Table of Contents

• This Week

• Business Use Cases

• Things to Pay Attention to

This Week

Bottom Line Up Front

• Anysphere's AI-powered customer support bot for its software coding assistant, Cursor, went rogue, providing false information to users and triggering a wave of cancellations.

• The bot's "hallucination" was a made-up explanation for a technical issue, highlighting the risks of relying on automation for customer support.

• Experts, including Cassie Kozyrkov and Amr Awadallah, warn that such incidents can undermine confidence in AI-powered customer support and slow the adoption of autonomous AI agents, particularly in highly-regulated industries.

Business Use Cases

Why is this important

• Researchers at MIT and elsewhere have developed a new approach to guide large language models (LLMs) in generating accurate and error-free code in any programming language.

• Their method, which uses a technique called sequential Monte Carlo, allows LLMs to allocate efforts toward outputs that are most likely to be valid and accurate, while discarding unpromising outputs early in the process.

• This approach has been shown to enable small LLMs to outperform larger models in generating accurate outputs, and has potential applications in programming assistants, AI-powered data analysis, and scientific discovery tools, according to researchers including João Loula and Vikash Mansinghka.

Bottom Line Up Front

• AI agents will not solve all business problems, and companies should not invest heavily in them with unrealistic expectations

• Companies should rethink their processes before deploying AI agents to unlock real value and transform organizational change

• Leaders should focus on reshaping opportunities to fundamentally transform functions and gain efficiency, speed, and effectiveness

• They should also focus on inventing new services and customer experiences, and combining predictive AI, generative AI, and agentic AI to unlock these opportunities

• Improving data infrastructure is crucial, and companies should start by identifying areas where data quality is poor and focusing on improving it

Bottom Line Up Front

• The impact of tariffs on the AI industry is expected to evolve over the course of 2025, with Moccia advising businesses to plan for various scenarios, including worst-case, best-case, and likely scenarios, to mitigate the heightened cost pressures, as the new US tariffs have set the stage for increasing IT costs, with vendors likely to develop and roll out new pricing strategies in response.

Things to Pay Attention to

Bottom Line Up Front

• China is rapidly developing its robotics and artificial intelligence industry, with a focus on "embodied AI", to drive economic growth and upgrade its military strength.

• The city of Shenzhen is at the forefront of this movement, with companies like Meituan using drones for food delivery and humanoid robots being developed by companies like Unitree.

• The development of China's robotics industry is closely tied to advances in AI, with companies like DeepSeek making breakthroughs in large language reasoning models, and the government providing funding and support for innovation centers.

Bottom Line Up Front

• Douglas Matty will assume the role of the Pentagon's Chief Digital and AI Officer (CDAO) on Monday.

• Matty previously founded the Army AI Integration Center and has a 30-year career as a commissioned officer in the Army.

• He will lead the CDAO at a time when it is a major player in various DOD-wide initiatives to deliver data analytics, automation, and AI capabilities.

Bottom Line Up Front

• The United Arab Emirates (UAE) plans to use AI to help write new legislation and review existing laws, a first in the world.

• The UAE aims to create a massive database of federal and local laws, and the AI will "regularly suggest updates to our legislation", according to Sheikh Mohammad bin Rashid Al Maktoum.

• The government expects AI to speed up lawmaking by 70 per cent, but researchers warn of potential challenges and pitfalls, including biases and reliability issues, and emphasize the need for human supervision.

Bottom Line Up Front

• Tamay Besiroglu, a famed AI researcher and founder of the non-profit AI research organization Epoch, has launched a startup called Mechanize with the goal of achieving "the full automation of all work" and replacing human workers with AI agents.

• The startup aims to provide data, evaluations, and digital environments to make worker automation possible, with an initial focus on white-collar work, and estimates a total addressable market of around $60 trillion per year.

• Besiroglu argues that full automation could lead to "explosive economic growth" and increased standards of living, but the idea has been met with criticism and concerns about the potential negative impact on human workers and the economy.