Thursday, August 13, 2026

Among today's technology signals, there are several news that caught my attention, especially about the official release of DeepSeek V4 Pro and the rapid development of AI technology. DeepSeek V4 Pro is known as a very powerful AI model, its capabilities are close to Fable 5 and has a big move. Advances in this technology mean that AI can already complete more complex tasks, and its scope of application is also expanding.

This is not just a matter of technological progress, but also a matter of the future of the entire software engineering and development industry. With the development of AI technology, many traditional software development tasks may be replaced, which is a challenge for programmers and developers. At the same time, this also means that developers need to constantly learn and adapt to new technologies to remain competitive in this rapidly changing industry. For example, the OmniRoute project is a good example, which provides an MIT AI gateway that supports multiple AI models and providers so that developers can more easily use AI technology.

In addition, recent discussions about the impact of AI on the middle class of software development are also interesting. Some people believe that AI will replace software development at the middle level, which will lead to changes in the structure of the software development industry. However, I think AI is more of a tool that can help developers improve efficiency and productivity rather than replace them. Developers need to learn how to work with AI and use AI to complete more complex tasks in order to remain competitive in this industry. For example, projects like Grok Bot and Sidekick provide AI-driven development tools to help developers improve efficiency and productivity.

In general, among today's technological signals, the development of AI technology is the most important theme. The release of DeepSeek V4 Pro and the impact of AI on the middle class of software development are important parts of this theme. As developers, we need to pay attention to these changes and need to learn how to use these new technologies to increase our productivity and productivity. At the same time, we also need to think about the challenges and risks posed by these changes and find appropriate ways to deal with them.

The release of DeepSeek V4 Pro is not only an iteration of parameters, but more like a penetration test of existing AI defenses. Combined with the hotly debated technology of stealing inference traces from APIs on HN, we see a new game: improvements in model capabilities are accompanied by exponential increases in defense costs. When Agents are able to complete long-cycle tasks on their own, even working 24 hours a day as Musk advertises, the middle layer of software development is rapidly being exhausted. Projects on GitHub that aim to allow AI to self-evolve and automate routing provide further evidence that code is changing from a human craft to a machine's self-expression. This is a devastating blow to junior developers, but for superindividuals who can control Agent armies, it is a nuclear fusion of productivity.

The collection objects of flow tax are undergoing a fundamental and historic shift. Tencent's capital expenditure surged 82% in the first half of the year, and Alibaba's integrated flash shopping business shrank its front line. These two incidents may seem isolated, but in fact point to the same trend: traffic dividends peaked, and computing infrastructure became a new ticket. When 60% of web traffic turns into AI robots, traditional advertising distribution logic fails, and the Internet becomes a dialogue between machines. Big manufacturers are frantically hoarding GPUs to collect tolls in this new world, but players who cannot pay the high computing power costs can only face the fate of being integrated or eliminated like Taobao flash shopping.

Hardware is trying to find a physical carrier for AI, but the direction is still chaotic. Glory Robot Phone is trying to put a body on AI, while various Agent interface applications are competing for the right to define the operating system. This is not like the end of a generation of products, but more like a painful moulting period from old species to new species. At this stage, AI phones that simply stack up functions are destined to be transitional products. Only Agent hardware that can truly understand and execute users 'intentions can survive. In the coming year, we will see more hardware overlords of the old era collapse due to their inability to carry the weight of AI.

Among today's technological signals, the two most worthy of digging into are: the bubble and cost black hole of the AI model arms race, and the strategic contraction of major manufacturers under the dual squeeze of traffic and capital. Behind these two clues lies the same question-who is paying for this feast?

The release of DeepSeek V4 Pro and Tencent's capital expenditure of 52.7 billion yuan surged 82% in the second quarter, which are two sides of the same coin. DeepSeek's propaganda line is that "capabilities are close to Fable 5," but this benchmarking itself reveals anxiety. Fable 5 has not yet been released, and DeepSeek is already eager to announce a "tie." This is not technical confidence, but a marketing strategy urged by capital. Historically, every round of AI model iteration has been accompanied by an arms race for performance indicators. From ImageNet to GLUE to today's Agent Arena, indicators have become more and more fancy, but the ROI of actual implementation scenarios has become more and more blurred. Of Tencent's 52.7 billion yuan capital expenditure, how much is actually used for AI infrastructure construction, and how much is GPU clusters built to demonstrate "technological content" in the financial report? More importantly, whose profits will these investments eventually be transformed into? At present, the biggest beneficiaries are Nvidia and Intel, not Tencent itself. The commercialization path of the AI model is much longer than imagined. Whether DeepSeek's "big move" can really leverage paying users or is just another free trial gimmick is worth observing.

Another warning sign is Alibaba's strategic contraction of Taobao flash shopping and Cloudflare's report that AI robot traffic exceeds humans. Taobao flash shopping has returned to the Taotian system after burning 90 billion yuan a year. On the surface, it is a "refined operation", but in essence, the capital market's tolerance for burning money and expansion is disappearing. This is in interesting contrast to Cloudflare's data: When AI robot traffic accounts for more than 60%, the business model of web pages is being reshaped. In the past, traffic taxes were collected by Google and Byte. Now, API call fees and data crawling costs are becoming the new "traffic taxes." This means that small and medium-sized websites not only have to face the traditional traffic distribution dilemma, but also bear the additional costs caused by AI training and reasoning. Ali's contraction is not an isolated example, but a microcosm of the entire industry's transformation from a "traffic dividend" to an "efficiency dividend." But the question is, when all major manufacturers start to shrink, who will provide the power for the next round of growth? Historically, every time the Internet bubble burst, a group of "survivors" became new giants through acquisitions and integration. But this time, the high cost of AI may make this integration even more brutal.

Finally, Walter's $12 billion sale of the Lakers may seem to have nothing to do with technology, but is actually a barometer of capital market sentiment. Behind the Lakers 'appreciation of US$2 billion in one year is private equity funds' pursuit of "scarce assets." This is exactly the same as the situation in the AI field-the influx of capital is not because of the clear profit model, but because of fear of missing the next "scarcity outlet." But history tells us that when a bubble bursts, the assets that appear to be the most stable are often the first to suffer. What AI and sports assets have in common is that their value depends largely on narrative rather than cash flow. When the narrative turns, the bubble bursts faster than when it formed.