Moderne unveils Prethink to boost AI code governance

Moderne has launched Prethink, a new feature that gives AI agents precomputed, structured knowledge about an organisation's software estate before they generate code changes or recommendations.

Prethink targets gaps that emerge when AI tools are applied to large, mature repositories. These environments often include long-running services, dense dependency graphs, internal frameworks and conventions, and complex build and deployment configurations. Many AI coding systems still ingest code largely as files and text, leaving models to infer architectural intent and operating constraints.

That limitation can carry a measurable cost. A study by the Model Evaluation and Threat Research (METR) group found experienced developers were 19% slower when using AI tools, largely due to time spent reviewing and reworking output that lacked architectural awareness. Prethink aims to reduce that review burden by surfacing system structure and constraints up front.

Structured context

Prethink provides AI agents with a structured knowledge layer derived from code and architecture. Agents start with an explicit map of dependencies, service boundaries, configurations, and other constraints, instead of reconstructing that context from raw files during a session.

According to Moderne, this shifts token use away from comprehension and towards analysis and implementation. The company also says a shared knowledge layer can reduce hallucinations and inconsistent results by letting agents reason against known relationships rather than infer from incomplete inputs.

In large organisations, architecture and governance can matter as much as syntactically correct code. Changes that compile can still violate design intent, create operational risk, or conflict with internal standards. These errors can slow delivery cycles, increase review requirements, and raise compute consumption when teams repeatedly ask models to revise output.

Code model

Prethink draws on Moderne's Lossless Semantic Tree code model, described as compiler-accurate. It captures types, dependencies, and metadata, and exposes information such as service boundaries, endpoints, dependencies, and architectural relationships.

The knowledge is stored alongside the code and can be used by both humans and AI agents. Teams can decide which facts and constraints are generated and shared, which may matter for organisations with strict policies or regulated environments.

Moderne also pointed to analyst commentary that emphasises architectural intelligence and decision speed over simple code translation in modernisation work. On this view, the challenge for AI in software engineering shifts from producing snippets to working reliably within real systems shaped by years of accumulated decisions.

Platform positioning

Prethink sits within Moderne's broader platform for managing change across large, multi-repository codebases. The platform has focused on deterministic automation for codebase evolution, including structured refactoring and policy-driven change at scale. Prethink extends that approach into AI agent workflows by giving models the same underlying representation used elsewhere in the platform.

For engineering leaders, the challenge is often less about generating code than about safely coordinating change across many teams and repositories. In environments with thousands of repositories, small misunderstandings can cascade into wider failures. A change might affect shared libraries, service contracts, or internal platform components, or conflict with conventions that are not obvious from individual files.

Moderne says Prethink is built for those conditions, framing the AI agent as a participant in a governed delivery system rather than an isolated assistant making suggestions without full context.

"Moderne is evolving into an agent tools company, and it's our mission to be your AI agent's best coding partner," said Jonathan Schneider, CEO and co-founder of Moderne.

Prethink is available on the Moderne Platform.