Author: lethuphuong

📌 AI is upgrading the research industry but simultaneously eroding the foundation of traditional training. As entry-level researchers shift toward the role of “system engineers,” they may lose the chance to hone their intuition and deep thinking. The future is not about AI replacing humans, but about the requirement for humans to become the bridge between data and emotion. Without investing in these skills, the research industry risks losing the essence of understanding humanity.

📌 AI is not killing the programming profession but is restructuring the entire industry. Demand is still rising but is concentrated in senior positions with salary increases of ~15%, while entry-level positions are nearly “frozen.” The work is shifting from writing code to managing AI and product thinking. The biggest challenge is not immediate job loss, but the risk of a shortage of the next generation of engineers in the future.

📌 AI agents only create value when managed like real personnel, with clear roles, KPIs, and supervision. Although 94% of tasks can be automated, AI currently handles only about 1/3, and less than 10% of businesses deploy them effectively. This shows the main challenge lies in management, not technology. Businesses need to shift from a “deploying AI” mindset to “operating an AI workforce” to fully leverage its potential.

📌 The Chinese government has banned Manus co-founders Xiao Hong and Ji Yichao from leaving the country after meetings with economic regulators. This move is related to the investigation into Meta’s $2 billion acquisition of the AI startup Manus in 2025. The incident highlights growing US-China tech tensions, particularly in the field of generative AI. With a $2 billion deal and a startup hitting $100 million in revenue in months, Manus has become a strategic focal point. The travel restrictions on the founders indicate that China is tightening its grip on technology. The outcome of the investigation could significantly impact…

📌 Research indicates that LLMs are not the neutral strategic advisors many leaders believe them to be. In thousands of tests and over 15,000 simulations with GPT-5, AI consistently recommended “trendy” strategies such as differentiation, cooperation, and long-term thinking, regardless of the business context. This tendency to provide identical strategic advice, favoring modern management jargon over specific situational analysis, is termed “trendslop” by researchers. This stems from the internet data and modern management culture that AI learns from. Therefore, LLMs should be used to generate ideas and analyze options, but the final strategic decisions must remain the responsibility of humans.

📌 According to AI expert Ayesha Khanna, the most important skill in the AI era is “learning how to learn,” which means knowing how to handle ambiguity, experiment, fail, and combine knowledge from many different fields. While AI can provide information rapidly, humans need to focus on soft skills such as creativity, critical thinking, and complex problem-solving. She argues that universities should shift from an exam-based model to one centered on discussion, debate, and real-world problem-solving. In enterprises, AI should be used to expand employee capabilities, helping them perform tasks that were previously impossible.

📌 The AI boom is creating a new demand for frontline deployment engineers—those who directly implement AI into businesses. In 2025, job postings for this position increased more than 10 times compared to 2024, while mentions in corporate financial reports rose from 8 to about 50. Due to the requirement for both deep technical skills and business operational understanding, only about 10% of engineers are willing to do this work. This scarcity makes engineers who directly deploy AI the key factor in determining whether AI can function in the real world.

📌 Delays in leadership decision-making primarily stem from messy inputs, vague options, and a lack of recorded rationale. AI can address this by standardizing summaries, clarifying choices, synthesizing data, and logging decision logic. When AI acts as support infrastructure instead of the decision-maker, leaders can focus on strategic judgment, helping the organization make faster and more accurate decisions in complex business environments.

📌 Silicon Valley is witnessing a major shift in work as programmers move from writing code to coordinating teams of AI agents. Many set up AI agents to work through the night or while at parties, checking progress like caring for digital “Tamagotchis.” Tools like the new Claude can complete tasks equivalent to 12 hours of human labor, causing many engineers to manage 4–5 bots at once. While accelerating software development, this trend also raises concerns about AI acting out of control and traditional coding skills becoming less necessary.

📌 This deal shows that Chinese tech companies are finding ways to access advanced AI computing power through international cloud infrastructure instead of directly purchasing restricted hardware. With a cluster of 36,000 Blackwell GPUs worth about $2.5 billion in Malaysia and the potential to expand by another 7,000 GPUs in Indonesia, ByteDance is building large-scale AI capabilities while still complying with US export regulations. This reflects the trend of “Global AI Compute Outsourcing,” where cloud access becomes a strategic factor in the AI competition.