saving six to twelve hours per employee per week.
Most AI conversations in small business are still science-fair conversations: which model is best, which tool is hottest, what the new feature does. Ryon’s argument is that all of that is the wrong starting point. AI starts working when it stops being interesting — when it disappears into the workflow and starts producing measurable savings nobody talks about anymore.
What he covered
Start with the problem, not the tool. The single message Ryon returned to throughout the session. Companies that succeed with AI map their workflows first, pick the one process where AI can move the needle, pilot it for 90 days, train the team on it, and only then scale. Companies that fail start by buying ChatGPT seats and hoping the team figures something out. The order matters.
AI as a first-draft accelerator, not a magic wand. Ryon’s framing for what AI actually does in production: it handles roughly 70% of a defined task, and the team refines the remaining 30%. This sets expectations honestly and is what makes adoption stick. The teams that expect 100% get burned; the teams that treat AI as a draft layer get compounding leverage.
RAG vs. model training. Two different builds for two different problems. Retrieval-augmented generation (RAG) lets a general model pull from your company’s documents at query time — fast to deploy, lower investment, good for knowledge work. Training a model on your data is heavier, slower, and locally hostable — the right call when the data is sensitive enough that it can’t leave the building. Ryon’s accounting client used the trained-model approach for exactly this reason.
The 90-day pilot discipline. Pick one workflow. Build the agent. Run it for 90 days against a known baseline. Measure hours saved, error rates, throughput. Then decide whether to scale, kill, or rebuild. Most companies skip this and either over-commit on hype or abandon a working pilot too early because the savings looked invisible at week three.
LangChain and LangGraph as the build layer. The frameworks Ryon’s team uses to build operational agents — not the consumer tools you’ve seen, but the engineering layer underneath them. He noted he’s not a fan of n8n. The distinction matters because business owners often equate “AI” with whichever chatbot they’ve used; the agents driving the case-study results live a layer deeper than that.
What attendees got
Three real AI case studies presented with the actual numbers, not testimonials. The HVAC company: an agent that builds quotes from years of historical estimate data — the owner saves six hours a week and closed estimates rose 23%. The accounting firm: a locally-hosted LLM (nicknamed “Gary”) plus client-onboarding automation — over $1M in annual savings on a $20M business. The training company: personalized AI-generated learning paths — onboarding cut from five weeks to 12 days, completion rates climbed from 22% to 73%. Each case followed the same map-pilot-train-scale sequence. Attendees can connect with Ryon’s Vir Technology team for AI roadmapping engagements that begin with workflow mapping rather than tool selection.
One story that landed
The accounting firm Ryon worked with was a $20M business drowning in client-onboarding paperwork. Their data was too sensitive to send to a cloud model, so the team trained a local LLM — internally nicknamed Gary — on years of the firm’s own documents and processes. Gary handles the first draft of every client onboarding packet. The team refines it. Result: more than $1M in annual labor savings. The unsexy version of that story is what makes it work — no demo days, no science fair. Just one workflow, one trained model, one team that knows how to use it.
“Start with the problem, not the tool.” — Ryon Brewer
“It’s boring. That’s when you know AI is actually working.” — Ryon Brewer
About the speaker
Ryon Brewer founded Vir Technology, an AI implementation firm for small and mid-market businesses based in Rexburg, Idaho, after 20 years in enterprise tech that included running an Oracle implementation consultancy he sold in 2018. He has also served as CEO of the Idaho Entrepreneur Center, where he runs the Entrepreneur Edge, Bootcamp, and Launch programs for startups across East Idaho. His Vir Technology team builds operational agents and trained-model deployments using LangChain and LangGraph.