Intent Models
Before generated code can be trusted, the old system has to be decomposed into a model that explains what it meant to do.
Holonic Research
6 min read
Intent is extracted, not invented
Legacy systems rarely describe themselves cleanly. Their business intent is distributed across procedures, dependencies, data formats, exception paths, and operational habits. A modernization layer has to read that system before it rewrites it.
A queryable intent model gives engineers and AI systems a shared substrate. Instead of guessing what a program should do from a prompt, the modernization process asks questions of the source itself and carries those answers forward.
The model becomes the control plane
Once the intent model exists, it can guide generation, structure verification, explain evidence, and persist as a baseline. That continuity is what turns modernization from a one-time rewrite into a governed engineering practice.
Share this post
Stay in the loop
Get new writing on deterministic modernization, evidence, and governed AI.
Related articles

AI Can Rewrite Your Legacy System. It Can't Tell You What It Changed.
LLMs can generate legacy code migrations fast, but they can't prove what they changed or account for what broke along the way. Here's why sampling and coverage reports aren't proof, and what deterministic verification actually requires.

Reading generated code backwards
The reviewer’s question is rarely “does it compile?” It is “why did this line have to exist?”

Source-derived beats prompt-derived
Prompts are useful instructions. Source-derived intent is evidence. The distinction determines whether AI output can be trusted.
HOLONIC
The deterministic evidence layer underneath legacy modernization and the agentic enterprise.
© 2026 Holonic Technologies, Inc. · Atlanta, GA · Tucson, AZ
CodeIntent® is a registered trademark of Holonic Technologies.