Software developers around the world are no strangers to all of the repetitive and time-consuming tasks that often consume their workdays, leaving them with less time for actual coding. According to Stack Overflow’s 2022 developer survey, a significant number of respondents (63%) reported spending over 30 minutes every day searching for solutions and answers, translating to a substantial time loss, especially in larger developer teams. Another survey conducted by Propeller Insights and Rollbar revealed that more than a third of developers allocate a quarter of their work time to fixing bugs, while 26% spend up to half of their working hours on bug resolution.
There must be a better way! Well, William Zeng and Kevin Lu, both with experience at Roblox, recognized this trend and decided to address it head-on. They co-founded Sweep, a promising platform that has been designed to autonomously handle development tasks like high-level debugging.
Sweep acts more like an AI-powered junior dev for software teams, as their platform allows developers to describe their requests in natural language, such as “add debug logs to my data pipeline,” outside of an integrated development environment (IDE). Then, Sweep generates the corresponding code and pushes it to the relevant codebase through a pull request. It can also address comments from code maintainers or owners, offering a more autonomous solution compared to GitHub Copilot.
Sweep specializes in writing Python code and employs a combination of AI models for code generation, including OpenAI’s GPT-4 and a custom “code search engine.” The code search engine is crucial for planning and executing “repository-wide” code changes, a feature that sets Sweep apart. Sweep can find literal matches or slightly varied portions of code using lexical search and locate loosely related code through vector search techniques.
According to the company, Sweep plans to enhance its code generation capabilities by integrating StarCoder, the open-source code-generating model developed by Hugging Face and ServiceNow.
However, the reliability of AI tools for code generation remains a concern, as they can sometimes introduce security vulnerabilities. The possibility of inadvertently incorporating copyrighted code into software also poses legal risks. Sweep’s approach is to prompt users to review and edit any generated code before integrating changes into the target master codebase.
Sweep’s developer services come at a premium, with a price of $480 per seat per month, compared to the business-focused tiers of competitors like GitHub Copilot and Amazon CodeWhisperer, which cost around $20 per user per month. Despite the higher cost, Sweep has found customers who value their offerings and are willing to pay for its services, ensuring a steady stream of income for the company’s growth. The recent $2 million funding round, led by Goat Capital and featuring notable investors, will primarily be used to expand the team and further improve the platform’s capabilities.
Sweep acknowledges the many challenges in AI developer tools, while emphasizing their commitment to ensuring reliability and managing large codebases. As they continue to focus on Python, the company aims to address various aspects of crushing tech debt, from unit testing to refactoring and handling incomplete tasks in the code.