Leading companies in artificial intelligence are betting on specialized applications, moving away from the generic chatbot approach. This week, OpenAI and Anthropic unveiled new versions of their models focused on the healthcare sector, each with distinct strategies but the same goal: to transform healthcare through advanced technology. The interest is undeniable: ChatGPT receives over 230 million health-related queries weekly. In the near future, the imminent launch of the DeepSeek V4 model stands out, promising significant advancements and potentially marking a turning point in global competition, especially in the programming sector, one of the most lucrative. Major Western labs like OpenAI, Anthropic, and Google do not seem willing to cede ground to DeepSeek or its Asian rivals. The sector is also closely watching the arrival of new post-smartphone devices, with relevant leaks already circulating. In research, DeepSeek and MIT CSAIL have agreed to tackle the challenge of managing large volumes of contextual information without sacrificing performance. DeepSeek will implement its solution in the upcoming model, while MIT's proposal could be adopted by other players. It is likely that, in the medium term, we will see a combination of both approaches. After the holiday break, OpenAI and Anthropic announced almost simultaneously health-focused initiatives. More than a race, this movement reflects that the sector has reached a point of no return: healthcare can no longer be sidelined. For years, excessive regulation and operational fragmentation hindered the adoption of AI in healthcare, but now the models are more powerful and, above all, more controllable and auditable. The economic appeal is clear: the search for profitable niches with high barriers to entry. Following success in programming, healthcare is shaping up to be the next big vertical, alongside generalist agents. AI does not aim to "decide," but to coordinate. In an environment where medical information is scattered, AI can efficiently reconstruct context, something unattainable for a human under pressure. OpenAI and Anthropic differ in their entry strategies. OpenAI expands its general assistant into the healthcare sector, adding layers of privacy and access control. Anthropic, on the other hand, opts for institutional integration, embedding Claude into the workflows of hospitals and insurers. While both strategies are likely to converge, the initial audiences are different: end users versus healthcare organizations. In this context, OpenAI has acquired the health app Torch for about $100 million, and Anthropic has made a similar move. Generalist agents are gaining prominence following the success of Manus. Anthropic has launched Claude Cowork, a tool that not only answers questions but also acts on user files, automating tasks like organizing documents or converting receipts into spreadsheets. Available on macOS for professional subscribers, Cowork could mark the transition from assistants that advise to operators that execute real tasks. DeepSeek is finalizing the launch of DeepSeek V4, the successor to V3.2, which already surpassed GPT-5 and Gemini 3 in some benchmarks. The new model is expected to excel in programming, especially in large codebases and error detection. Its launch, scheduled for the upcoming lunar new year, will incorporate the Engram technique, a static memory that allows data querying without recalculation, published as open source. Apple, for its part, has confirmed that the next major update for Siri will be based on Google's Gemini models, in a deal valued at $1 billion annually. This integration will combine on-device processing with Apple's private cloud, maintaining privacy standards. Apple will continue to collaborate with OpenAI for complex queries, while Gemini solidifies its role as a pillar of the Apple Foundation Models. Additionally, the "Sweetpea" project has leaked, an audio device with an advanced chip that could function as a voice assistant and phone remote. Its launch could occur in September, paving the way for a new family of AI-first devices, following the modest results of the first generation of AI gadgets. In terms of revenue, OpenAI and Anthropic account for 85% of the total sector, a figure that rises to 94% when including the top ten companies, such as Cursor, Cognition, ElevenLabs, Midjourney, Suno, xAI, and Perplexity. Many of them already exceed $100 million annually, with Suno (over $200 million), Cognition ($400 million), and Cursor (over $1 billion) standing out. In research, DeepSeek and Peking University have introduced Engram, a memory module that allows models to efficiently query basic facts, improving not only factual knowledge but also reasoning and programming. By freeing the model from mechanical tasks, AI can tackle complex problems and lengthy texts with greater reliability. On the other hand, Recursive Language Models (RLMs) enable working with much longer prompts, fragmenting text and managing inputs up to one hundred times larger than the usual context, with better results and similar costs. AI promises productivity and growth, but, as O'Reilly warns, prosperity will only be real if value is distributed. Without widespread purchasing power, technological abundance can concentrate wealth and weaken the system. The key will be to build a system that equitably distributes the benefits of AI.