The Oakland Ballers let an AI manage the team. What could go wrong? (techcrunch.com)

The Oakland Ballers, a minor‑league team formed in response to the A’s departure, ran a live experiment letting an AI manage a game in real time. Partnering with AI firm Distillery, they fine‑tuned OpenAI’s ChatGPT on more than a century of baseball data — including Ballers games — so the model could make per‑pitch managerial decisions like pitching changes, lineup construction, and pinch‑hitting. During the game the AI’s calls matched what manager Aaron Miles would have chosen, and Miles only overrode it once to replace a sick catcher, underscoring that the system functioned largely as an optimization and pattern‑recognition tool rather than a source of new strategy. For the AI/ML community this is a compact, practical case study in deploying LLMs for closed‑loop, high‑frequency decision support: it demonstrates feasibility of fine‑tuned models making operational calls in a live setting, the value of rich historical datasets, and the continued need for human‑in‑the‑loop oversight. It also highlights non‑technical risks — fan distrust, cultural pushback, and concerns about corporate optics — reminding practitioners that adoption depends as much on social acceptance and safety testing as on model accuracy. The Ballers’ experiment shows promise for AI as a tactical assistant, but also that real‑world pilots must anticipate human judgment, ethics, and community reaction.