A few days ago, the world’s top VC, a16z, released a rather interesting AI product list. This list did not include products based on hard metrics such as website traffic or revenue. Instead, 24 practitioners in the AI industry were invited. Each person selected their three favorite AI products of the year and gave their evaluations.
Among these 24 “big names”, the representatives of investors included Marc Andreessen, co-founder of a16z, and Garry Tan, the current CEO of Y Combinator. The representatives of practitioners included Aravind Srinivas, CEO of Perplexity, Danny Trinh, product design director of Meta, and Scott Belsky, chief strategy officer of Adobe. In addition, the quite active partners of a16z on social media, such as Olivia Moore and Justin Moore, also participated.
Products that have been chosen more than 2 times include ElevenLabs (audio generation, 5 times), Suno (music generation, 4 times), and Claude (ChatBot, 3 times). These three products all have global web AI product traffic levels within the Top 50.
Surprisingly, ChatGPT, as the ‘backbone’ of AI products, was only mentioned once by a16z partner Justin Moore. The author believes that the reason for this situation is partly due to ‘aesthetic fatigue’ towards ChatGPT, and on the other hand, new developments in ChatGPT such as AI search functionality and o1 inference models in 2024 are not so impressive. In addition to these overall observations, the author categorized all 45 products mentioned by ‘big shots’, from which we can also see some trends: ‘AI Tools’ mainly refer to comprehensive tools launched for a specific scenario in work or life, such as writing tools, meeting summary tools, presentation tools, and so on. 1. AI tools are the most numerous type of product, with a total of 10, accounting for 22%. Efficiency-enhancing products for various work/life scenarios are also a continuous hot topic, as improving the capabilities of large models is important, but without application scenarios, they are merely castles in the air. 2. The second-ranking in terms of product quantity is AI video, with a total of 8, accounting for 18%. As a modality that started relatively late, it has seen explosive growth this year, with multiple ‘image/text to video’ products appearing on the market, and product types such as Talking Video and long to short tools have produced a number of star products. 3. Although Perplexity was picked the most times, the AI search track it is in is quite bleak, with no other products on the list except Perplexity. 4. Other product types with a share of more than 10% are AI ChatBot/Agent and AI Image, which have also been popular AI tracks. However, we can also see that niche industries/scenarios such as AI programming, AI psychology/medicine/parenting, have also produced many products and are beginning to be noticed by practitioners. 5. In addition to AI software, AI ‘hardware’ such as smart glasses co-launched by Meta and Ray-Ban, AI companion product Curio, and intelligent driving ride-hailing Waymo, have also entered the field of vision of practitioners.For this topic selection, we will still focus on four main types of AI products: AI tools, AI ChatBot/Agent, AI images, and AI videos. We will also focus on products that have the potential to be selected but are not so well-known. Following the “subjective perspective” of practitioners, we will observe these potential products. Newer product types such as AI hardware and AI in psychology/medicine/parenting will be the subject of separate topic selections for observation in the future, so we won’t go into details here.
AI Tools: The application scenarios are becoming increasingly rich, and most of the mentioned ones are in the initial stage. The AI tool category mainly refers to comprehensive tools launched for a specific scenario in work or life, such as writing tools, meeting minutes tools, AI presentation tools, etc. Among the products in the AI tool category, the only relatively well-known product mentioned is Gamma. From the traffic data, Gamma’s traffic reached its peak in November but declined in December. We once observed Gamma in the topic selection of AI + PPT. Interested readers can refer to “AI +?, Attracting Traffic and Making Money”. Gamma’s traffic data from May to December 2024 | Picture source: AI Pure Another AI tool that was discussed frequently this year is NoteBookLM launched by Google. In May 2024, Google officially launched the AI note-taking product NoteBookLM powered by the Gemini large model. It can complete tasks such as answering questions, generating summaries, and extracting keywords based on information such as documents, videos, and audios uploaded by users. In particular, the function of generating podcasts (supporting two-person conversations) based on the uploaded information has been widely praised. NoteBookLM’s traffic data from June to December | Picture source: AI Pure NoteBookLM experienced a rapid growth in traffic from September to October. By December, its traffic had reached the tens of millions mark. Adobe Acrobat AI Assistant was only mentioned once by Adobe’s Chief Strategy Officer, which may seem like a token mention. As a product with a market share of over 50% in the global PDF reader market, adding AI capabilities is only natural.Now, let’s shift our focus to products recommended by industry leaders but less well-known…
Granola Use Case: Online Meetings Main Features: Granola is an AI meeting minutes tool. When users conduct online meetings, it can transcribe the meeting verbatim and also generate brief meeting summaries based on the content. It also saves commonly used meeting templates for future use. Ben’s Bites founder Ben Tossell’s evaluation: ‘The best note-taking app for meetings. I’ve tried almost all other tools, and I like it because it doesn’t cause awkwardness during the meeting, and the product design is very exquisite. Among the tools I’ve used, it offers the best transcription and summary features.’ Lindy Use Case: Email Management, Online Meetings, Online Marketing, Online Customer Service, Content Creation, etc. Main Features: Lindy is an AI efficiency tool for work scenarios, offering a variety of functions. Lindy’s main distinction is providing users with a process editor to customize their workflows and have AI complete related tasks, simplifying workload. Lindy currently offers over 100 workflow templates, primarily for scenarios such as meetings, marketing, and content creation. The Rundown AI founder and CEO Rowan Cheung’s evaluation: ‘This tool can help sales teams accomplish the most tedious and time-consuming tasks, allowing the team to focus on truly important matters, such as planning influential events and communicating with customers.’ Flow Use Case: Text Entry Main Features: Flow is an AI dictation tool currently only supporting MacOS. Users can use Flow for voice input to transcribe into text in any scenario requiring text input. Flow currently supports 100 languages, and its official website indicates that this tool can triple the speed of text input. Ben’s Bites founder Ben Tossell’s evaluation: ‘I use this transcription app every day. Just set a key and start speaking; the app will transcribe the voice and format it into text (including generating summary points, quoted content, or making minor corrections when you repeat yourself), and automatically paste it into the app.’ ElevenLabs Design Director Ammaar Reshi’s evaluation: ‘This voice transcription feature is excellent; I just need to speak to my Mac.’It automatically removes all filler words such as ‘uh’ or ‘um’, and when it detects that I am listing items, it automatically formats them into bullet points.
Cora Usage Scenario: Email Management Main Features: According to the official website, Cora provides users with concise email summaries twice a day, allowing users to quickly find important emails that require a response and categorize related emails in the form of a storyline. In addition to this, Cora also offers smart reply and spam filtering functions. Currently, Cora is gradually opening up registrations, and users need to join the Wishlist to gain access. Ben’s Bites founder Ben Tossell’s evaluation: This is a tool developed by Every that has not yet been released. It handles a large volume of emails and sends me reports daily, processing over 90% of my emails. It can set rules according to different types of emails, automatically archive emails, and draft replies when needed. Spiral Usage Scenario: Writing Main Features: Spiral is an AI tool for repetitive writing. Users need to provide a piece of content, and the AI can rewrite it into content with specific style, structure, and tone. For example, users can provide a podcast outline and let Spiral write the text summary or related social media promotion content. Spiral’s target audience includes content creators and marketers who need to process a large amount of text content. It is worth mentioning that Spiral, like the previously mentioned Cora, both come from Every, a startup based in New York. Newsletter author and podcast host Lenny Rachitsky’s evaluation: We use this tool in our podcast workflow to generate possible titles, thumbnail ideas, and key content summaries. It is trained to match our style and reference our most successful episodes. Particle.News Usage Scenario: News Browsing Main Features: Particle.news is an AI news reading application developed by a former Twitter engineer. Particle integrates reports from various news media according to major daily news events and provides summaries and related sources, striving to offer users a comprehensive perspective. In addition, the App version of Particle also offers more functions such as AI Q&A, media political, and text-to-speech.At present, Particle has reached cooperation with news media organizations such as Reuters, AFP, and Fortune Magazine. It has also completed two rounds of financing totaling $15.3 million. Currently, Particle has been launched on the web and on the iOS and iPadOS platforms in the US market. However, the download volume in the past 30 days is only 2,257. Adobe Chief Strategy Officer Scott Belsky commented: This is the news app I use every day.
It combines perspectives from various sources and then pushes the most objective summary content in a way that increasingly suits my preferences. Cubby Usage Scenarios: Research, Information Organization Main Functions: Users can store and manage various content types in Cubby, such as articles, PDFs, YouTube videos, podcasts, tweets, presentations, and e-books. The product provides functions such as marking comments, AI summaries/Q&A, and intelligent recommendation of relevant content. In addition, Cubby also offers browser plugins and team collaboration-related functions. a16z partner Zach Cohen commented: Having a centralized workspace designed specifically for research is a magical experience. It can switch between different modes, search across multiple websites, and also supports embedded AI functions. Moreover, the “Generate Tweet Thread” function is simply tailored for venture capitalists. AI ChatBot/Agent: The Main Battlefield of Big Companies, with the Only Exception Being the “Live Version” C.AI The products of AI ChatBot/Agent are not too surprising. ChatGPT, Claude from Anthropic, Meta AI, and Grok from Musk’s X.ai are basically products deployed by major companies. Top 3 ChatBot Traffic List in December | Picture Source: AI Product List However, Google Gemini, which ranks second in the global ChatBot traffic list, has not been mentioned by anyone. Although Gemini itself has not been mentioned, NotebookLM, which is equipped with Gemini, has made it onto the list (introduced above). In addition to NoteBookLM, Gemini has also been integrated into Google’s products and services such as Workspace, YouTube, Gmail, and advertising. Although the evaluations of these “integrations” are mixed, Google’s strategy of developing AI applications relying on its own ecosystem is relatively clear.Delphi Official Homepage | Image Source: Delphi
Another ChatBot/Agent product on the list is Delphi, which defines itself as a “digital cloning tool”. Users can upload information related to themselves or others (including social media content, articles, speeches, etc.). Delphi will create an “Agent” containing elements such as thoughts and voices based on this content. It can be used to chat or reply to emails instead of real people. This model is a bit similar to creating characters on c.ai, except that Delphi is aimed at real people. AI Images: Awarded Products Demonstrate Their Specialties Note: Currently, some image generation products include video generation functions. However, if a product mainly features image generation, it will be classified into the image generation category. Products such as Coling and Runway, which are mainly focused on video and also have image generation functions, will be introduced in the following text. Among AI image products, the “big brother” Midjourney, the well-received ideogram, and Photoroom, the absolute leader in the product image production track, were selected mainly for their fame and reputation. Judging from the traffic data, in the past six months, the traffic of these three products has been basically stable, without obvious growth or cliff-like decline. The other three products were selected for their “special features”. Playground Creation Interface We observed Playground in the topic “Exactly One Year Later: What Happened to the Image Generation Products on the First a16z List?” launched during this year’s National Day holiday. It has changed from the previous traditional text-to-image mode to an “image editing” mode. Users can input a prompt to let the AI modify the picture, reducing the “gacha” problem caused by the instability of the model and making it more in line with user needs than direct generation. However, we also noticed that it has not been designed and optimized for specific scenarios, and its traffic has dropped from around 4 million to around 2 million. Krea Creation Interface Krea’s most distinctive feature is “real-time image generation”. As shown in the figure above, after users input a prompt, they can also add graphics of different colors and shapes to the left picture to more stably control the color, position, relative relationship, etc. of each object in the generated picture, adding more certainty to image generation. In addition to this, Krea also has other functions such as text-to-image, object-background integration, and text-to-video.New Feature of Krea’s 2D to 3D Conversion | Image Source: Krea
On January 17, Krea launched a new function of converting 2D images to 3D images in the real-time image generation section. However, perhaps because it is still in the testing phase, the author is currently unable to use this new feature. Remix Landing Page (left 1), Creation Page (left 2), Discovery Page (left 3), Chat Page (left 4) Different from other products, Remix’s main platform is not the web version but the app version. In addition to basic functions such as text-to-image/video generation and image stylization, Remix has relatively complete community functions. These include a content sharing page that can be scrolled up and down, a complete work ranking and search function, as well as chat/group functions. Many products focus on building an image generation community on the web version, but on the app version, Remix is probably the only one that uses the TikTok-style swiping card information flow. AI Video: Text-to-Video Still Losing Money, Talking Video and Long-to-Short Editing Making Money Among the selected AI video products, there are currently three mainstream types of landed products: text-to-video, talking video, and long video to short video conversion. Text-to-video is the hottest AI track this year, especially for overseas companies and Chinese startups, which are extremely active. When Kling was launched in July, it sparked large-scale discussions both at home and abroad. The website’s monthly traffic increased from 2 million to 15 million within a month. Even today, although Kling’s traffic has decreased somewhat, products such as Minmax’s HailuoAI, Shengshu’s Vidu, and Chinese startup Haiper have entered the top 10 of video generation products and achieved good results. However, although AI video generation may be one of the paths to AGI, as a product, it is still in a very early stage. Although Kuaishou gained a huge amount of traffic through Kling, its ROI is not very positive. In November, Kuaishou stated that Kling’s monthly turnover reached tens of millions of RMB (1.36 million US dollars), but people familiar with the matter revealed that from June to September alone, its overseas advertising expenses reached 330,000 US dollars, and Kling’s model training costs and daily inference costs (presumably data as of November) even reached hundreds of millions of US dollars. Compared with the costs, the income is negligible.Other WenSheng video products are largely in the same boat, still in the phase of making losses for publicity. When it comes to profitability, long-to-short video editing and Talking Video are more pragmatic choices. Taking Captions, which is oriented towards the Talking Video scenario, as an example, its global monthly revenue reached $1 million in December 2024. Coupled with the positive web-end data, the income is quite considerable.
HeyGen, which has better traffic, has even better commercial achievements. In June this year, founder Xu Zhuo revealed that HeyGen’s Annual Recurring Revenue (ARR) has reached $35 million, with more than 40,000 paying users. With its good commercial performance, HeyGen has also been favored by capital, with a total funding amount of $65 million and a valuation of $500 million. Similarly, the long video to short video product OpusClip, founded by Chinese founders, has also achieved an ARR of tens of millions of dollars. Looking forward to 2025, the biggest variable in the video race may be the ban on TikTok. How this ban will affect the AI video industry, we will continue to pay attention.
