Fireside Chat, Interview
Where does consumer AI stand at the end of 2025?
- Models have reached a quality level sufficient for building real scalable applications, with 2026 predicted as a pivotal year for consumer builders and creators.
- The general LLM assistance market is expected to trend toward a "winner take all" or "winner take most" outcome, with only 9% of consumers currently paying for more than one service among top providers.
- Competitive dynamics may see a bifurcation where infrastructure providers like Google leverage distribution advantages to adopt innovations pioneered by application-layer startups like Krea and Higgsfields.
- ChatGPT is projected to evolve into a general "everything app" through proactive data nudges, while enterprise usage has grown 8 to 9 times year-over-year, potentially driving further consumer adoption.
- Major labs face structural incentives to release safe, incremental product improvements (e.g., group chats, new tools) rather than risky, opinionated standalone interfaces, leading to many new consumer products potentially failing.
- Application layer startups hold an advantage due to an ability to focus on product sensibility and template formats without the compute tension between training and inference faced by big labs.
- Multimodal capabilities are expected to merge into "mega models" handling text, image, and video, with video generation costs dropping to allow for video-first products and "Ghibli moment" viral styles.
- Specific product launches and improvements are forecasted: Meta for 2026 on SAM3 series models, Grok releasing interactive video game content by year-end, and Perplexity launching prosumer interfaces with agentic capabilities.
- Google's distribution may help it compete despite the "Kleenex of AI" status of ChatGPT, while Anthropic may need to simplify accessibility to expand beyond hyper-technical verticals.
- Consumer AI products are expected to demonstrate over 100% revenue retention via usage-based pricing, separating exceptional products from good ones.
- Social apps using AI-generated content face a diminished "status game" due to a lack of real human representation, shifting the focus toward humor and prompting skills.
- The integration of search with image models (e.g., "Nano Banana" style interactions) is considered essential for accuracy and will likely become a standard for user engagement.
- Big labs face a compute bottleneck forcing decisions between viral entertainment use cases and coding intelligence, potentially creating a "pincer movement" between the infrastructural layer and the entertainment layer.
- The ChatGPT apps directory is expected to become a major new channel for consumer adoption, while the "supply chain of ideas" will see application companies leading innovation before major providers adopt these templates.