(An audio, live-discussion version of this post can be found on the Interconnected YouTube channel):
Agora is a platform-as-a-service (PaaS) company for developers to build apps with real-time audio and video capabilities. As a high-growth developer-focused firm, it’s a company I’ve been tracking for quite some time and briefly wrote about before.
Given its connection with Clubhouse (it’s built on Agora), the hottest social media app in Silicon Valley since perhaps Snapchat, Agora’s brand recognition may become more recognized and more people may want to know what it actually does. Thus, it more than deserves its own deep dive. (FYI, you can follow me on Chubhouse: @kevinsxu, for more Interconnected audio events there.)
Agora’s founding story, core technology, API product model, and global market reach are also representative of a larger trend, where platforms built for developers will thrive, because the number of developers worldwide will double in the next five years.
[For my updated thoughts on Agora, in light of new regulatory measures from both the US and China, please see "Revisiting Agora" published on August 1, 2021.]
Agora was founded in 2012. Its founder and CEO, Tony Zhao, has been a veteran technologist of the video and audio networking space for more than 15 years before starting Agora. He was a founding engineer at Webex and worked with Eric Yuan, who founded Zoom. Instead of building a better Webex like Zoom, Zhao wanted to build a better networking infrastructure for audio and video workloads, so more developers can build their own Zoom, Clubhouse, or virtual classrooms, concerts, doctor visits, and even places of worship -- all of which are now real use cases on Agora.
Side note: last year when there was a real possibility that WeChat might be banned in the US, I spent half a day spinning up an Agora instance that can be used for video chatting with family in China. While I didn’t end up needing to use my makeshift video chat tool, learning how to use Agora was easy and straightforward.
Zhao’s insight to enable “living real life online” (analogous to the Metaverse, except in a Metavase your identity is a fictional avatar) came from his time as the CTO of YY.com. He found his way to the CTO position by selling his first company to YY. While most of the US investment community probably have never heard of YY until Muddy Waters’ short report on the company published last November, it is one of the first livestreaming platforms to reach scale in China, if not the world. It is arguably one of the first instances, where people were “living real life online.” To fully appreciate YY’s cultural impact, I recommend watching the documentary “People’s Republic of Desire”. Understanding YY, and livestreaming in general, is also a crystal ball into how people will live post-COVID; livestream e-commerce is already broadcasting from China to Eastern Europe.
YY is not only culturally consequential, but also technologically important, because scaling a system to handle real-time livestream feeds concurrently with millions of users, while also enabling features like gifting or tipping with virtual goods or real money is no easy feat.
That’s what gave Zhao the inspiration (or maybe desire?) to build Agora -- solving the problems he had to solve for YY for developers everywhere.
(There’s not much biographical information on Zhao available in English. For a glimpse into his thinking and personality, I recommend listening to this GGV Capital’s podcast interview with him.)
So how is Agora’s technology an improvement?
Without getting too deep into the technical mumbo-jumbo, Agora provides a so-called Software-Defined Networking (SDN) solution. This approach improves the existing Internet infrastructure, because the Internet operates as a “best effort” network. It means pretty much what it sounds like. The Internet will try its best to deliver some data (a piece of file, text, audio, or video), but it can’t guarantee delivery nor that the quality of the data when delivered would be the same. You can see how this “best effort” approach would make the user experience of an audio or video chat terrible, especially during the 90s when Zhao was working on Webex and when most of the Internet was still dial-up.
This was a hard technical problem to tackle, because the data that flows through the Internet (technical term: data plane) and the network control of that data flow (technical term: control plane) used to be together. SDN is a game changer, because it separates them into two layers. Now engineers can use software to build algorithms to control the Internet’s data flow without touching the hardware that’s delivering all that data (e.g. servers, routers, and switches in cloud data centers). More control, more flexibility.
Agora is one of these SDNs, but built specifically to make sure audio and video data is delivered with good quality, no matter how good or bad your Internet situation is. So whether you are at home connected to strong wifi, jogging outside and connected to a 4G or LTE network, or traveling to a developing country where the network may be 3G or worse, the audio quality of listening to your favorite Clubhouse rooms should be about the same.
That’s really hard to do, but it looks like Agora’s technology is meeting this challenge and need. Clubhouse is gaining users not just in the US, but also Nigeria, Germany, India, and Australia, according to its release notes. Enabling Clubhouse’s global growth is possible, because Agora’s solution is distributed in more than 200 data centers on most continents, as this map from one of its whitepapers shows.
Developer-Focused API Model
We all know the best technology, unfortunately, rarely wins in the market without a good business model. While Agora’s technology looks pretty solid, I think it's developer-focused API model may be what sets it apart from the crowd.
As Zhao somewhat shyly admitted on the GGV podcast episode, in the early days of Agora, there was no revenue or business model. The company just opened its SDN via a layer of APIs for developers to use with 10,000 free minutes every month. If a developer’s usage exceeds that limit, then charge something. The goal is just to encourage developers to try its service. This deceptively nonchalant approach to making money happens to exhibit the two hallmarks of a well-designed, developer-friendly API product: self-service and pay-as-you-use, which I discussed in-depth in Part II of my Global by Nature series.
By supporting a wide range of developer platforms (Android, iOS, macOS, Web, Windows, Linux) and frameworks (Electron, Unity, Flutter, React Native, Cocos Creator), most developers can use Agora to hack on an idea -- say hosting a virtual audio dinner party -- for free.
Developers can self-service by building something functional on Agora just by reading the product’s documentation (like I did to build my video chat clone) without talking to any sales people. And no developer likes to talk to salespeople.
Because Agora is distributed via APIs - a very convenient distribution model of technology -- it can generate revenue on a pay-as-you-use basis. If your crazy virtual dinner party idea takes off like a rocketship (Clubhouse), pay Agora more money to support the growth. If the idea is a dud, no big cost and no big deal. In this model, the customer-vendor relationship is very aligned. Agora only succeeds if the developers succeed. That’s why all of Agora’s services are priced on a “per 1,000 minutes” basis. The prices only vary depending on how heavy and complex the workload is on Agora’s SDN, e.g. live video streaming is more expensive than live audio streaming.
It’s kind of like angel investing, except with APIs.
One last twist to Agora’s model is its Experience Level Agreement or XLA, announced in Q4 of 2020. It's a clever spin to Service Level Agreement or SLA, a concept that’s already familiar with most developers. A typical example of a SLA is when a cloud vendor, say AWS or Azure, guarantees uptime and availability up to a threshold, like 99.99% of the time per year. If that uptime was not delivered due to server outage or other problems, the cloud vendor would have to compensate its customers somehow. Some form of SLA exists between almost every kind of enterprise tech company and its customers.
In the world of delivering audio and video streaming, just keeping the service up and available may not be enough. The data could be technically delivered but the quality would be so bad the receiving end could barely hear or see it. So this XLA goes one step beyond a SLA to guarantee that services are both delivered and done at some level of guaranteed quality.
To be clear, this XLA is still in pilot phase, so who knows if it will end up helping or hurting Agora. The worst thing you can do in infrastructure technology is over-promise and under-deliver. But this move does communicate a high level of confidence Agora has in its own core product.
Clubhouse + Agora: a Classic “Global by Nature” Story
Having started in China, roughly 80% of Agora’s revenue comes from China, according to its most recent Q3 2020 earnings call. The Chinese Internet market also has more demanding workloads for real-time audio and video features, from livestream entertainment to online education. One of Agora’s largest customers in the education space, New Oriental, had to bring all class activities online during the height of the pandemic in China. Agora’s technology supported classrooms that had up to 20,000 students in one session, streaming and interacting to learn. In a way, China is a better market than the US to battle-test real-time infrastructure technology of all kinds.
Since an API is just a common way to get some resources or tools to build things, Agora’s APIs can be used by any developer in the world. That’s what gave birth to Clubhouse. When Clubhouse released its first version in March 2020, Agora was already an established, soon-to-be-public company (its IPO happened in June 2020). Without the flexibility, affordability, and maturity of Agora’s APIs, Clubhouse may not have been created or would at least have taken a much longer time to build.
Agora’s non-China revenue is already growing. Since the model is pay-as-you-use, Clubhouse will no doubt contribute to that growth as its usage skyrockets. It’s a case of Agora’s “angel investing” working out. It’s also a story we will see more in the future -- a battle-tested, well-designed, and well-documented API product from one market giving birth to the next rocket ship in a totally different market.
Agora is only eight years old -- a young company with many operational, strategic, and execution challenges ahead. I’ve personally had a subpar experience with its sales team (still waiting for a demo of its App Builder product). But overall, Agora’s core technology, model, and problem space are all promising. So full disclosure: I do own shares in Agora. Some of you may view this as a conflict of interest, which I fully understand and appreciate, but I also believe strongly in having “skin in the game.”
The word, Agora, is an ancient Greek term that refers to a public “gathering place”. The technology, Agora, of the 21st century has the potential to become a “gathering place” to build many “gathering places.”
Clubhouse is simply one among many.
声网成立于2012年。其创始人兼CEO赵斌在创办声网之前，已经是视频和音频网络领域里有15年经验的资深技术专家。他曾是Webex的创始工程师之一，与创立Zoom的袁征肩并肩共事过。于袁征想做Zoom来代替Webex不同，赵斌想做的是把音视频技术负载打造成一个更好的基础设施平台，让其他开发者做出自己的Zoom、Clubhouse，或者是线上教育、演唱会、健康医疗，甚至宗教礼拜 -- 这些都已经是声网上的真实用例。
小插曲：去年，当微信真的有可能在美国被封的时候，我花了半天的时间用声网 hack 出了一个可以用来和国内家人视频聊天的小东西。虽然最终没需要用，但学习怎么用声网还是很简单直接的。
赵斌对实现 "线上真实生活"（类似于Metaverse，但在Metavase里你的身份是虚构的）的设想来自于他担任YY的CTO时的经历。他把自己第一家公司卖给了YY，当上了CTO。虽然在美国投资界里，在Muddy Waters去年11月发表关于YY的做空报告之前，大多数人可能根本没听说过YY，但它是在中国，乃至全球，最早达到规模的直播平台之一，也可以说是最早的"线上真实生活"的实例。要想充分了解YY对老百姓日常生活和文化的影响程度，我推荐大家看纪录片《People's Republic of Desire》。了解YY，以及更个直播行业，也是一个水晶球，可以预测疫情后人们将如何生活；直播电商已经从中国播放到东欧了。
这也给了赵斌创建声网的灵感 -- 为全球各地的开发者解决他必须在YY解决的问题。
为了让本文更简单易懂些，我不想用太多的技术行话，简单来说，声网提供了一套所谓基于软件定义网络（Software-Defined Networking，SDN）的解决方案。这种方法改善了现有的互联网，因为互联网其实是个 "尽力而为" （“best effort”）的网络。意思其实就是日常理解的"尽力而为"的意思。互联网会尽力传送数据（比如一个文件、一条短信、音频或视频），但不能保证能传到，也不能保证传到后能保持数据质量，真的就是"尽力而为" 。不难想象，这种 "尽力而为" 给音频或视频聊天的用户体验一定很糟，尤其是在90年代的时候，那是赵斌还在做Webex，大部分互联网还是拨号上网（dial-up）。
这是一个很难解决的技术问题，因为互联网里流动的数据 (技术术语：数据平面，data plane) 和控制数据流的网络环节 (技术术语：控制平面，control plane) 过去是黏在一起的。SDN贡献的一大技术突破就是把它分成了两层。分开以后，工程师就可以用软件构建算法来控制互联网的数据流，而不需要触碰传输数据的硬件（如云数据中心里的服务器、路由器和交换机），更好，更灵活的控制
通过广泛支持各种开发者平台（Android、iOS、macOS、Web、Windows、Linux）和框架（Electron、Unity、Flutter、React Native、Cocos Creator），大多数开发者都可以免费使用声网来hack出一个新点子的实例 -- 比如说举办一个虚拟的音频晚宴。
因为声网是通过API模式分发的，也一种非常方便的技术分发模式，可以支持“随用随付”的付费方式。如果你奇葩的虚拟晚宴想法突然变成了艘火箭（也就是Clubhouse）腾飞起来，那你也赚了，声网也赚了。如果这个想法没人用，实验的成本也不大，不伤大雅。在这种模式下，客户与供应商的关系是非常一致的。只有开发者成功了，声网才会成功。所以声网的所有服务都是按 "每1000分钟" 来定价的。价格区分只是根据声网的SDN上的工作负载有多大、多复杂，比如视频直播比音频直播贵些。
声网模式里的最后一个环节就是其2020年第四季度宣布的体验水平协议或XLA（Experience Level Agreement ）。这是在服务水平协议或SLA（Service Level Agreement）基础上的一个巧妙提升。SLA这个概念大多数开发者已经熟悉。SLA的一个典型例子就是，当一个云厂商，比如AWS或Azure，必须保证服务正常运行时间和可用性达到一个阈值，比如每年99.99%的时间。如果由于服务器故障或其他问题导致正常运行时间无法实现，云厂商必须给客户提供补偿。几乎每一家企业技术服务公司与其客户之间都有某种形式的SLA协议。
Clubhouse+声网：一个经典的 "生来全球化" 故事
声网在中国以外的收入已经在增长。由于模式是随用随付，随着Clubhouse的使用量暴增，无疑也会对声网的业务增长做出贡献，也算是声网的 "天使投资" 成功的案例。这也是一个我们未来会看到更多的故事 -- 某个市场里打磨历练、完善成熟、文档完备的API产品支持在一个完全不同的市场中新一艘火箭的诞生。
声网只有八年的历史，还是一家年轻的公司，未来还有很多运营、战略和操作方面的挑战。我个人对其销售团队的体验其实不尽如人意（还在等其App Builder的产品演示）。但总的来说，声网的核心技术、模式和解决的问题都很有潜力。所以郑重披露：我持有些声网的股份。您可能会认为这是一种利益冲突，我完全理解这个观点，但同时我也坚信另一个道理：“skin in the game”。
声网的英文名，Agora这个词，来自于古希腊，指的是一个公共的 "聚集场所"。21世纪里科技版的Agora，也许即将会成为一个能建立更多 "聚集地" 的 "聚集地"。