Author | Xu Xiaofei Editor | Mai Guangwei Last year, Google made a remarkable comeback with its large model capabilities, stealing the spotlight and even leading investors to predict a "market cap of $10 trillion." Strangely, however, Google has remained relatively unknown in the critical field of AI coding.
Powerful as Google is, why has it stumbled in AI coding?
Even more surprisingly, Google is not alone. Looking across the internet giants in both China and the United States, their performance in AI coding has been mediocre. Among the well-known products in the programmer community—such as Cursor, Claude Code, Codex, Zhipu AI, and MiniMax—very few come from major internet companies.
It seems everyone has fallen into a strange cycle: major internet companies cannot do AI coding well.
AI coding is not only the most mature and fastest-growing commercial scenario for large models to date, but also a critical entry point in the era of AI software.
Why are the internet giants, who are most eager to seize "entry points," collectively absent this time?
01 Why Has Google Fallen into the "Big Company Curse"?
"The strategic significance of AI coding goes far beyond the tool itself," multiple senior architects from major companies emphasized to Leifeng.com.
It can be said that whoever controls the developer entry point controls the production methods of next-generation software (including Agents).
In fact, over the past four decades, the iteration of each generation of software platforms has essentially been a game of "whoever defines the development rules harvests the ecological value." AI coding is becoming the "definer" of development rules in the AI era.
Moreover, one of the aforementioned architects told Leifeng.com that AI coding platforms can also capture full-chain data in the production of AI software like Agents: which APIs are frequently called, which business logic is most commonly implemented, what the developers' decision-making paths are, and so on.
"Controlling the entry point for Agent production is equivalent to controlling the entry point for AI To B."
Given the importance of AI coding, why is Google struggling with it?
Fei Lianghong, former Chief Architect at Amazon Web Services, believes that Google is not "bad at" AI coding, but rather has consistently failed to translate its model advantages into a product with a clear developer mindshare.
"Google's problem is essentially not technical, but rather product fragmentation, lack of entry points, and a failure to concentrate organizational efforts," he candidly stated.
In fact, Google has numerous AI coding products, such as Gemini Code Assist, Jules, Gemini CLI, Firebase Studio, and AI Studio; one could even say they are emerging one after another.
However, these products are developed by different departments, featuring different brand names, different entry points, and even different pricing models, leading to internal conflicts and frequent iterations.
"This is a typical result of 'feudal fragmentation' under the internal horse-racing mechanism of major companies," Fei Lianghong noted. For development tools, stability and continuity are vital. Google's chaotic layout leaves users unsure of what to choose, thereby weakening brand recognition.
Secondly, Google also faces the entry dilemma of "fighting on away turf."
"GitHub Copilot was the first to make a name in AI coding because it had a natural foundation. It was only later, when it failed to push forward on the model side, that it was overtaken by competitors," Zhou Lei, a senior architect at a major digital technology company, told Leifeng.com.
In the development toolchain sector, veterans like Microsoft have the VS Code editor and GitHub, which are used daily by over 70% of programmers globally. Newcomers like Cursor have also secured the core IDE entry point by heavily modifying the underlying architecture of VS Code.
In contrast, Google is more like "duckweed without roots" in the programming field, lacking equivalent code repositories and IDE distribution channels.
To avoid boosting Microsoft, Google once vigorously promoted a web-based Cloud IDE, attempting to change programmers' muscle memory.
However, programmers have strong stickiness to local debugging environments. A web-based IDE means network latency and complex permission configurations, so this attempt by Google fell flat.
When it had to turn back and write plugins for Microsoft's VS Code, it was reduced to a "second-class citizen" fighting on away turf, constrained by the competitor's underlying framework.
Furthermore, hidden behind these issues is the even more fatal commercial orientation issue where "where you sit determines where you stand."
"For a long time, Gemini Code Assist has been more like an upselling tool for Google Cloud rather than a product independently refined around the developer experience. After all, the cloud team's goal in pushing AI coding was to prevent customers from defecting to competitors due to service outages," revealed Sarah, a former Senior Key Account Product Manager at Google Cloud.
This is just the tip of the iceberg.
Tang Xiliu, CEO of Yuanxu Intelligence and a former senior engineer at Google, told Leifeng.com that for major internet companies like Google with mature core businesses, it is often hard to escape the curse of the "Innovator's Dilemma."
Driven by the cash cows of search and advertising, Google's resources and attention naturally lean toward its core business. It is difficult for a new species like AI coding to gain high priority and corresponding resource allocation internally.
Moreover, giants like Google are accustomed to being large and comprehensive, pursuing scale and stability. In contrast, a new field like AI coding requires small teams, extremely fast iterations, and relentless focus on developers' real pain points. This is precisely the pace of startups, not the pace of major companies.
Additionally, major companies usually carry heavier burdens, such as concerns over legacy code, compliance, and brand risks, making them dare not be as aggressive as Cursor.
"It's not that major companies lack the capability, but rather that the success factors for innovation—such as being small, fast, specialized, and willing to take risks—are contrary to the DNA of major companies. It's not that Google isn't strong enough; rather, its advantages have turned into disadvantages in the context of AI coding," Tang Xiliu summarized.
02 Why Have Domestic Giants Also Collectively "Gone Silent"?
Turning the focus back to China, the phenomenon that "major internet companies cannot do AI coding well" is even more pronounced.
Feedback from some programmers and secondary market investors shows that the top-reputed AI coding products in China are basically from independent model providers like Zhipu AI and MiniMax.
Many programmers at domestic major companies report that they mostly use overseas AI coding tools for daily development, and many domestic tools also rely on calling overseas large model APIs to achieve code capabilities.
Why is this the case?
An unnamed head of engineering efficiency at a major company candidly told Leifeng.com that there have been certain strategic missteps.
"Domestic large models started relatively late. In a catch-up state, there is pressure to achieve high benchmark scores, and the focus is entirely on reporting upwards. By the time they realize it, it is already a bit late. Moreover, many major internet companies prioritize the integration of AI with their core businesses, even subordinating themselves to serve the core business, leaving AI coding without priority."
Furthermore, similar to Google, domestic internet giants lack the "pressure of self-use."
Each business group and department has its own KPIs. It is difficult for them to abandon better overseas tools just to refine the group's self-developed AI coding tools and models.
"Everyone's KPIs are different. Even if we sacrifice R&D efficiency to polish internal AI tools, whose performance will it count towards?" a senior IT engineer at a major company told Leifeng.com. Their team directly provides each member with a reimbursement quota of several thousand dollars per month, allowing them to use any AI tool they want. When money is not an issue, everyone basically uses the best in the world.
When major companies do not use their own products, the products cannot gain an "efficiency-improving flywheel," making it even harder for client companies to adopt them.
A seemingly counterintuitive phenomenon is that small and medium-sized enterprises (SMEs), often criticized for "lacking the habit of paying," are unusually willing to spend money when it comes to R&D tools, which are core productivity drivers.
"The differences between tools are still quite significant. With world-class tools, I can still get off work at a normal time, but with some domestic AI coding tools, I have to work overtime until 2 AM."
"We use whatever the company buys. The company bought GPT and MiniMax. If GPT runs out of quota, I will use the Codex tool paired with the MiniMax model."
"When AI coding tools first came out, we tried many of them, but later we were too lazy to bother switching. These domestic tools came out relatively late, and their initial evaluation performance was not very good either. If we can use the top-tier models, we will definitely prioritize using the top-tier ones."
The above is feedback from multiple engineers at various SMEs to Leifeng.com.
"Actually, many domestic teams probably only realized the importance of AI coding around July and August 2025, and several major companies started developing IDE tools," revealed Zhou Lei, the aforementioned architect at a major company.
Who would have thought that after the release of Claude Opus 4.6, CLI tools caught fire.
"Before this, everyone was still competing in IDEs, still believing that AI coding meant AI assisting humans in coding. But after Opus 4.6 came out, everyone realized that AI coding is truly about AI doing the coding entirely on its own, and the entire definition changed."
"The biggest problem with making IDEs is that IDEs are meant for humans, not for AI. Therefore, the vendors competing in IDEs basically haven't achieved any particularly good results," Zhou Lei added.
"Domestic companies have never been strong in developer tools, and major companies are no exception," Zhang Song, a consultant at an IT consulting firm in Shanghai, told Leifeng.com. Besides facing "lack of roots and insufficient ecosystem" like Google, there is also a kind of "inertia of thinking" in business.
Whether it is Google or these domestic internet giants, they are more accustomed to the traditional internet mindset of traffic distribution and enterprise sales—either relying on resource bombardment or having sales teams sign bundled contracts with the leaders of public cloud clients. However, AI coding is a typical PLG (Product-Led Growth) track.
"Programmers are a group that extremely worships technology, resists commercial brainwashing, and holds core discourse power over productivity tools. The tactics that internet giants used to rely on, such as 'branding, subsidies, and bundled sales,' are hard to influence them," Zhang Song explained.
So, do internet giants still have a chance?
"Don't expect major companies to build good AI coding tools using the 'army group' approach," Tang Xiliu candidly stated. Space must be given to small and specialized teams. The best domestic AI coding tools will most likely come from relatively independent teams that can iterate quickly.
"Don't just focus on the arms race of model parameters. The winning hand in AI coding is increasingly in engineering and product experience, which is precisely what Chinese teams are good at, but also where they tend to fail to go deep due to impetuousness."
03 Conclusion
Google's dilemma in AI coding is not a technical one, but an organizational one.
The company possesses powerful AI research capabilities, yet it has consistently failed to stably translate this capability into a product that programmers open every day.
The reasons are multiple. Product fragmentation, lack of entry points, dispersed organizational efforts, deviated commercialization orientation, the Innovator's Dilemma... these reasons combined constitute a greater warning: AI capability does not equal AI product.
And domestic companies may be walking a similar path.
Fortunately, the competitive landscape of AI coding has not yet completely solidified. But this window of opportunity may not last more than two years.