By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
a16zThe Top 100 GenAI Products, Ranked and Explained
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 2:53
How the Gen AI 100 Is Built
Steph Smith opens the fourth edition of the Gen AI 100 and asks how the list is tracked. Olivia Moore explains the purpose of finding mainstream AI products and the data-only methodology: top 50 on web by Similarweb visits, top 50 on mobile by Sensor Tower MAU, plus a new mobile revenue ranking.
- •The report tracks consumer AI through startup pitches, viral trends and products reaching mainstream users
- •Many mainstream products may not market themselves as AI
- •Web list uses Similarweb monthly visits for January 2025
- •Mobile list uses Sensor Tower monthly active users
- •A top 50 by mobile revenue was added for the first time
- 2:53 – 5:37
Pivotal Moments in Consumer AI
Olivia Moore walks through the moments that brought AI into consumer awareness, from Midjourney and Character.ai before ChatGPT to the Balenciga pope, BBL Drizzy, Coke's ad and DeepSeek. Anish Acharya explains why DeepSeek's free reasoning model and visible chain of thought surprised the market.
- •Consumer activity lags research by roughly 6 to 12 months
- •Snapchat's My AI reached about 150 million users
- •Balenciaga pope, BBL Drizzy and the Coke Christmas ad were cultural milestones
- •DeepSeek released its reasoning model for free and showed its thought process
- •DeepSeek upended the belief that a horizontal model couldn't reach mass scale quickly
- 5:37 – 8:19
Broken Assumptions and How Early We Are
Steph Smith frames each pivotal moment as a broken assumption and asks where the industry sits on the adoption curve. Olivia Moore says we are still early adopters, and Anish Acharya offers two assumptions that may prove backwards.
- •Assumptions keep being broken within months of each other
- •Olivia Moore says we are moving from infrastructure to application building
- •AI video currently handles only short clips
- •Anish Acharya suggests AI may be better at relationships than expected
- •AI might delegate work to humans rather than the reverse
- 8:19 – 11:53
Newcomers: AI Video and Vibe Coding
Steph Smith notes 17 new web entrants, and Olivia Moore highlights AI video models and vibe coding products as the main newcomer trends. The group discusses Cursor, Bolt and Lovable, and ideas of personal and disposable software.
- •Hailuo, Kling and Sora joined the list, with Veo 2 expected next
- •Cursor and Bolt made the list; Lovable made the Brink List
- •Many users are technical people prototyping
- •Anish Acharya describes DIY and disposable software
- •AI lowers the cost of creation
- 11:53 – 16:15
The Brink List and Products Not Yet Seen
The hosts explain the new Brink List of five web and five mobile near-misses. They then discuss what they expected but didn't see, including style transfer video and voice or agentic products, and the Deep Research primitive.
- •Brink List shows how fast rankings change, with Runway, Otter and Umax edged out
- •Rising products like Krea and Lovable made the Brink List
- •Anish Acharya expected more style transfer for video
- •Voice, screen-aware and Operator-style products may appear on future lists
- •Deep Research is a primitive whose uses are underexplored
- 16:15 – 19:43
Surprises: Saturation, Companions and Consistency
Steph Smith and Anish Acharya share surprises from the list: rapid saturation of vibe coding tools among developers, high-ranking companion apps, and 16 web companies appearing on every list. They also note that most Lovable traffic is from creators, and joke the App Store will be chaos.
- •Gary Tan's tweet said roughly 95% of YC companies use these tools
- •Three companion products in the top 10, two NSFW-oriented
- •16 companies have made every web list
- •Anish Acharya wanted to see more multimodality in companion apps
- •Lovable.dev visits far exceed Lovable.app visits
- 19:43 – 22:47
ChatGPT's Flat Period and Resurgence
Olivia Moore explains that ChatGPT has topped every list but was flat on web for a year, with much traffic from students. It has since resurged to 400 million weekly users, driven by new models and products, and Olivia describes her own growing daily usage.
- •Web visits were flat from February 2023 to February 2024
- •Over 50% of traffic was students at that point
- •Weekly users went from 200 million to 400 million in six months
- •Growth tracks releases like o1, 4o, Advanced Voice Mode, Operator and Canvas
- •Reasoning models increased confidence for serious work
- 22:47 – 25:31
Claude, DeepSeek and Retention
Anish Acharya describes Claude as a beloved non-traditional number two, strong in creative writing and coding. Olivia Moore details DeepSeek's rapid rise and its retention being slightly below ChatGPT's, with caveats about China and VPN-restricted markets.
- •Claude is loved by a smaller group and strong in coding and writing
- •Multiple models can augment each other
- •DeepSeek reached number two on web with only about 10 days of data
- •Day-30 retention was 7% for DeepSeek versus 9% for ChatGPT
- •Many DeepSeek users are in markets where ChatGPT is unavailable
- 25:31 – 27:25
Chinese Video Models and Aggregators Like Krea
Olivia Moore discusses two Chinese video models on the list and why they may be strong. Anish Acharya and Olivia explain why aggregators like Krea matter in a multipolar model world, since users chain tools and pay for many subscriptions.
- •Chinese models are less copyright sensitive and have captioning advantages
- •Sora was somewhat disappointing to some
- •Krea aggregates many models into one canvas
- •Models are each good at specific things
- •Workflows chain image generation, upscaling and video
- 27:25 – 29:23
Opinionated Models and Ideogram
Steph Smith asks whether video models are becoming more opinionated, as image models have. Olivia Moore says both models and apps are, with camera-control prompting, while the group praises Ideogram for text generation and its image-to-text and prompt rewriting features.
- •Video tools now let users prompt camera angles and shot width
- •Different models suit different parts of a video
- •Ideogram excels at text generation
- •Image-to-text can turn a reference image into a prompt
- •Ideogram modifies prompts, helping users learn what they want
- 29:23 – 30:16
Mobile vs. Desktop Use Cases
Steph Smith asks how mobile and desktop differ for video and other AI products. Olivia Moore says mobile wins where users are on the go or the asset is easily captured by a phone.
- •Avatar apps work on mobile because selfies are on the phone
- •Voice-first products suit mobile
- •Language learning and companionship benefit from speaking to a phone
- •Homework helper apps are blowing up on mobile
- 30:16 – 34:20
Mobile Revenue vs. Usage
Olivia Moore explains the new revenue ranking, which overlapped only 40% with the usage list. Categories match, but smaller prosumer apps earn far more per user, while low-cost paid acquisition drives large user bases at lower revenue.
- •Revenue excludes ads and counts in-app purchases and subscriptions
- •Same categories dominate, but different companies
- •Apps like Speak, Otter and Captions earn high revenue per user
- •Some apps make $50-100 million ARR with only a million or two users
- •Plant identification apps appear in revenue but not usage
- 34:20 – 36:24
Key Learnings: Product First
Steph Smith asks what the last few lists teach. Olivia Moore says it all comes back to the product, and that researchers can struggle in consumer because complexity isn't the same as utility.
- •The more you know, the less you know in consumer
- •Without retention, a business is a leaky bucket
- •PhD researchers can struggle building consumer products
- •Focus on the pain point or unique experience
- •Let the data guide choices, even using an older model or a single AI feature
