From Single-Repo Bots to Cross-Repository Coordination
AI coding agents are software development automation tools that write, modify, and reason about code while coordinating related tasks, but they have historically been limited to a single repository and lose context when work spans multiple services or libraries. This restriction makes multi-repo changes slow, brittle, and heavily dependent on human coordination, especially in organizations where services sit in dozens or hundreds of separate codebases. Even when an agent edits one service correctly, it often cannot see the downstream consumers in other repositories, so cross-repository coordination still falls back to engineers passing around tickets, documents, or chat threads. The result is that agent sessions tend to stall once they hit a boundary they cannot see across, handing work back to people long before the broader change is complete or safely deployed.
Monorepo Architecture as an On-Ramp for AI Agents
Monorepo architecture gives AI coding agents a single, consistent view of services and shared libraries, which makes cross-service development far more manageable. Block’s migration from about 450 JVM-based repositories to one monorepo shows why: dependency drift, duplicated upgrade work, and runtime incompatibilities had turned routine rollouts into “heroic” efforts. In the consolidated codebase, atomic updates across services in one commit and a shared dependency graph mean agents and humans alike work against the same source of truth. According to Block’s engineering team, the monorepo now supports approximately 8,800 builds per week with p90 CI times of around 10 minutes on a reliably green main branch. For AI coding agents, those same dependency graphs and selective build rules can define which projects to test or refactor, turning large-scale change into a scoped, trackable task rather than a risky multi-repo campaign.
Nx Polygraph’s Synthetic Monorepo: A Virtual Single Source of Truth
Nx Polygraph targets organizations that cannot or will not move to a physical monorepo but still want AI coding agents to act across many repositories. It analyzes internal repositories and the open-source packages they depend on, then builds a dependency graph describing which projects publish which packages and which APIs they provide or consume. Nothing moves; the code stays in its original repos while Polygraph presents a synthetic monorepo view to the agent. Co-founder Victor Savkin argues that the limits of agents are set by how autonomous they can be, and that traditional agents stall because “the agent can only work on the repo it sits in and forgets everything between sessions.” With Polygraph, the agent can modify an API producer in one place and see every consumer across the graph in the same logical session, without manual cross-repository coordination.
Shared, Portable Memory for AI Coding Agents
Beyond its synthetic monorepo abstraction, Nx Polygraph adds a shared memory layer that lets AI coding agents carry context across tasks and sessions. Every conversation an engineer has with an agent is captured and linked into what the team describes as a meta graph of how work was done. When a new session starts, Polygraph can surface past work based on intent—what the session tried to achieve—rather than only by file paths. This portable memory is crucial for cross-service development: a follow-up change in another repository can reuse plans, decisions, and explanations from earlier sessions, even for different engineers. For organizations, this turns agent usage into a collective knowledge base, rather than a set of isolated chats that reset every time. Combined with the synthetic monorepo, it gives AI coding agents both the code context and the organizational memory they lacked.
The Future of Cross-Repository Software Development Automation
Taken together, monorepo consolidation and synthetic monorepo services mark an infrastructure shift that clears a major blocker for AI coding agents: fragmented repositories and lost context. Physical monorepos like Block’s reduce dependency drift and give agents a single, authoritative place to reason about builds, tests, and deployment impact. Synthetic approaches like Nx Polygraph extend similar benefits to teams that remain on polyrepos, offering a unified dependency graph and shared memory without moving code. As these patterns spread, cross-repository coordination stops being a human-only job and becomes a space where agents can plan, execute, and verify multi-service changes. The likely outcome is not a tenfold speedup in typing, but a meaningful reduction in coordination overhead, letting engineers and AI systems share the work of maintaining large, interconnected systems.
