Developer velocity (the speed at which an organization ships code) is often influenced by necessary but time-consuming processes such as code reviews, documentation, and testing. Inefficiencies can make these processes even longer. according to According to one source, developers waste 17.3 hours a week on technical debt and bad code, or code that doesn’t work.
Machine learning Ph.D. Matan Greenberg and Eno Reyes, previously a data scientist at Hugging Face and Microsoft, thought there had to be a better way.
During a hackathon in San Francisco, Greenberg and Reyes built a platform that could autonomously solve simple coding problems. This is a platform they later came to believe had commercial potential. After the hackathon, the two expanded the platform to handle more software development tasks and founded a company. factoryto monetize what they built.
“Factory’s mission is to bring autonomy to software engineering,” Grinberg told TechCrunch in an email interview. “More specifically, Factory helps large engineering organizations automate parts of their software development lifecycle through AI-powered autonomous systems.”
Factory systems – Greenberg calls them “droids” in Lucasfilm terminology there may be a problem — Built to juggle a variety of repetitive, mundane, but typically time-consuming software engineering tasks. For example, Factory has “Droids” for reviewing code, refactoring or rebuilding code, and even generating new code from a prompt like GitHub Copilot.
Grinberg explains: “Reviews Droid leaves insightful code reviews, providing human reviewers with context for every change to the codebase. Documentation Droid generates documentation as needed and continuously updates it. Test Droid creates tests and maintains test coverage percentages as new code is merged. Droid knowledge resides in communication platforms (such as Slack) to answer deeper questions about engineering systems. Project Droid also helps you plan and design requirements based on customer support tickets and feature requests.”
Factory’s droids all have what Greenberg calls a “droid core,” an engine that ingests and processes a company’s engineering system data to build a knowledge base, and an engine that extracts insights from the knowledge base to perform various engineering tasks. It is built on algorithms that solve problems. . His third Droid core component, his Reflection Engine, acts as a filter for third-party AI models that Factory utilizes, allowing Factory to implement its own safety measures, security best practices, etc. based on these models. I will make it possible.
“The enterprise perspective here is that this will enable engineering organizations to output better products faster, while also boosting engineering morale by offloading tedious tasks such as code reviews, documentation, and testing. It’s a suite of software that makes it better,” Greenberg said. “Additionally, the autonomous nature of the droid requires little user education or onboarding.”
Now, if Factory can consistently and reliably automate all these development tasks, the platform will certainly pay for itself. According to 2019 investigation According to Tidelift and The New Stack, developers spend 35% of their time managing code, including testing and addressing security issues, and less than a third of their time actually coding.
But the question is, can it be done?
Even today’s best AI models can make fatal mistakes. Generative coding tools can also introduce insecure code, and a Stanford study found that software engineers who use code generation AI are more likely to introduce security vulnerabilities in the apps they develop. It is suggested.
Greenberg was candid about the fact that Factory doesn’t have the capital to train all its models in-house, so it’s at the mercy of third-party limitations. But while relying on third-party vendors for some of its AI capabilities, he argues that the Factory platform still provides value.
“Our approach is to build these AI systems and inference architectures, leverage cutting-edge models, establish relationships with customers, and deliver value now,” Greenberg said. Masu. “For early startups, training is a losing battle. [large] model. There is no financial advantage, no chip access advantage, no data advantage, and (almost certainly) no technological advantage compared to incumbents. ”
Factory long play teeth Greenberg said the company will further train its AI models to build an “end-to-end” engineering AI system and differentiate those models by collecting engineering training data from early customers.
“Over time, you have more capital. Chip shortage The problem is solved and we have direct access (with permission) to a treasure trove of data (i.e., the historical timeline of the entire engineering organization). ” he continued. “We build robust and fully autonomous droids with minimal human intervention, customizing them to our customers’ needs from day one.”
Is that too optimistic? perhaps. Competition in the AI startup market is increasing day by day.
But to Greenberg’s credit, Factory already works with a core group of about 15 companies. Mr. Greenberg declined to name names, but the size of his clients, which have used Factor’s platform to date to perform thousands of code reviews and create hundreds of thousands of lines of code, is from “seed stage.” It covers a wide range of topics, including “public”.
And Factory recently closed a $5 million seed round co-led by Sequoia and Lux with participation from SV Angel, BoxGroup, DataBricks CEO Ali Ghodsi, and Hugging Face co-founder Clem Delangue. Greenberg said the new funding will be used to expand Factory’s six-person team and platform capabilities.
“The main challenges in this AI code generation industry are trust and differentiation,” he said. “Every VP of Engineering wants to use AI to improve their organization’s outcomes. This is hindered by the unreliability of many AI tools and the lack of confidence that this new futuristic sound A large labyrinthine organization that refuses to trust its technology…Factory is building a world where software engineering itself is an accessible, scalable commodity.”
