Don Barger, Mark Lay, and Darren Taillie connect organizational boundaries to everyday AI operations in Practical AI’s second Midwest AI Summit compilation. Barger describes writing an AI charter before building: drafting, translation, research, and human-reviewed coding were permitted; deceptive impersonation, undisclosed monitoring, and replacing human counseling were off limits. That commitment helped him refuse an urgent request for an AI counselor, despite the ease of building a prototype. His faith-oriented chatbot also exposes a capacity problem: personally taking over critical conversations stopped being feasible as its audience grew. Lay keeps people accountable for data protection, output quality, agent behavior, and business outcomes. Department-built AI applications resemble Excel macros: teams gain useful tools while taking on maintenance obligations. He recounts a vendor upgrade that could not convert existing applications and asks what reversing a deployment would cost. Taillie supplies two deployed examples at AeroCore, a jet-engine cleaning company with a two-person analytics team. A text-to-SQL agent, which turns questions into database queries, uses MCP connections to tools and answers colleagues inside Microsoft Teams. A morning vision check compares aircraft photographs with work-order entries and flags discrepancies for human review. A noon digest adds operational notes and anomalies. Taillie reports canceling one monthly reconciliation meeting, saving five or six person-hours, because no incorrect work orders remained. Formal resolution tracking was still being built; inspecting truck settings with vision was a future project. These are practitioner reports without a controlled baseline or published error rates. Useful automation still needs named owners, maintained data, and enough human capacity to handle what it flags.
“We drew the boundaries before we wrote the code.” — Don Barger, 07:22
“Do they have the capacity to keep running what they deployed?” — Mark Lay, 28:35
“We are building in a resolution system as we speak.” — Darren Taillie, 43:49