What It Takes for AI to Become a "Real Colleague" — A Survey Paper from Tencent and Others

A survey paper from Tencent and several Chinese universities maps out a path for evolving AI from chatbots into "digital colleagues." The paper's argument is clear: for an AI system to be seen as a trustworthy colleague, it needs more than the ability to generate answers — it must be able to carry tasks through to completion within a persistent working environment.

The key lies in combining a persistent workspace with reusable skills. Today's AI assistants reset with every conversation; if they could instead retain their working environment and context while accumulating and reusing skills, their practical value inside enterprises could change dramatically.

The observation offers a reality check on the hype surrounding today's AI agents, revealing a gap that still needs closing between the expectation that they can "act autonomously" and the reality of actually "seeing tasks through to the end."

Coinbase Adopts Chinese AI Models — Western Labs Face a Pricing Stress Test

Reports indicate that Coinbase, the cryptocurrency exchange, is moving to adopt Chinese-made AI models — a development that throws the harsh price competition facing major Western AI labs into sharp relief.

Chinese AI models are increasingly offered at high performance and low cost, which lines up directly with corporate cost-cutting demands. For companies beyond Coinbase that pick models on practical grounds, price-to-performance is becoming a decision criterion they cannot avoid.

This shift signals that the AI model market is moving from a competition of "brands" to one of "price and performance." For Western labs to preserve their value, they will likely be tested not just on raw performance but on cost competitiveness as well.

Austria Pushes to Bring Anthropic into the EU — Countering US Access Restrictions

It has emerged that Austria is lobbying the EU to bring Anthropic within an EU member state. The backdrop is a US move to restrict foreigners' access to advanced AI models.

Austria's aim appears to be securing European access to the major AI players and preserving geopolitical influence. With US restrictions tightening, this is further evidence that European countries are accelerating their search for independent AI strategies.

It is one more example of how the AI development race keeps expanding from a purely corporate contest into political territory where nation-states are involved.

An Individual Develops an Alzheimer's Treatment at Home Using AI

A case is drawing attention in which an individual developer used AI to create a candidate Alzheimer's disease treatment at home. Without specialized laboratory equipment, the individual reportedly leveraged AI tools to carry out part of the drug discovery process.

While the case shows AI democratizing drug discovery, it also raises safety and ethical concerns. Some argue that democratizing biotechnology will require phased deregulation similar to the legalization of cannabis. As AI lowers the barrier of specialized expertise, how far individual-level drug development should go — and how it should be governed — will become an important debate in the years ahead.

75% of Dutch Responses to EU Tobacco Regulations Were AI-Generated

An investigation has revealed that roughly 75% of the responses submitted from the Netherlands to the EU's public consultation on tobacco regulations were generated by AI tools. Tobacco giant Philip Morris is said to have been involved.

The case shows that large-scale, AI-driven manipulation of public opinion is already a reality. The risk of democratic decision-making processes being polluted by AI-generated opinions is a new challenge for regulators, and detecting AI-generated content while safeguarding the integrity of public consultations has become a matter of urgency.

Cate Blanchett Unveils a Free Tool to Protect Your Identity from AI

Actress Cate Blanchett has unveiled a free tool for protecting personal identity from AI. It aims to shield everyday users from AI-driven deepfakes and impersonation.

AI protection initiatives led by celebrities carry significant influence in terms of visibility. As tech companies and governments struggle to keep up, a movement is spreading in which individuals take self-protection into their own hands.

AgentWatch: Runtime Budget Management to Keep AI Agents from Running Amok

AgentWatch was announced on Hacker News. Designed to prevent runaway loops and unexpected spending by AI agents, it enforces budgets before requests ever reach the model.

In running AI agents, "how much execution do you allow?" is a critical practical question. Alongside prompt and context design, this tool deserves attention as an approach that provides runtime-level safeguards.

News from the Japanese-Speaking Community

Zenn and Qiita saw a wave of practical AI development articles. Notable posts include:

"Before I Knew It, the Walls Between Industries Had Disappeared — The Art of Article Jacking"**: a substantial read collecting the insight that AI is producing common patterns across three different domains — drug discovery, IT, and EdTech

A build log of a multi-AI conference system that runs Gemini, Claude, and Codex simultaneously on Discord**

A practical guide to GEO (Generative Engine Optimization)**: an explanation of techniques for making your content more likely to be cited by AI search engines such as ChatGPT and Perplexity

Why AI Self-Review Is Dangerous, and Implementing Role Separation**: an article warning of the dangers of having AI both write and review code, and arguing for the importance of human-in-the-loop