Cognition Labs Launches AI Software Engineer Devin

The AI market is undoubtedly skyrocketing, and rising with it is coding assistantship. Coding assistants like GitHub Copilot are valuable tools used to support software developers and engineers, mostly utilized to reduce the time spent on code writing and streamline coding processes in general through automation.

According to The Business Dive, the AI coding assistant market is estimated to reach $2.3B in value by 2028, indicating a 41.2% compound annual growth rate since 2021. Based on research conducted by Microsoft, developers are able to carry out their tasks 55.8% faster when they use an AI coding assistant.

Taking things much further, applied AI startup Cognition Labs, founded by CEO Scott Wu, CTO Steven Hao, and CPO Walden Yan, creates Devin, the first AI software engineer able to perform complex engineering tasks independently with a natural language prompt.

Reasoning Training

The boom of existing generative AI tools like OpenAI’s ChatGPT or Google’s Gemini has, for quite some time, been a slight reason for worry for software engineers and other careers in the software development field. However, most of these AI tools are mainly utilized to assist with coding without much training in basic reasoning.

Devin’s training, on the other hand, was focused on reasoning to take the leap and not stop at just code. Though Cognition has not revealed whether Devin was trained using the startup’s proprietary model or a third-party model, it has stated that Devin’s advanced properties come from long-term reasoning and planning.

Devin’s expertise in reasoning is said to unlock new possibilities in a wide range of disciplines, created to become a tireless and skilled teammate for software engineering teams, building alongside humans. What differentiates Devin from other models is the fact that it undoubtedly could perform independently and autonomously if one instructs it to do so.

Instead of merely providing coding suggestions, performing auto-completion, or identifying and rectifying bugs, which engineers have utilized AI models to reduce the time and effort taken in development, Devin can take a single natural language command and turn it into a functioning website or software.

Human Assistance

In courtesy of Devin’s reasoning training along with its own developer tools such as the shell, code editor, and web browser within a sandboxed compute environment, Devin is essentially its own AI software engineer, able to solve complex engineering tasks like any human engineer could.

Users can simply type a prompt into Devin’s chatbot as they would with LLMs, and the model will develop a detailed plan followed by execution. Devin will write its own code, fix any bugs, and conduct testing, live-reporting its progress to its users in real-time.

This revelation, which Cognition has only just announced and previewed less than a week ago, caused a shift in the software engineering community, with many questioning whether Devin would end up not only out-performing them but essentially doing all their work for them, rendering their jobs futile.

However, Cognition assures that Devin’s live reporting feature was built so users, as human observers, can interfere whenever needed by commanding the AI to fix an issue or make a change. The model was made not to replace human engineers but to assist them. By having an AI model by their side whom they could delegate a variety of complex tasks to perform well, Cognition hopes that engineering teams can focus more on creative tasks that would require irreplaceable human intelligence.

Cognition tested Devin on the SWE-bench, a benchmark platform that lets models resolve issues found on open-source projects, essentially to test its effectiveness and accuracy levels. Devin correctly solved 13.86% of the issues given to it end-to-end, outperforming other existing models such as GPT4 with 1.74% resolve and Anthropic’s Claude 2 with 4.80% resolve.

The startup claims that Devin has been used to complete real freelancing jobs on platforms like Upwork, and according to demos of Devin’s use cases, it seems the model really can build and deploy apps from start to finish. With Devin’s capabilities, there is no telling what the future holds for both the AI and the software engineering community. In staying optimistic, we can only hope AI will maintain its position and stay alongside humans, not up and above.

Cognition has raised $21M in a Series A funding round led by Peter Thiel’s Founders Fund, which the startup plans to use to further improve its model by ramping up capacity. At the moment, Devin is only available for early access to select users, and those interested in testing out the model can contact Cognition’s team through its website.

Source: Cognition Labs

Meme & AI-Generated Picture

Job Posting

• Cognition - General Application - San Francisco, CA (In-Office/Hybrid)

• Cognition - Machine Learning Researcher - San Francisco, CA (In-Office/Hybrid)

• Cognition - Software Engineer - San Francisco, CA (In-Office/Hybrid)

• Capital One - Senior Manager, Product Management - Chicago, IL+ (Remote/Hybrid)

Promote your product/service to Digger Insights’ Community

Advertise with Digger Insights. Digger Insights’ Miners are professionals and business owners with diverse Industry backgrounds who are looking for interesting and helpful tools, products, services, jobs, events, apps, and books. Email us [email protected]

Your feedback would be greatly appreciated. Send it to [email protected]