Developers often invest a significant amount of time in monotonous, repetitive tasks, and surprisingly, they allocate less time to actual coding.
According to Stack Overflow’s 2022 developer survey, 63% of respondents reported spending over 30 minutes a day searching for answers or solutions to problems. This amounts to 333 to 651 hours of lost time per week for a team of 50 developers. Another survey by Propeller Insights and Rollbar revealed that more than a third of developers dedicate approximately 25% of their time to fixing bugs, with 26% spending up to half their time on bug fixes.
William Zeng and Kevin Lu, both veterans of Roblox, decided to address this trend. They created a platform called Sweep earlier this year, designed to autonomously handle development tasks, including high-level debugging.
Zeng, Sweep’s CEO, stated, “Sweep is like an AI-powered junior developer for software teams.” The platform enables developers to describe their requests in natural language, such as “add debug logs to my data pipeline,” outside of an Integrated Development Environment (IDE). Sweep can then generate the corresponding code and push it to the appropriate codebase using a pull request. It can also address comments made on the pull request, similar to GitHub Copilot but more autonomous.
Sweep recently secured $2 million in funding from Goat Capital, Replit CEO Amjad Masad, Replit VP of AI Michele Catasta, and Exceptional Capital at a post-money valuation of $25 million.
Sweep specializes in writing Python code and utilizes a combination of AI models for code generation, including OpenAI’s GPT-4 and a custom “code search engine” that is not trained on customer data. This code search engine aids in planning and executing “repository-wide” code changes.
Zeng mentioned, “We have one of the best unit test generation abilities available and will run and execute tests in real time.” In the future, Sweep plans to enhance its code generation capabilities with StarCoder, an open-source code-generating model from Hugging Face and ServiceNow.
However, some concerns exist regarding the reliability of AI tools like Sweep in the long run. Research has shown that AI tools can inadvertently introduce security vulnerabilities into applications. Additionally, copyright issues may arise when AI models generate code that is derived from copyrighted or restricted-licensed sources.
Sweep addresses these concerns by prompting users to review and edit the generated code before pushing changes to the master codebase. Zeng acknowledges the challenges related to AI developer tools, emphasizing the importance of reliability and managing large codebases.
Sweep’s services are relatively expensive, priced at $480 per seat per month, in contrast to the business-focused tiers of GitHub Copilot and Amazon CodeWhisperer, which cost around $20 per user per month. Despite the high cost, Sweep claims to have enough capital from clients to sustain the company for years. They plan to use the new funding to expand their team and continue focusing on Python, improving various aspects of tech debt, including unit testing, refactoring, and handling remaining tasks in the code.
