"His central message was that reliability comes from combining probabilistic components with deterministic boundaries. Erickson argued agentic AI becomes more interesting when it is treated as a layer over real operational systems rather than a replacement for them. The model can interpret questions, retrieve evidence, classify situations, and suggest actions. Deterministic systems execute the actions, enforce the constraints, and provide the telemetry that allows the whole loop to be evaluated."
"He described a common trap in natural language to SQL and similar query generation patterns. The first few demos work because the questions are simple and the schema is small. Accuracy falls sharply when the schema is complex and the query space includes many joins, edge cases, or overloaded fields. One mitigation he emphasized was reducing degrees of freedom: flatten the schema, constrain the query forms, and treat expressiveness as a cost that must be paid for with more evaluation and more safeguards."
Agentic AI should be treated as an engineering problem requiring probabilistic model components bounded by deterministic systems to ensure reliability. Agentic layers should integrate with operational systems, where models interpret queries, retrieve evidence, classify situations, and suggest actions while deterministic components execute actions, enforce constraints, and provide telemetry for evaluation. Natural-language-to-SQL and similar query generation suffer accuracy loss as schema complexity and query space expand, so reducing degrees of freedom—flattening schemas, constraining query forms, and accepting expressiveness costs—improves reliability. Classification maps well to small known categories; code generation over large search spaces increases error rates. Tool catalogs and interfaces must be designed to avoid a 'paradox of choice.'
Read at InfoQ
Unable to calculate read time
Collection
[
|
...
]