1009: 54% AI-Generated and Climbing — State of AI

The hosts Wes and Scott introduce the episode focused on discussing the recently released 'State of AI 2026' survey, which explores AI adoption among web developers. They highlight the survey's role in revealing how developers are feeling about AI, what tools they're using, and the impact of AI on their workflows. The episode starts with some personal banter and details about upcoming events like the syntax meetup at JS Nation and React Summit.

A significant theme is the rapid increase in AI coding adoption, with data showing many developers now writing a majority of their code with the help of AI tools. Respondents report using AI for code generation, assistance, review, learning, and debugging, with a notable rise from previous years. Scott and Wes share their personal usage levels, noting that while some are more heavily reliant on AI, others maintain a balance. The survey indicates that around 75% of code is often created or refactored with AI, marking a monumental shift in industry practices.

The podcast delves into which AI models and providers are most used. ChatGPT and OpenAI models dominate, with Claude and Gemini also being popular. Many developers are unclear about what models they’re actually paying for or using—often just opting for 'Copilot' without understanding the backend. There's skepticism about local AI use, with Scott pointing out that most 'local AI' setups are expensive and technically demanding, not the casual, small-scale models many assume them to be. The discussion covers the disparity between perceived and actual use of local versus cloud-based AI, and the popularity of various agents like Copilot and Claude.

The hosts examine the tools developers are using for code and asset generation. TypeScript remains the dominant AI programming language, with Python also widely used. Image generation tools like MidJourney, Nano-Banana, and ChatGPT for images are discussed, noting their roles in creating assets and thumbnails. Video generation, however, is deemed less practical and not widely adopted. Scott criticizes the current state of AI-generated videos, viewing them mostly as novelty rather than useful production tools.

Scott highlights emerging tools from Sentry like Warden and Spotlight that assist in bug tracking, security, dependency analysis, and code review. These tools exemplify how AI can proactively help maintain code quality and safety. There’s also discussion around the use of agents like CodeRabbit and Codex in coding workflows and the need for better code review and security tools to prevent issues like hallucinations and inaccuracies.

The episode discusses the costs associated with AI tools, with some developers spending hundreds or thousands of dollars monthly if fully leveraging high-end models. Many are still on free tiers or low-cost plans, but the trend suggests costs will rise as tools become more integral and sophisticated. Scott mentions the financial risks for AI providers, citing huge expenses like Peter Steinberger’s reported $1.5 million month on OpenAI tokens, emphasizing that AI is becoming a significant, costly part of development workflows.

There’s skepticism about the actual use of local AI by most developers, recognizing that demanding hardware and technical knowledge limit their adoption. Scott notes that many rely on cloud models due to practicality. They discuss the potential future where smaller models handle specific tasks while large models focus on core coding. The trend toward API-free, integrated AI services in enterprise settings is also highlighted.

A major concern discussed is job displacement, with AI’s role in automation worrying many. Other pain points include hallucinations, inaccuracies, security risks, and environmental impacts of large models. The psychological effects on developers—such as stress, frustration, and cognitive overload—are also mentioned. The hosts advocate for better tools that detect bugs, security issues, and improve error handling, emphasizing the importance of responsible AI use.

The episode concludes with reflections on the survey results: a majority see AI as integral to their workflows, with optimism about AI’s staying power despite bubble concerns. The hosts critique the survey for missing details on skills, strategies, and real-world workflows, urging for more comprehensive data. They stress that AI’s influence on software development is profound and ongoing, with a call for more nuanced understanding and better tools to support developers.