Monorepos, Synthetic Monorepos, and the Next Step for AI Coding Agents
Monorepo AI agents are AI coding tools that operate on a unified view of many projects—either in a single physical monorepo or in a synthetic monorepo built from multiple repositories—so they can make consistent, cross-repository code changes with shared dependency management and portable memory instead of working in narrow, isolated repos with no long-term context. The key takeaway is blunt: until agents see and remember code across boundaries, they will stay as smart autocomplete rather than true collaborators. Monorepo architectures and synthetic monorepo layers are emerging as the missing infrastructure that gives agents the context they need to act autonomously on complex systems. The story here is not about build tools; it is about reorganizing code and knowledge so agents can participate in real multi-repository coding, where dependency management and AI code context are first-class design concerns.
Block’s 450-Repo Migration: Why Consolidation Matters for Humans and Agents
Block’s decision to migrate around 450 JVM repositories for Cash App and Square into a single monorepo was not a tooling vanity project; it was a hard admission that the polyrepo status quo had broken down under scale. Dependency management in the old model had become “unmanageable,” with constant breaking changes, runtime failures, and library rollouts that demanded heroic cross-team coordination. By consolidating into one codebase, Block reduced dependency version drift, duplicated upgrade efforts, and diamond-dependency surprises while simplifying cross-service development and improving dependency visibility. The resulting monorepo now supports approximately 8,800 builds per week, with p90 CI times of around 10 minutes on an often-green main branch. According to Block, “what started as a complex, large-scale migration became a step-change in developer experience: a modern, cohesive codebase with an optimized IDE workflow, dramatically faster CI, and a foundation built for long-term velocity”. This is exactly the kind of unified environment in which monorepo AI agents thrive, because they see the entire dependency graph and can reason about atomic updates instead of juggling brittle versioned libraries.
From Physical to Synthetic Monorepos: Nx Polygraph’s Shared Context Layer
Of course, not every organization can, or should, cram hundreds of services into a single git repository. This is where synthetic monorepo ideas become important. Nx’s Polygraph connects multiple repositories—internal and open source—into a synthetic monorepo that AI coding agents can work on as if it were a regular monorepo. It does not move code; it builds a dependency graph that maps who publishes which packages and which APIs they define and consume. That graph gives agents multi-repository coding powers: they can change the producer of an API in one session and update every consumer because all of the code is accessible through the synthetic monorepo. Crucially, Polygraph adds portable, shared memory. Every conversation an engineer has with an agent is captured in a meta-graph of how the work came to be, so new sessions surface relevant past work by intent, not only touched files. This shift—from stateless helpers to agents with organizational memory—starts to look like dependency management for knowledge itself, giving AI code context that spans both repos and time.
Coordination Overhead, CVEs, and Why Cross-Repo Context Defines Agent Value
The deeper story behind both Block’s migration and Polygraph’s launch is coordination overhead. Nx’s founders point out that a solo developer using agentic tools might see about a 4.3x speedup, while developers in large organizations constrained by coordination only see around 1.3x. In other words, faster local coding does not matter if every cross-repo change still requires meetings, manual alignments, and spreadsheet-driven dependency audits. Monorepos fix part of this by allowing atomic updates across services in a single commit and resolving shared dependencies directly from source. Synthetic monorepos extend the same benefit to teams that keep multiple repositories, letting agents act across boundaries without human glue. The practical impact goes beyond developer happiness. For enterprises, Polygraph can respond to security issues by finding everything that depends on a vulnerable library version and opening as many pull requests as needed, cutting CVE response times “from days or weeks to something that could be done in hours”. That is what real AI autonomy looks like: agents not only write code, they orchestrate multi-repository change and shipping.
Conclusion: Monorepo Thinking as the Foundation for Agentic Engineering
Both Block’s monorepo consolidation and Nx’s Polygraph make the same argument in practice: AI agents cannot be fully useful until we stop treating repositories as sealed silos. Monorepos reduce dependency drift and coordination overhead by centralizing code and dependency management, and they give agents rich AI code context to make confident changes. Synthetic monorepos generalize that idea to polyrepo realities, providing a shared graph and memory so agents can participate in multi-repository coding without being blocked by missing context. The next competitive advantage in software engineering will not come from slightly smarter agents inside a single repo; it will come from teams that restructure their systems—physically or virtually—so agents can read everything they need, write everywhere they should, and remember what the organization has already learned. Monorepo thinking, whether implemented through one large repository or a synthetic monorepo layer, is quietly becoming the infrastructure for agentic engineering.
