New capability helps AI agents reason more reliably and efficiently by shifting token use from code comprehension to actual development work.

Moderne, the enterprise code management platform for evolving large, multi-repository codebases, today announced the launch of Moderne Prethink, a new capability designed to give AI agents the structured software knowledge they need to operate reliably inside large, complex, and interconnected real-world codebases.

As AI agents become part of everyday software development, organizations are increasingly relying on them to reason about changes, trade-offs, and impact across complex codebases. But many teams are discovering a fundamental limitation: AI systems are being asked to make decisions without a shared, explicit understanding of how the software actually works. Industry analysts, including Gartner®, have observed that the “modernization value proposition has shifted from code translation to architectural intelligence and decision velocity.” Without consistent, authoritative context, even powerful models are forced to infer intent from partial data—leading to hallucinations, inconsistent outcomes, rising compute costs, and skepticism among engineering teams.

The disconnect: AI agents are now embedded throughout the development workflow, but software delivery velocity and quality are not improving at the expected pace. In fact, one study shows a 19% slow down of experienced developers using agents. Prethink is Moderne's answer to improving agent productivity.

Rather than feeding AI agents more raw data and expecting them to reconstruct system understanding on demand, Prethink provides a structured knowledge layer derived directly from an organization’s code and architecture. This knowledge is pre-computed and made available before an agent begins reasoning, so tokens are spent on analysis and change, not on rediscovering how a system works. The goal is not to replace AI decision-making, but to make it reliable, efficient, and usable in production environments.

“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 the ‘call a friend’ for agents when they need expert context about the code.

The problem with context-blind AI

Modern software systems are dense, interconnected, and shaped by years of accumulated decisions. Dependencies, frameworks, build configurations, and internal conventions all matter. Yet most AI tools treat codebases as collections of files rather than living systems, forcing agents to repeatedly scan, infer, and approximate how software behaves.

Without deeper context, AI agents may recommend changes that technically compile but violate architectural intent, introduce subtle risk, or fail to align with organizational standards. In regulated or large-scale environments, these mistakes slow delivery, increase review burden, drive up token consumption, and can block software delivery efforts entirely.

Prethink was built to address this.

A knowledge layer designed for AI agents

Prethink functions as a structured knowledge layer for AI agents, capturing explicit information about software estates and making it available as decisions are being made. This knowledge is derived from the compiler-accurate Lossless Semantic Tree (LST) code model that captures types, dependencies, and other metadata, exposing context such as service boundaries, endpoints, dependencies, and architectural relationships. This context is stored alongside the code and can be shared by both humans and AI agents. Instead of relying on inference alone, agents can reason against known relationships, constraints, and patterns that already exist within an organization’s systems.

Teams can customize what knowledge is produced and shared, deciding which architectural facts, dependencies, and constraints AI agents should take into account. By grounding reasoning in this shared, intentional context, Prethink helps ensure recommendations are consistent, explainable, and aligned with enterprise realities.

When agents reason from incomplete or noisy inputs, outcomes are noisy as well. Prethink improves the signal before reasoning begins.

Making AI useful at scale

Prethink builds on Moderne’s broader platform, which enables safe, large-scale change across complex codebases through deterministic automation. Prethink extends this approach by ensuring AI agents working alongside those systems operate with the same discipline, accuracy, and governance.

Rather than allowing agents to propose changes in isolation, Prethink helps them reason within a framework that reflects how software is actually built, maintained, and governed across an organization. This becomes increasingly important at scale. In environments with thousands of repositories and hundreds or thousands of developers, even small misunderstandings can cascade into significant operational and financial risk.

By anchoring AI decisions in structured, pre-computed knowledge, Prethink helps organizations move faster without sacrificing trust or control.

AI agents becoming practical partners

Prethink supports engineers and platform teams by improving the quality and reliability of AI-generated insights, not by making AI an unquestioned authority. The insights generated using Prethink remain reviewable, traceable, and grounded in explicit knowledge rather than opaque inference or probabilistic guesswork.

As organizations adopt more AI-driven tooling, the difference between useful automation and risky automation will increasingly come down to whether AI systems are grounded in reality. By giving AI agents a clearer, more accurate understanding of the systems they analyze, Moderne aims to make AI a practical partner in building new software on top of complex, real-world systems rather than another source of uncertainty.

Prethink is available today on the Moderne Platform. Read our detailed blog for more information.

Gartner, Predicts 2026: AI Potential and Risks Emerge in Software Engineering Technologies, 3 December 2025

GARTNER is a trademark of Gartner, Inc. and its affiliates.

About Moderne

Moderne is an agent tools company that enables AI coding agents to operate faster, more accurately, and more cost-effectively across enterprise software systems. Moderne helps organizations scale AI-driven engineering with organization-wide intelligence and governed execution, continuously improving software quality while maximizing the value of their AI investments. Based in Miami, Moderne investors include Acrew Capital, Intel Capital, True Ventures, Mango Capital, Allstate Strategic Ventures, Morgan Stanley Ventures, Amex Ventures, and TIAA Ventures. To learn more, visit www.moderne.ai.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260210110758/en/

Prethink shifts AI agent tokens from rediscovering system context to doing actual development work—making agent-driven engineering more reliable at scale.