Bigger Possibilities, Better Questions: September 30 at UA Little Rock
What becomes possible when we stop asking only how AI can save time? And how do we know whether the systems we build are actually getting better? Those two questions connected our September 30 gathering at UA Little Rock.
AI in Production: What Arkansas Is Building Now brought the community together around two complementary perspectives. John McCrary of Ouachita Labs presented “Evals.” Owen Parker, CTO, and Kurtis Cochran, CFO, of Arkana Laboratories presented “Previously Impossible.” The evening paired bigger possibilities with a practical responsibility: understanding what we build, how we test it, and where human judgment belongs.
Evals: define what better means
John’s talk made evaluations approachable without treating them as an afterthought. An evaluation combines a model or agent, an environment, and a grader. The grader might be code, another model, or a person. What matters is choosing a test that tells you something useful about the work your system is meant to do.
His examples started close to real work: production transcripts, bug reports, support tickets, hand-written cases, and synthetic cases. For constrained outputs, code can provide a clear check. For more open-ended work, an LLM judge may help, but the scores still deserve scrutiny. Read the scored examples before trusting the grader.
The practical lesson was to make improvement measurable. Define the performance axes you care about, break a large problem into smaller pieces, and examine where added complexity helps or hurts. In an iterative optimization loop, make one change at a time and keep it only when the training and test results improve. A convincing demo is a beginning, not a substitute for a repeatable test.
Previously Impossible: look beyond efficiency
Owen and Kurtis invited the room to think beyond faster summaries and quicker drafts. Their presentation focused on rebuilding core processes and exploring work that an organization could not previously take on. The starting point was the business itself: its data, its constraints, its people, and the systems they depend on.
Arkana’s examples spanned revenue-cycle work, intake, physician applications, and logistics. The presentation connected ownership of the underlying system with opportunities to capture better data and support future capabilities. That is a different question from “Which tool should we buy?” It asks what could change when the people closest to the work can help shape the software around it.
One important distinction deserves to stay with the story: the billing-code automation example had been tested against de-identified cases and was not yet in production. Promising test results and a deployed system are different milestones. Keeping that distinction clear is part of responsible AI communication and responsible implementation.
Questions worth carrying forward
What would better performance look like for the work you actually need done?
Which real examples would reveal where your system succeeds or fails?
Where are you adapting your work to the software, rather than building software around the work?
What review and testing must happen before an experiment becomes something people rely on?
These are questions for developers, business leaders, researchers, students, and anyone helping a team make sense of AI. You do not need the same background to contribute to the conversation. You do need room to ask honest questions and learn from the people around you.
Thank you for being part of it
Thank you to John McCrary, Owen Parker, and Kurtis Cochran for sharing their work and perspectives. Thank you to Matmon, UA Little Rock, and Arkana Laboratories for supporting the evening, with special thanks to UA Little Rock for hosting. And thank you to everyone who made time to be in the room.
This recap is based on the presentations and photos supplied after the event. It is not a transcript or an attendance report. The questions above are prompts for our continuing community conversation.
For the original event details, see the September 30 event listing.
Next: AI, Distilled
On Wednesday, October 21, join Phil Brandon, founder of Rock Town Distillery, at Skyline Event Venue on the 30th floor of 400 West Capitol. AI, Distilled explores custom business software, mobile apps built from an existing website, and a local AI assistant. The gathering runs 5:00–7:30 PM Central, with free admission and free parking in the attached deck.
Bring your curiosity and someone you would enjoy exploring these ideas with. Reserve your free spot at AI, Distilled.
