Editor’s note: Robert Brennan is speaking at Odsc you have west this October 28th-30th. Check out his talk, Automating Massive Refactors with Parallel Agentsthere!

Today’s coding agents excel at small, atomic tasks—the kind that fit neatly into a single commit. But what about the massive refactoring projects that have traditionally consumed months of developer time? Tasks like migrating from Java to Golang, porting from VueJS to React, or breaking apart monolithic architectures into modular systems?

The Evolution of AI-Powered Development

We’ve witnessed remarkable progression: from ChatGPT’s context-unaware snippets in 2022, to GitHub Copilot’s editor integration in 2023, to autonomous coding agents like OpenHands in 2024 that could run code, debug errors, and iterate until completion.

Now in 2025, we’re entering the era of parallel agent orchestration—deploying fleets of AI agents to tackle large-scale refactoring projects that would otherwise take months to complete manually.

The Challenge of Scale

The key insight is that massive refactors aren’t actually single tasks—they’re collections of smaller, automatable subtasks. The challenge lies in intelligent decomposition and coordination. Simply telling an agent “port this codebase to Go” results in incomplete solutions. Instead, we need to break these projects into bite-sized pieces that individual agents can handle autonomously.

Strategic Approaches to Task Decomposition

Piece by Piece: Move through your codebase systematically—directory by directory, file by file. This works well for incremental changes like adding type annotations to Python applications.

Dependency Tree Navigation: Start with leaf nodes in your dependency tree and work downward. Port utility functions first, then classes that depend on them, eventually reaching your main application logic.

Scaffold and Iterate: Build temporary scaffolding that allows partial migrations. Create shims or compatibility layers that let different parts of your system migrate independently while maintaining functionality.

Orchestrating Agent Workflows

The magic happens in the coordination. Each agent works on its own branch, contributing individual pull requests to a central refactoring branch. This allows for human review and validation at each step while maintaining parallel progress across multiple components.

Context sharing becomes crucial as agents encounter roadblocks or discover better approaches. Advanced implementations can include message-passing systems where agents broadcast learnings to their peers, creating a collaborative intelligence network.

The Human-in-the-Loop Advantage

While these systems can compress months-long projects into days, they’re not fully autonomous. Success requires a skilled engineer who understands the full project context—someone with business intuition, user empathy, and deep knowledge of the existing codebase.

The sweet spot is 80-90% automation with strategic human oversight. Engineers can manage fleets of 3-5 agents simultaneously, with the potential to scale to dozens as tooling improves.

Real-World Impact

Organizations report dramatic improvements: 10x faster execution, higher accuracy through automated validation, and consistent code quality across massive codebases. This transforms resource-intensive, error-prone manual processes into manageable, systematic operations.

The Future of Large-Scale Development

We’re moving toward a collaborative model where developers focus on architectural decisions while agents handle implementation details. This partnership amplifies human creativity while eliminating tedious, repetitive work.

Ready to revolutionize your approach to large-scale refactoring? Attend Robert Brennan’s talk live at ODSC AI to see these concepts demonstrated and get your questions answered by experts pushing the boundaries of AI-assisted software development.

About the Author

Robert Brennan has been writing software for 15 years, with a focus on natural language processing and developer tools. He is currently the CEO of All Hands AI, the company behind OpenHands (formerly OpenDevin), a fully autonomous AI developer. Previously he was VP of Product Development at Fairwinds, ran a startup called Datafire, and worked as a Senior Software Engineer at Google.