Case Study

Factory is building Droids for software engineering with Claude

Factory logo

Factory uses Claude to power autonomous droids that streamline software development for enterprises, saving customers over 500,000 engineering hours and millions of dollars across each step of the software development lifecycle.

  • 550,000 hours of development time saved across customers
  • Reduced development cycle time 20%
  • 3x reduction in code churn

Automating labor-intensive coding tasks

Factory, founded in 2023, is on a mission to bring autonomy to software engineering. The company builds AI-powered systems called "Droids" that automate labor-intensive tasks throughout the software development lifecycle. These Droids can handle everything from code review and documentation to end-to-end coding tasks like refactors, migrations, and feature requests.

"As software is built and companies grow, organizational structure and engineering practices become inconsistent. It leads to bottlenecks and slowdowns," explains Eno Reyes, CTO and Co-Founder at Factory. "There really wasn't any one tool or system that could handle the ambiguity required to navigate the entire software lifecycle, especially one that scaled with the organization."

Factory's solution is to create autonomous systems that can understand and navigate the complexities of large-scale software development. Their Droids use AI to analyze codebases, make decisions, and execute tasks with minimal human intervention.

Building agentic AI systems with Claude

To create these autonomous systems, Factory needed AI models capable of complex reasoning, planning, and decision-making. They found their solution in the Claude 3 model family.

"We're trying to model human cognition," says Reyes. "We need systems that are agentic in nature. They have to have planning, subgoal decomposition, reflection on failure, refinement of tasks as they act. They need memory, both short-term and long-term. They need to be able to retrieve information and use tools. Anthropic and Claude have consistently delivered what we need to get there."

Factory uses Claude 3 Opus for tasks requiring heavy reasoning and complex decision-making. For high-volume tasks that still require good reasoning but need to be very fast, they employ Claude 3 Haiku. This combination allows Factory to create AI systems that can handle the ambiguity and complexity inherent in large-scale software development.

The Claude 3 model family is best-in-class for agentic engineering tasks. Claude 3 Opus outperformed comparable models by a fair margin across a variety of our internally-developed benchmarks for software engineering. In addition, Claude 3 Haiku delivered the most impressive combination of performance and speed in its class.

— Eno Reyes, CTO and Co-Founder at Factory

Streamlining development for enterprise customers

Factory's Droids are already making a significant impact for their enterprise customers. The company typically works with organizations that have hundreds to thousands of engineers.

One of their key products is the Review Droid, which automates the code review process. When deployed, it analyzes pull requests, provides contextualization, leaves inline comments, and helps other reviewers understand the changes. Another product, the Code Droid, can take a ticket from a project management system like Jira or Linear, work on the task, and create a pull request to implement the requested feature or fix.

App screen from the Factory platform
Figure 1: A screenshot of Factory’s product dashboard showing different droids available and their status to the user.

Both of these Droids, which are powered by Claude, dramatically reduce the time engineers spend on routine tasks—enabling engineers to focus on higher-level problem-solving and new opportunities. Factory estimates that across all their customers, they've saved approximately 550,000 hours of development time. On average, each organization saves about 2,300 hours, which translates to a 20% reduction in development cycle time.

Pushing the boundaries of AI in software development

Factory's use of Claude goes beyond simple task automation. They're developing sophisticated systems that can handle complex, multi-step processes in software development.

This focus allows Factory to create AI systems that can handle the immense complexity and variability of real-world software development tasks. Their systems can plan, break down problems into sub-steps, and make decisions at each stage of the process.

App screen on software development lifecycle
Figure 2: An illustration of the current state of today’s Software Development Lifecycle (SDLC) as a complex interconnected set of stages from initial idea to final product deployment

Factory’s commitment to advancing the state of the art in AI-powered software development is evident in their recent technical report, where their Code Droid—which is powered in part by Claude—delivers state of the art performance on software engineering benchmarks.

Looking to the future

Factory continues to grow and refine their AI-powered droids. Their plans include both deepening the skillset for existing droids while also further transforming the software engineering landscape.

With their innovative use of LLMs and their focus on solving real-world problems for enterprise customers, Factory is well-positioned to lead the way in bringing true autonomy to software engineering.