2024: The Year the GPT Wrapper Myth Proved Wrong

The hosts recall the view that OpenAI’s GPT Store and foundation models would leave little room for startups. They point to Perplexity, Glean, Harvey, and Photoroom as examples of applications built outside OpenAI, and say open source models such as Llama widened the field. With more model choices, product quality, sales, user feedback, and low churn matter more to application companies.

The hosts say enterprise pilots that once drew skepticism have turned into paid deployments, including systems handling thousands of support tickets a day. They attribute better reliability to agent techniques and the infrastructure around models. They also report that YC’s summer and fall 2024 batches grew about 10% per week in aggregate, roughly tripling during the program.

The hosts describe voice AI as useful across many markets, with workflows varying between airlines, banks, and software companies. They expect both shared infrastructure and specialized applications, comparing the relationship to Stripe and the many products that use payments. They also see growing interest in robotics, while noting hardware remains costly and difficult.

The hosts say Cursor and other AI coding tools became common among YC founders, while Replit agents and Claude Artifacts let nontechnical users build prototypes. They expect hiring interviews to test how candidates use these tools and assess their output. They cite an Amazon language migration that reportedly changed hundreds of thousands of code lines in weeks instead of taking an engineering project six months or longer.

The hosts discuss concerns about AI regulation, including California’s SB 1047 and federal policy, and say the regulatory outlook was still uncertain. They describe Apple Vision Pro adoption as limited by hardware constraints and a shortage of apps, while Meta Ray-Ban glasses offered a useful audio and voice experience. They close on YC’s return to in-person Demo Day and renewed optimism about San Francisco.