2012: I Left Baidu Before The Wolf Culture
· 24 min read · Views --

2012: I Left Baidu Before The Wolf Culture

Author: Alex Xiang

Essays

I am an old-school programmer. After leaving big tech I have spent my time in startups, and I am still doing hands-on AI-related development. This column will slowly trace the changes in IT from the 1990s to today, with some personal memories mixed in. I publish fuller updates on the WeChat account “字与码”; if these recollections and observations are useful to you, feel free to follow there.

In 2012, I left Baidu and joined Weibo.

In my memory, it was a subtle moment. Baidu was still powerful and technically deep, but the company’s atmosphere had begun to change. People later often talked about “wolf culture”; when I left, I could already feel that direction beginning to appear.

I did not leave because of one specific incident. More accurately, I had reached a fork: stay inside a search giant, or move to a place closer to mobile internet and social media.

I chose the latter.

A Thread To Connect Before Leaving

When I wrote about 2010 earlier, I mentioned that I owed my entry into Baidu to Liancheng, the leader who hired me.

Before I left Baidu, he went to Silicon Valley. Later he became an early employee at Databricks. Databricks was founded in 2013 by the team behind Apache Spark, and would become an important company in data and AI infrastructure.

That thread is interesting to me.

Liancheng later researched Spark at Intel’s IoT research institute and made many open-source contributions. He also translated two Erlang-related books, and I was lucky enough to read early drafts. Looking back now, the line running through him, distributed systems, data processing, functional concurrency, and Silicon Valley startups, connects naturally with my later work on Weibo’s big data platform and with how I look at AI data infrastructure today.

I moved from Linux system software into Baidu, and then from Baidu into Weibo’s big data platform. The person who brought me into Baidu went to Silicon Valley and joined a company closely tied to Spark, Lakehouse, and data platforms. Years later, it feels like we were all moving along the line of “data is becoming more important,” just by different paths.

At the time, I of course could not see the whole picture. I only felt that the people around me, the direction of the industry, and the choices of companies were all moving somewhere new.

Weibo Stood In The Mobile Internet Wind

Weibo was hot in 2012.

It was no longer just a product for posting short messages. It had become a major information field on the Chinese internet. Breaking events, celebrities, media distribution, public discussion, fan relationships, and brand marketing all happened there.

More importantly, mobile internet was pushing it into a larger scene.

In the PC era, using Weibo meant sitting in front of a computer. In the mobile era, it meant opening, posting, and watching at any time. That difference was huge. Information no longer waited for the user to return to the desktop; it moved with people.

For backend systems, that meant higher-frequency requests, more complex push, more fragmented access, more real-time feedback, and a stronger need for data capability.

Social data platform

After joining Weibo, I worked on the big data platform. That direction felt natural then: if a social media platform cannot understand its own data, it is hard for it to understand users, content, relationship chains, propagation paths, and commercialization opportunities.

The Stack Became More Mixed

During my Weibo years, I used a more mixed set of languages than at Baidu.

Big data platform APIs used Java. Data processing used Python. Redis secondary development used C. Mobile APIs used PHP. Later I gradually used more Go. Today many people like unified stacks, but real business sites are often not like that.

When a company grows quickly, different teams, different historical stages, and different business pressures leave different technical choices behind. You cannot always start elegantly from scratch. You can only solve problems inside real systems.

That is a kind of training for engineers.

You need to understand the boundaries of different languages, understand why the system became what it is, know which places can be refactored, which places can only be reinforced for now, and which places must stay compatible. Much of the time, engineering ability is not about writing a beautiful new architecture; it is about finding a viable improvement path through a complicated old system.

The Feeling Of An IPO Sprint

Weibo later went public in 2014. When I joined in 2012, the company was already clearly sprinting toward that stage.

A pre-IPO company has a special atmosphere: the business must grow, the data must look good, the system must be stable, and the story must make sense. Engineering teams have to support daily business while constantly filling in platform capabilities.

That same year, Toutiao launched, and Xiaoju Technology, the predecessor of Didi, was founded. One represented algorithmic information distribution; the other represented mobile transportation and location services. They were not traditional portals, search engines, or social networks, but new species growing out of mobile internet. When I walked into Weibo in 2012, I happened to be standing at the handoff between the previous generation of social platforms and the next generation of mobile platforms.

That pressure was different from Baidu’s.

Baidu was a mature big company, strong in infrastructure and systems. Weibo felt more like a platform standing in the wind, with product form, user behavior, media attributes, and commercialization all changing quickly. For engineers, Weibo’s problems were closer to the business front line, and more mixed.

Later, when I worked on Weibo Toutiao, I would see the competition between Weibo and Toutiao more directly. That is a later story. In 2012, it was only the beginning: I moved from a search giant into a social media platform, and from a relatively stable engineering system into a business site closer to the mobile internet wave.

IT Events Of 2012

  • Mobile internet kept growing quickly. Social media, mobile clients, push, and always-with-you content consumption entered a deeper stage. Users no longer accessed the internet only in front of computers; they stayed connected in fragmented time.
  • Weibo entered its pre-IPO sprint. Weibo was in the stage before going public, with social media, big data platforms, and mobile entry points intertwined. Hot topics, relationship chains, celebrities, media, and mobile usage together formed the platform story.
  • ByteDance was founded and launched Toutiao. Toutiao pushed algorithmic recommendation into the main line of content distribution. Its direction was not traditional portal, search, or social networking, but an information-flow machine driven by user behavior feedback.
  • Xiaoju Technology was founded and launched Didi Dache. Mobile internet began reshaping urban transportation. Location, payment, dispatch, supply-demand matching, and mobile experience combined to platformize offline services.
  • AlexNet won ImageNet. Deep learning achieved a landmark breakthrough in computer vision. It brought neural networks back into industrial attention and laid a key foundation for later AI waves.
  • Big data platforms became core capability. Mobile and social platforms generated large amounts of logs, relationships, and behavior data. Whether a company could collect, clean, compute, and understand that data began to directly affect product iteration and commercialization.

References

More in this column

View all posts in "Thirty Years in IT and Me"