Y Combinator — searchable YouTube transcripts & AI Q&A

How AI Is Changing Design at Shopify

AI tools like Codex and Claude Code have shifted the designer's role from crafting layouts in Figma to curating and editing AI-generated output, argues Marvin Schwaibold. He demonstrates this with a weekend project where he used AI to create a realistic grass shader, breaking the task into subtasks and using a custom tool called DialKit to expose and tweak variables. The key takeaway is that designers must understand the complexity of their projects and be able to decompose them into manageable prompts for AI, while maintaining quality control and decision-making.

Not All Open Model Tokens Are the Same | YC Paper Club

Open model tokens vary significantly across inference providers, and the quality gap between the worst and best provider for the same model can exceed the gap between open and closed models, argues Jeff. He recommends monitoring endpoints through distribution tests, log probs, and tool-call success rates to detect silent changes in serving stacks. He also notes that labs can improve consistency by publishing their own quantized models and inference stacks, such as DeepSeek's DeepInfer.

This Is How You Should Actually Make Launch Videos

Farza Majeed argues that founders should treat launching as a continuous process of sharing demos rather than a single polished event. He recommends posting short, unpolished videos daily, focusing on a clear hook and story, and measuring success by conversion rates rather than views. He emphasizes that consistent practice and starting with simple iPhone videos are key to improving.

How To Stand Out In A World Of AI Designs

Alex Danalovich argues that in the age of AI-generated design, standing out requires focusing on conversion and user experience rather than flashy animations or generic AI aesthetics. He advises founders to prioritize clear calls-to-action, avoid scrolljacking, and establish a brand book early to avoid the 'vibecoded' look. He also emphasizes the importance of talking to users and using social proof to build trust, especially for AI products that handle sensitive data.

AI Safety Fact vs. Fiction | YC Paper Club

The speaker argues that much of the current AI safety discourse is hysteria rather than science, and that the field should focus on falsifiable claims and peer review. He highlights that the OpenAI Hugging Face incident was misreported, as the AI was prompted and safety measures were disabled, not acting on its own. He concludes that the most pressing real concern is the misuse of open-weight models by rogue actors, and suggests investing in cyber defense as a practical solution.