Andreessen Horowitz's Olivia Moore on top 100 consumer AI apps: agents rise, DeepSeek fades, Sora surprises

Summary

Andreessen Horowitz partner Olivia Moore publishes a top 100 consumer AI apps list every six months, ranking AI-native and majority AI-enhanced products by web traffic and mobile usage via SimilarWeb and Sensor Tower. This is the sixth edition. Moore notes that desktop products like Cursor and Granola are meaningfully undercounted by that methodology, and says future editions will weight payment data more heavily, singling out Yipit's credit card-based panel as more reliable.

Agents on the list

The clearest new development is the emergence of horizontal consumer agents. Genspark and Manus both made this edition. OpenClaw would have ranked at number 30 had the data window extended into February rather than cutting off in January, a strong debut for a product that requires terminal access, limiting its potential user base to roughly 1% of the population.

General-purpose chat

ChatGPT, Claude, and Gemini lead general-purpose chat. DeepSeek has effectively collapsed in the US, though it still ranks highly globally because it dominates in China and Russia. Moore draws a partial parallel to Claude: before either product got mainstream press coverage, DeepSeek had near-zero US brand awareness, and Claude was at roughly 2% recognition in surveys. Big headlines drive trial, but DeepSeek did not retain those users. Whether Claude does remains an open question.

Moore points to Canva and Notion as the strongest examples of growth-stage companies moving fast enough to matter in the incumbent-versus-AI-native debate. Notion has disclosed that 50% of its ARR now comes from AI features, enough for Moore to include it on the list as credibly majority AI. Even so, she expects AI-native companies to win most categories over a 20-year horizon, because cannibalizing existing products is structurally hard. Google's Gemini integration into Gmail and Slides illustrates the incumbent trap: impressive models attached to demo use cases that don't reflect how anyone actually works.

Sora

Sora is the biggest narrative surprise. Downloads peaked at roughly 6 million per month when it topped the US App Store for 20 consecutive days, and have since dropped to around 1.5 million. Daily active users are still climbing, however, and the product now has 3 million global DAUs, which Moore believes is the highest of any mobile video generation product. Sora did not become the social network OpenAI may have envisioned, but it held as a creative tool because the underlying model is genuinely good and a Cameo-style video feature drives repeat engagement. A pending Disney partnership could push it back toward the top of the charts. Moore argues that Sora combined with Disney IP could be what finally unlocks consumer AI for kids and families, a segment held back so far by hallucination and content-safety concerns.

Generative video

Chinese models including Hailuo, Kling, and others from ByteDance are training on copyrighted data and are generally ahead of the pack. Sora and Veo 3 are close behind. Moore's view, which her colleague Justine Moore has written about, is that no single video model will dominate because the requirements for a two-hour film and a ten-second marketing clip are genuinely different. The more interesting opportunity is tools that let creators switch between models depending on the task. Kling and Higgsfield both made this edition of the list on that basis.

The missing social layer

Moore sees a genuine gap in AI-native social products. Skeuomorphic social networks with AI content bolted on don't replicate the emotional stakes of posting real photos of yourself, and they haven't worked. What is actually spreading is narrowcast, private sharing: meme songs made with Suno for a group chat, LoRA-generated images of friends, Sora videos that only land as in-jokes. Moore thinks that use pattern could give ChatGPT group chats, currently seeing weak adoption, a reason to exist. OpenClaw being added to family group chats is the rough prototype, and she believes someone needs to productize it for non-technical users.