The video challenges the common belief that AI tools make developers faster, presenting findings from a study by Meter, a company with ties to OpenAI and Anthropic. Contrary to expectations, the study revealed that AI actually slowed down experienced developers by 19%, rather than speeding them up. This slowdown was attributed to additional tasks such as reviewing AI output, prompting, and waiting for AI responses. The study involved 16 senior open-source maintainers with significant experience, using Cursor Pro with Claude 3.7 as the primary AI tool. The experiment was a randomized control trial involving 246 GitHub issues, where developers either used AI or worked without it, and the time taken was measured from start to merge of pull requests.
Several reasons were identified for why AI slowed developers down. Senior developers often relied on AI for simple tasks they could perform faster manually, and many already knew the solutions due to their deep familiarity with large codebases, which also overwhelmed the AI models. The video’s creator critiques the study’s limitations, including its small sample size, focus on senior developers with established workflows, and use of only one AI tool. It was also noted that AI tends to benefit beginners more, especially when navigating unfamiliar codebases, and that AI tools provide leverage only when used skillfully.
Despite the study’s surprising results, the video encourages developers to embrace AI, highlighting a recent GitHub study that found AI boosts developer ambition, technical fluency, and job satisfaction. AI is predicted to write 90% of code within the next few years, and new skills like agent orchestration and iterative collaboration are becoming essential. The video’s creator promotes his own AI task management tool, Vectal, which he claims can increase productivity significantly. He emphasizes that AI is particularly effective for building new projects or MVPs, offering massive speedups, while improvements in production code are more modest.
To use AI effectively, the video offers seven practical tips: choose the right AI model for the task, leverage AI for new feature development, make small incremental changes on large codebases, thoroughly understand the code and feature before coding, maintain disciplined workflows with frequent commits, engineer clear and specific prompts, and always remain the decision-maker rather than blindly trusting AI. The creator stresses the importance of using AI to upskill oneself by learning faster and encourages developers to maintain their judgment and intuition, especially when making product decisions.
As a bonus, the video introduces custom hooks for Claude Code, scripts that automate and enhance AI interactions by appending specific instructions to prompts, improving reasoning and output quality. Three example hooks are demonstrated, including one that encourages the AI to “think harder” and another that helps interpret logs more effectively. The video concludes by promoting a comprehensive workshop on mastering Claude Code, offering extensive resources and community support for developers serious about integrating AI into their workflows. The creator encourages viewers to join the AI developer community and participate in future studies to further advance the field.
