Moderne Unveils Prethink to Boost AI Code Governance
Moderne has launched Prethink, a new feature designed to provide AI agents with precomputed, structured knowledge about an organisation’s software estate before they generate code changes or recommendations.
Addressing Gaps in AI Code Applications
Prethink specifically targets gaps that arise when applying AI tools to large, mature repositories. These environments often consist of:
- Long-running services
- Dense dependency graphs
- Internal frameworks and conventions
- Complex build and deployment configurations
Many AI coding systems typically ingest code as files and text, leading models to infer architectural intent and operating constraints. This limitation can result in measurable costs. A study by the Model Evaluation and Threat Research (METR) group found that experienced developers were 19% slower using AI tools, primarily due to time spent reviewing and reworking outputs lacking architectural awareness. Prethink aims to reduce this review burden by surfacing system structure and constraints at the outset.
Structured Context for AI Agents
Prethink provides AI agents with a structured knowledge layer derived from code and architecture. Instead of reconstructing context from raw files during a session, agents begin with an explicit map of:
- Dependencies
- Service boundaries
- Configurations
- Other constraints
This innovation shifts token use from comprehension to analysis and implementation. According to Moderne, a shared knowledge layer can also reduce hallucinations and inconsistent results, enabling agents to reason against known relationships rather than infer from incomplete inputs.
Importance of Governance in Code Changes
In large organisations, architectural governance is as critical as syntactically correct code. Changes that compile can still violate design intent, create operational risks, or conflict with internal standards. Such errors can:
- Slow delivery cycles
- Increase review requirements
- Raise compute consumption due to repeated model revisions
Prethink leverages Moderne’s Lossless Semantic Tree code model, known for its compiler accuracy. This model captures types, dependencies, and metadata while exposing crucial information such as:
- Service boundaries
- Endpoints
- Dependencies
- Architectural relationships
The knowledge is stored alongside the code, available for both human and AI use, allowing teams to determine what facts and constraints are generated and shared, especially relevant for organisations with strict policies or regulated environments.
Platform Positioning and Future Directions
Prethink is part of Moderne’s broader platform for managing change across extensive, multi-repository codebases. The platform has focused on deterministic automation for codebase evolution, including structured refactoring and policy-driven change at scale. Prethink extends this approach into AI agent workflows, providing models with the same foundational representation used throughout the platform.
For engineering leaders, the challenge often lies not in generating code but in safely coordinating changes across multiple teams and repositories. In environments with thousands of repositories, even small misunderstandings can lead to significant failures. A change might impact shared libraries, service contracts, or internal platform components, or conflict with conventions not immediately apparent from individual files.
Moderne asserts that Prethink is built for these complexities, framing the AI agent as a participant in a governed delivery system rather than a standalone 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,” stated the CEO and co-founder of Moderne.
Prethink is now available on the Moderne Platform.
