Developer velocity, the rate at which an organization deploys code, often faces delays due to essential but time-consuming processes such as code review, documentation writing, and testing. These inefficiencies can further prolong these processes. According to a source, developers lose approximately 17.3 hours per week dealing with technical debt and poor-quality, nonfunctional code.
During a San Francisco Hackathon, Machine learning PhD Matan Grinberg and Eno Reyes, formerly a data scientist at Hugging Face and Microsoft, recognized the need for a better solution. They developed a platform capable of autonomously handling simple coding tasks, a platform that they later realized had commercial potential. Following the hackathon, they expanded the platform to manage a broader range of software development tasks and established a company called Factory to commercialize their creation.
Grinberg stated, “Factory’s mission is to introduce autonomy into software engineering. More specifically, Factory aids large engineering organizations in automating parts of their software development lifecycle through AI-powered autonomous systems.”
Factory’s systems, which Grinberg humorously refers to as “Droids” (a term reminiscent of Lucasfilm), are designed to handle various repetitive and time-consuming software engineering tasks. For instance, Factory includes Droids for code review, code refactoring, code restructuring, and even generating new code based on prompts, similar to GitHub Copilot.
Grinberg elaborates on the Droids’ functions: “The review Droid provides insightful code reviews and offers context to human reviewers for every codebase change. The documentation Droid generates and continuously updates documentation as needed. The test Droid writes tests and maintains test coverage as new code is merged. The knowledge Droid resides in your communication platform (e.g., Slack) and answers more in-depth questions about the engineering system. The project Droid helps plan and design requirements based on customer support tickets and feature requests.”
All of Factory’s Droids rely on the “Droid core,” a foundational engine that processes a company’s engineering system data to construct a knowledge base. An algorithm extracts insights from this knowledge base to resolve various engineering problems. A third component, the Reflection Engine, serves as a filter for third-party AI models that Factory employs, allowing the company to add its own security measures and best practices on top of those models.
The enterprise value of Factory lies in its software suite, which empowers engineering organizations to produce higher-quality products more quickly while reducing the burden of tedious tasks like code review, documentation, and testing, thereby boosting engineering team morale. Furthermore, the autonomous nature of the Droids minimizes the need for extensive user training and onboarding.
If Factory can consistently and reliably automate these development tasks, it would provide a substantial return on investment. A 2019 survey by Tidelift and The New Stack revealed that developers spend 35% of their time managing code, including testing and addressing security issues, leaving less than one-third of their time for actual coding.
However, a critical question remains: can Factory’s system deliver on its promises? Even the most advanced AI models are not immune to significant errors. Generative coding tools can introduce insecure code, as a Stanford study suggested that software engineers using code-generating AI were more likely to create security vulnerabilities in their applications.
Grinberg acknowledged that Factory lacked the resources to train all its models in-house and is subject to limitations of third-party vendors. Nevertheless, he emphasized that Factory continues to deliver value by leveraging third-party AI muscle while building its own AI systems and reasoning architectures. As an early startup, competing with established players in terms of model training, monetary resources, hardware access, and data access is challenging.
Factory’s long-term strategy is to develop more of its AI models and establish an “end-to-end” engineering AI system. It intends to differentiate these models by soliciting engineering training data from early customers.
While this ambitious vision may appear optimistic, the AI startup landscape is growing increasingly competitive. To their credit, Factory has already engaged with approximately 15 companies, ranging from seed-stage startups to public companies, which have utilized Factory’s platform to conduct thousands of code reviews and generate hundreds of thousands of lines of code.
Factory recently secured a $5 million seed round co-led by Sequoia and Lux, with participation from SV Angel, BoxGroup, DataBricks CEO Ali Ghodsi, Hugging Face co-founder Clem Delangue, and others. The additional capital will be invested in expanding Factory’s six-person team and enhancing platform capabilities.
Grinberg emphasized, “The primary challenges in the AI code generation industry are trust and differentiation. Every VP of engineering desires to enhance their organization’s output with AI, but many AI tools are unreliable, and large, complex organizations are often hesitant to trust this new, futuristic technology. Factory is shaping a world where software engineering becomes an accessible, scalable resource.”
