Leading AI companies are rushing out automation features in a race to build AI agents that can automatically handle specific tasks without humans, as they seek to expand their share. After OpenClaw, an open-source AI agent tool that carries out various tasks on a user’s behalf, drew strong interest, tech companies’ AI agent strategies have also shifted toward greater automation.

Anthropic updated Claude Code and Claude Cowork, adding functions that can carry out tasks directly on a user’s computer. The update allows Claude to take more proactive actions such as opening files, launching a browser and using development tools.

Anthropic adds user PC control feature to Claude Cowork and Claude Code

Anthropic also introduced an "auto mode" in Claude Code. Auto mode lets the AI autonomously decide permission levels on a user’s behalf. With auto mode, Claude Code can carry out tasks independently for the user. It is a useful feature, but it also carries risks because it could lead to unintended actions such as deleting files, sending sensitive data externally, or running malicious code or hidden commands. Auto mode is offered as a research preview for team plan users. Anthropic plans to offer it soon to enterprise and API users as well.

OpenAI also added plugin support to its coding AI tool Codex, expanding it beyond a simple coding tool into an enterprise collaboration platform. OpenAI’s strategy is to use plugins to make it easier for companies to implement AI-based work automation.

OpenAI introduces plugins to Codex, expanding beyond coding into a collaboration tool

Competition between OpenAI and Anthropic over next-generation models is also drawing attention. Sam Altman (샘 알트먼), OpenAI’s CEO, recently announced an organisational reshuffle and said it has also completed pre-training for a next-generation model codenamed "Spud." He said, "In a few weeks a very powerful model will come out, and the whole team believes this model can meaningfully accelerate the overall economy."

OpenAI, which scrapped video-generation AI Sora, completes training for next-generation model Spud

Anthropic is also developing a new AI model with performance greatly improved from its existing models and has begun testing with some customers, according to reports. The new model, known as Claude Mythos, is reported to have scored far higher than its previous top model, Claude Opus 4.6, in evaluations including software coding, academic reasoning and cybersecurity. The leak of information about Anthropic’s new AI model again led to a drop in U.S. cybersecurity-related shares. Altman said it has also completed pre-training for a next-generation model codenamed "Spud." He said, "In a few weeks a very powerful model will come out, and the whole team believes this model can meaningfully accelerate the overall economy."

Reports say Anthropic is testing new AI model Claude Mythos with different-level performance; security firms’ shares fall on news of new model

OpenAI’s restructuring has also emerged as an issue. As part of a broader reset of its business strategy, OpenAI abruptly shut down its video-generation AI app Sora and also halted development of an adult mode for ChatGPT. OpenAI also scaled back ChatGPT’s instant payment feature and revised its strategy to instead strengthen product discovery functions.

OpenAI shuts down short-form video app Sora as it cuts costs aimed at an IPO; OpenAI puts ChatGPT adult mode on indefinite hold; OpenAI revises ChatGPT e-commerce strategy toward an information hub over instant payment

The article also summarised moves by companies in and outside South Korea around AI.

Anthropic’s AI service Claude is rapidly increasing paid subscribers in the U.S. consumer market. An analysis says Super Bowl advertising and a dispute with the Department of Defense instead served as a catalyst for boosting consumer awareness. Meta effectively acquired AI agent startup Dreamer by hiring its talent.

Anthropic’s Claude sees surge in paid sign-ups, more than doubling this year; Meta steps up AI agent startup acquisitions, also hires Dreamer team

SAP will acquire master data management company Reltio. It is part of efforts to optimise corporate data for AI and strengthen SAP Business Data Cloud.

SAP to acquire master data management firm Reltio to optimise enterprise data for AI

Apple is pushing a plan to shift voice assistant Siri to an open platform. WhatsApp, a messenger platform operated by Meta, announced a major update that includes AI-based auto-reply functions. ByteDance will add a new AI video generation model, Dreamina Seedance 2.0, to CapCut. Google will expand its AI-based translation function "Live Translate" to iOS and significantly increase the number of supported countries. Google also unveiled its AI music generation model "Lyria 3 Pro."

Apple pushes to turn Siri into an AI platform and allow integration with external AI apps; WhatsApp announces major update featuring AI-based auto-replies; ByteDance to add AI video generation model Dreamina Seedance 2.0 to CapCut; Google expands real-time earphone translation to iOS and adds many more supported countries; Google releases AI music generation model Lyria 3 Pro

Enterprise AI company Cohere launched an open-source automatic speech recognition model called Transcribe. French AI company Mistral also unveiled an open-source text-to-speech model, Voxtral TTS.

Cohere releases voice-focused open-source model to strengthen ASR; Mistral unveils open-source voice generation AI model

Microsoft held an AI Tour Seoul event and proposed frontier transformation, embedding AI as a growth engine beyond productivity improvements, as a key theme. It emphasised Copilot plus agents as technologies that can make frontier transformation a reality. AI MSP company Cloucus also unveiled its enterprise AI platform ClooSphere at the event. Wanted Lab will fully open to external users its agentic design system Montage, which it has advanced in internal working environments.

Microsoft says agent-heavy Copilot will transform business app experience; more companies in South Korea are shifting their makeup through AI-driven frontier transformation; existing security models cannot block threats from AI spread, with ambient and autonomous agents as answers; Cloucus unveils enterprise AI platform ClooSphere focused on supporting real operations; Wanted Lab to 공개 agentic design system Montage externally

As the centre of gravity in AI competition shifts from model performance to token prices, a survey found China has begun to take the lead in this fight. Data from OpenRouter, which tracks AI model usage, showed that from February this year token consumption by Chinese AI models such as DeepSeek and MiniMax surpassed U.S. rivals.

Chinese AI overtakes U.S. in token consumption, expanding global share with price competitiveness