Artificial intelligence is entering CNC programming — but not in the way most people expect. The most valuable application of AI in CNC is not autonomous machining or self-optimising toolpaths. It is a much simpler and more immediately useful capability: translating natural language into exact geometry operations.

The Translation Problem

CNC programming requires precise geometric specification. "Create a point at the intersection of line L2 and circle C1, on the upper side" is a precise geometric instruction. But expressing it in operator syntax — Point/YLarge,Intof,L2,C1 — requires knowledge of the operator library.

An AI agent that can translate natural language into operator syntax removes this barrier. The programmer describes what they want in plain English. The agent translates it to the correct operator and executes it.

The Catalog-Driven Architecture

The XyzPilot Chat Agent uses a catalog-driven architecture. The agent is not taught to compute geometry — it is taught to translate commands into validated function calls that XyzPilot executes natively.

This separation is critical:

  • The geometry stays inside XyzPilot — computed by the exact analytic engine
  • The agent handles the interface — translating language to operators
  • The agent cannot introduce geometric errors — it only calls validated functions

"The geometry remains entirely inside XyzPilot. The agent becomes a translator."

Proposal-Controlled Execution

The XyzPilot Chat Agent uses a proposal-controlled workflow. Every geometry creation command is presented as a proposal before execution. The operator must approve the proposal before any geometry is created.

This is not a limitation — it is a safety feature. The operator remains in control. The agent cannot modify committed geometry without explicit approval.

Deterministic Routing

For structured commands that match the audited operator corpus, the agent uses deterministic routing — the tool and parameters are selected without consulting the AI model. This eliminates the risk of the model selecting an incorrect tool by similarity.

For example, "create point on L0 mid" is deterministically routed to Point/L0,Mid — not to any other operator, regardless of what the model might suggest.

What AI Cannot Do in CNC

It is important to be clear about what AI cannot do in CNC programming:

  • AI cannot verify that the geometry is correct — the operator must verify
  • AI cannot guarantee that the machining program is safe — the operator must prove out
  • AI cannot replace the operator's judgment about setup, tooling, and workholding
  • AI cannot compute geometry itself — it calls the analytic engine

The operator remains responsible for the final accuracy and safety of all machined components. AI is a tool that makes the operator more productive — not a replacement for the operator's expertise.

The Training Application

The most transformative application of AI in CNC programming may be training. The XyzPilot Chat Agent supports four training modes — Demonstration, Prediction, Construction, and Diagnostic — that allow new setters to develop operator proficiency through structured interaction with the AI.

A new setter can reach programming competency in weeks rather than months, with the AI providing immediate feedback and guidance at every step.