Building machine-readable ontologies to properly constrain AI Agent outputs

On November 12, 2026, the AI section of DEV Community published a post titled “Building a formal ontology for AI agents instead of relying on long prompts.” It takes up the problem of invalid outputs that agents produce when they carry out business actions. The author argues that long prompts rarely constrain that behavior reliably, and that a machine-readable ontology encoding the rules of the business domain is the more dependable fix. In the post, the agent demo always works: a customer asks for a refund, the agent calls the right tool, and the money moves. Then it ships, and a week later it refunds an order that was never delivered, or it issues a refund larger than the purchase, or it approves an expense for a “manager” who has no reports. The author stresses that none of these are reasoning failures. The model reasoned fine; it acted on something that isn't true in the business, because nothing in the system told it what is true. Only this one report was available at the time of writing, and no follow-up coverage has appeared, so it is unclear whether any team or product has adopted this architecture in practice.