Tech News

The Pentagon's blacklist, self-replicating AI, and broken security pledges—yet still #1 on the App Store. Anthropic is ahead of the curve. 5 Key Points Bioweapon alert issued 10 days before launch — Claude 3.7 Release on emergency hold due to experimental results showing Sonnet could aid terrorists in creating biological weapons. A company valued at $380 billion, surpassing Goldman Sachs —Coding Agent alone generates $2.5 billion in revenue, with the software industry's market cap evaporating by $300 billion with each new product release. Deployed in the Maduro capture operation and broke with the Pentagon — becoming the US government's first classified AI, then branded a "supply chain security risk." Claude makes Claude — AI writes 70-90% of future model code, 427x faster than humans, and one year to fully automated research. Safety pledge retreats, but user surges — The company voluntarily removed its core safety promise, but saw a million new users sign up in a day after the Pentagon clash. An emergency meeting began with the hotel bed flipped over. In February 2025, five members of the Anthropic Red Team received urgent news during a conference: controlled experiment results suggesting that the soon-to-be-released Claude 3.7 Sonnet could aid in the creation of biological weapons. They rushed to their hotel rooms, flipped their beds over to makeshift desks, and spent hours analyzing the data. When no conclusions were reached, the company postponed the launch for ten days. Red Team Leader Logan Graham (31) recalls the day as "a fun and interesting day." Only someone who deals with risk on a daily basis can say that. "You think there's a room somewhere with adults who know the solution. There's no such room. The responsibility lies with you." The philosopher came before the product. Anthropic was founded in 2021 by siblings Dario and Daniela Amodei, both former OpenAI alumni. Altman's decision to go independent stemmed from his feeling that the company was rushing to market without a thorough safety review. The company created a "social impact" team before it even launched its product. Amanda Askell, the company's in-house philosopher, designs Claude's moral sensibility. "It's like teaching a six-year-old what kind of goodness is. By the time they're 15, they'll be smarter than you in every way." Employees call themselves "ants," and job interviews ask, "Are you willing to lose stock value if we abandon the model because we can't guarantee its safety?" All seven co-founders have pledged to give away 80% of their assets. But this idealism has drawn suspicion from both Silicon Valley and the Trump administration. Trump's AI czar, David Sachs, has accused Anthropic of engaging in a "fear-based regulatory capture strategy," and Elon Musk has called the company "misanthropic." Meanwhile, Nvidia's Jensen Huang, despite disagreeing with nearly all of Amodei's views, has acknowledged Claude as an "amazing" model and invested $10 billion in it. Programmers who no longer write code directly In September 2024, Ukrainian engineer Boris Chernyi created Claude Code, a game-changer. While chatbots could only speak, this tool could access files, run programs, and even write code like a human. When Cherny asked, "What music are you listening to?", Claude opened the music player, took a screenshot, and told him the name of the song. Chernyi no longer writes code himself. Coding agent revenue soared from $1 billion in late 2025 to $2.5 billion in February 2026, and by the time the non-coder plugin was released, $300 billion had vanished from the software company's market cap. CEO Amodei himself warns that AI could replace half of entry-level white-collar jobs within one to five years. Deep Ganguly, leader of the social impact team, accurately captures this tension: "It feels like we're talking incoherently." The AI ​​that captured Maduro and the broken alliance On the dawn of January 3, 2026, a US Army helicopter entered Venezuelan airspace and arrested President Maduro. Claude was reportedly involved in the planning and execution of this operation. The US government's first classified Frontier AI—it was Anthropic's greatest, and most dangerous, success. The rift began when the Department of Defense demanded contract renegotiation to allow "all lawful uses." Amodei's red lines were twofold: first, a ban on fully autonomous lethal weapons where AI makes the final targeting decision; second, a ban on mass surveillance of American citizens. Emil Michael of the Department of Defense countered: "We cannot operate an organization of 3 million people under these unthinkable exemptions." On February 24th, Secretary of Defense Hexes summoned Amodei to the Pentagon and issued an ultimatum: Accept the offer by 5 p.m. on February 27th or face designation as a supply chain security risk. Just before the deadline, Anthropic believed they were close to a compromise, but the Pentagon demanded Amodei's presence on the call, and when he was unable to attend, the negotiations were terminated within minutes.

Cameron Adams, Canva's co-founder and Chief Product Officer (CPO), appeared on theCUBE's floor at the New York Stock Exchange (NYSE). He unveiled Canva's latest product, Magic Layers, and spoke candidly about the company's current scale, AI strategy, and future direction. This article summarizes the key points of the interview. Canva, where have we come from? Let's look at the numbers first. Annualized revenue of $4 billion 265 million monthly active users AI products have been used more than 27 billion times. Enterprise sales growth of over 100% for two consecutive years These figures demonstrate that Canva, which launched 15 years ago to make design easy for anyone on web browsers and mobile devices, has now established itself as a global AI visual platform. Cameron Adams describes Canva not simply as a "tool," but as a **Creative Operating System**. This refers to a system that combines three layers: a content layer, an AI layer (its own design model), and a brand and content management platform layer. New Product: Magic Layers The highlight of this interview was the unveiling of Magic Layers . Anyone who's ever used AI to generate images has likely experienced this: After generating an image, even if there's a typo in the text or you want to change the color, you can't edit it. Re-entering the prompt results in a completely different image, and it's nearly impossible to edit just the part you want. Magic Layers tackles this problem head-on. Bring any flat image—whether PNG, JPEG, or AI-generated—into Canva, and it will automatically break down into text, image elements, and more, making them individually editable. Edit text, change colors, and move elements—it's all possible. The core engine of this feature is the Canva Design Model . This foundation model, developed by Canva itself, is the world's first complete design model capable of handling both design creation and editing within an AI workflow. Cameron Adams emphasized that this feature is particularly powerful in multilingual marketing . For example, marketing assets created in English can be instantly localized for Mexican or Indonesian markets by simply changing the text. Even if you lose the original design files, you can restore them to an editable state with just the images. It is currently in public beta , initially available in English, and will be rolled out to international markets sequentially thereafter. A leap forward in our own AI capabilities following the acquisition of Leonardo AI. The decisive turning point that enabled Canva to achieve its current AI capabilities was the acquisition of Leonardo AI . Leonardo was a foundational model company with a world-class research team in the field of visual AI, and the addition of this team to Canva significantly strengthened our AI research team. Canva's AI research team now consists of over 100 researchers . In addition, Canva has acquired several AI companies over the past year. MangoAI and Cavalry have contributed to Canva's strengthened model capabilities, and the acquisition of MagicBrief has expanded its AI automation capabilities. With MagicBrief, users can create entire marketing campaigns with just a few clicks, simply by entering a single product link. Partnerships with OpenAI, Anthropic, and MCP

A16z 2026 Gen AI Consumer Market Report Analysis More than 10% of the global population uses ChatGPT weekly. But the real story behind these numbers is different. The AI ​​market is evolving from a "winner takes all" situation to an ecosystem war between platforms with completely different philosophies. Let's explore five key shifts captured in a16z's latest report. Five Key Takeaways ❶ ChatGPT is the overwhelming leader, but Claude and Gemini are catching up, with paid subscribers growing by 200-258%. The AI ​​market is moving from a search war (90% monopoly) to a mobile OS war (two-party ecosystem). ❷ The global AI market is fragmented into a tripolar system: the West, China, and Russia. Singapore ranks first in per capita AI adoption, South Korea fourth, and the United States 20th. ❸ Image generation is bundled with ChatGPT and Gemini, and standalone products are shrinking. Meanwhile, video, music, and voice AI are building their own, defensible domains. ❹ The era of agents, which began with Vibe coding, has truly arrived. OpenClaw became GitHub's most-starred project and was acquired by OpenAI. ❺ As AI becomes integrated into browsers, desktops, and existing tools, we have reached an era where actual AI usage cannot be measured solely through web visits and app MAU. 1. ChatGPT leads, but the race for the "basic AI" spot has begun. Overwhelming first place, but not a solo runner-up ChatGPT remains the undisputed leader in the consumer AI market. It boasts 2.7 times the web traffic of second-place Gemini and 2.5 times the mobile MAU. Its weekly active users have grown by 500 million in just one year, now reaching 900 million, representing over 10% of the global population. However, competitors are starting to make serious moves. As of January 2026, Claude's paid subscribers are growing by over 200% year-over-year, and Gemini's by 258%. While ChatGPT is still eight times the size of Claude and four times the size of Gemini, its competitors are growing much faster. What's Changed: Competitors Strike Back Google has achieved a resounding victory in creative modeling. Nano Banana generated 200 million images in its first week, bringing 10 million new users to Gemini, and Veo 3 has been hailed as a turning point in AI imaging. Anthropic has strengthened its prosumer strategy. With Cowork, Claude in Chrome, Excel and PowerPoint plugins, and most notably, Claude Code, it's moving toward deeper integration with the everyday tools of professional users. The App Store Wars: The Beginning of Ecosystem Lock-in True lock-in begins in the app ecosystem. ChatGPT boasts 220 apps across 13 categories, while Claude boasts approximately 160 connectors and over 50 community MCP servers. However, the two platforms share only 41 apps, or about 11% of the total. ChatGPT employs a "superapp" strategy, with over 85 apps across consumer transaction categories like travel, shopping, food, and health. Claude, on the other hand, focuses on specialized areas like financial data terminals (PitchBook, FactSet), development infrastructure (Sentry, Supabase), and scientific and medical tools (PubMed). The report predicts that the outcome of this competition will not be a 90% market share for one player, like in the search market, but rather a mobile OS war, with each player building a trillion-dollar ecosystem based on a different philosophy. 2. The global AI market is fragmenting into a three-polar system. West vs. China vs. Russia

Want to implement AI solutions but are tired of signing separate contracts for each tool? Now, your worries may be over. Anthropic has officially launched the Claude Marketplace . Simply put, it's a B2B marketplace that gives you instant access to multiple Claude-based partner solutions for the same Anthropic contract price . Why should you pay attention? Companies know what they want to solve with LLMs. The problem is that building them on their own is too difficult. Conversely, AI-native startups are already developing effective solutions, but they struggle to reach enterprise customers. Claude Marketplace connects the two. The most groundbreaking point is the billing structure . New payment methods? Separate procurement processes? No need. The cost of partner tools is covered by the Anthropic commitment you're already paying . Which partner came in? The lineup is strong from the very first entry. GitLab — Automating the entire software development cycle with AI. Harvey — Handle complex legal matters with Claude Lovable — Build and deploy real apps, even for non-developers. Replit — Develop production-grade software using natural language. Snowflake — Enterprise Data Analysis, Transformation, and Action Financial-specific tools —from IR decks and financial models to corporate and market research. Cox Automotive's CPO said, "We've accelerated our AI transformation by streamlining procurement while extending it to partner tools." If you're an AI company, now is your chance. If you're a company building a Claude-based product, you can apply to join our partner waitlist . This opens a direct channel to enterprise customers who have already invested heavily in Anthropic. 👉 Corporate Customers : Talk to Sales 👉 Partner Application : Join the waitlist

Key Summary A new measurement method emerges : A new metric called "Observed Exposure" has been developed that combines theoretical AI capabilities with real-world usage data. Theory ≠ Reality : There is a huge gap between what AI can theoretically do and what it actually does. The most dangerous jobs are : computer programmer (74.5%), customer service representative (70.1%), and data entry worker (67.1%). No change in unemployment rate yet : No systematic increase in unemployment in AI-exposed occupations has been observed since the end of 2022. However, youth employment is showing signs of slowing : Early signs of a 14% decline in AI-intensive occupations for young people aged 22-25. Characteristics of highly exposed workers : highly educated, high-income, and a high proportion of women—targeting white-collar workers, not low-skilled, simple workers. "What AI can do" and "what it actually does" are completely different. On March 5, 2026, the Anthropic research team (Maxim Massenkoff and Peter McCrory) published an intriguing report. While previous AI labor market studies focused on "how many jobs could theoretically be replaced by AI," this study poses a completely different question: "How much are people actually using AI in their work?" The research team combined the O*NET occupational database, real-world usage data from Claude AI, and theoretical exposure assessments by Eloundou et al. to create a new metric they call " Observed Exposure ." The key to this metric is that it measures not simply whether AI can perform a task, but whether it is actually being used in an automated manner in the workplace. The results are astonishing. For example, in computer science and mathematics, while AI can theoretically perform 94% of tasks, its actual observed usage rate is only 33%. This represents a significant gap between theory and practice. Tasks like distributing prescription information to pharmacies are perfectly capable of being handled by AI, but hurdles like legal restrictions, software requirements, and human verification processes prevent its practical use. Which profession is most exposed to AI? The report's top 10 occupations with the highest AI exposure include some surprising findings, as well as some expected ones. Ranking Job Observed exposure Key automation tasks 1 Computer programmer 74.5% Software writing and maintenance

AI is everywhere, but it's invisible in productivity statistics. February 2026 | Comprehensive summary of key research from Fortune, NBER, UC Berkeley, and more. 1. Key takeaways: What's happening? **Approximately 90%** of 6,000 CEOs and executives responded that AI had no impact on employment or productivity in the past three years ( NBER, February 2026 ) Executives who use AI spend an average of just 1.5 hours per week 25% of respondents said they do not use AI at all at work. 374 S&P 500 companies mentioned AI in their earnings calls, but there was no change in actual productivity metrics. Apollo's chief economist, Torsten Slok, quoted Nobel laureate Robert Solow's observation from nearly 40 years ago: "AI is nowhere to be seen in employment data, productivity data, or inflation data." 2. The Solow Productivity Paradox: History Repeats Itself Lessons from the IT era of the 1980s In 1987, Nobel Prize winner in economics Robert Solow discovered a paradoxical phenomenon of the information age: with the advent of transistors, microprocessors, and integrated circuits in the 1960s, companies expected a surge in productivity, but the opposite happened. Period Productivity growth rate Situation 1948~1973 2.9% Before IT introduction Since 1973 1.1% Paradoxes arise after IT spreads

On December 10, 2025, something interesting happened in the AI industry: Antropic donated its core technology, the Model Context Protocol (MCP), to the Linux Foundation. But there's something else that's truly surprising about this news. A foundation created by rivals The MCP donation is being made to a new foundation called the Agentic AI Foundation (AAIF). The membership of this foundation is truly astounding. Co-founder Anthropic (the company that made Claude) Block OpenAI (the company that created ChatGPT) Supporting companies Google Microsoft AWS Cloudflare Bloomberg ChatGPT and Claude are fierce rivals in the current AI market. However, they have decided to form a foundation together and share their technology for the public good. It's akin to Samsung and Apple joining forces to create a smartphone standard. What is MCP? Model Context Protocol, or MCP for short, is a general-purpose, open standard for connecting AI applications with external systems. Simply put, it's a "common language" for AI to communicate with external data and programs. This technology, first introduced by Anthropic a year ago, has grown by leaps and bounds. Achievements in One Year Spread throughout the ecosystem There are currently over 10,000 active public MCP servers in operation, covering virtually every area, from small developer tools to systems used by Fortune 500 companies. Selection of major platforms

AI company Antropic conducted an interesting survey of 132 of its employees. It explored how people working at companies that create AI are actually using AI and how it's changing the way they work. Amazing productivity gains The most striking result is a dramatic increase in productivity. Just a year ago, employees were using AI in 28% of their tasks and experiencing a productivity increase of around 20%. Now, they use AI in 60% of their tasks and report a 50% increase in productivity. Some power users have even reported productivity increases of over 100%. What's interesting is that while the time spent on each task has decreased slightly, the overall workload has increased significantly. In other words, AI has enabled us to get more done in the same amount of time. What I do, what I leave to AI Employees most commonly use AI for bug fixing (55%) and code understanding (42%). They are particularly inclined to entrust AI with the following tasks: Something that is not complicated but that I am not familiar with Things that make it easy to check the results Repetitive and boring work Code quality isn't that important "The more exciting and fun the task is, the less likely you are to use AI," said one engineer. "But for tasks that feel challenging, it's easier to start by talking to AI." Everyone becomes 'full stack' The most exciting change is that employees are now working beyond their own areas of expertise. Backend developers are creating UIs, security teams are analyzing unfamiliar code, and researchers are creating data visualization tools. "I can now handle front-end, database, and API code," said one backend engineer. "I used to be afraid to touch areas I wasn't an expert in." Employees report that AI has enabled them to tackle small improvement tasks that were previously considered low-priority. Twenty-seven percent of all tasks would not have been completed without AI. But there are also concerns. It's not all good news. Many engineers have concerns: Concerns about a decline in technical skills : Some say, "Because AI provides immediate answers, we have less time to deeply understand how systems work." Developers, particularly those with less experience, worry about losing opportunities to learn skills "the hard way." Reduced interaction with colleagues : With AI becoming the first point of contact for questions, opportunities for collaboration and mentoring with colleagues have diminished. Some people have commented, "I enjoy working with people, but it's sad that they're 'less necessary' now." An uncertain future : "I'm optimistic in the short term, but in the long term, I think AI will do everything and I and many other people will no longer be needed," he confessed. The meaning of coding is changing

Key Summary ⚡ Vision : Building a fairer, stronger Australia with AI - seizing opportunities, spreading benefits, and ensuring safety. Investment Size : Currently over $460 million in AI-related investments + $10 billion in data center investments 3 Goals × 9 Actions : Infrastructure, Talent, and Investment / Adoption, Education, and Public Services / Safety, Responsibility, and Global Cooperation Establishment of the AI Safety Institute (AISI) : Support for AI Risk Monitoring and Regulation A Worker-Centered Approach : Ensuring Union and Worker Participation in AI Implementation and Fair Benefit Distribution Why AI Plan Now? 🚀 Australia is already an AI powerhouse, ranking third globally in AI tool usage per capita and second globally in data center investment (after the United States) by 2024. However, seizing this opportunity requires a strategic approach . The government is pursuing AI policies based on the core values of fairness, inclusion, and opportunity, driven by the philosophy of creating a world where "technology works for people." Success is measured not simply by technological advancements, but by the benefits enjoyed by all Australians . 🎯 GOAL 1: Seize the opportunity Action 1: Building Smart Infrastructure Data centers are the answer Australia has announced over $100 billion in data center investment between 2023 and 2025. Major projects are lined up, including $5 billion from Microsoft, $20 billion from Amazon, and $73.3 billion from Firmus. Why Australia? Stable legal environment Abundant renewable energy potential Proximity to the Asia-Pacific region Advanced chip accessibility 15 international submarine cable connections Sustainability is key The government is working with state governments to develop "Data Center Principles." These principles encourage environmentally friendly growth, including the use of renewable energy and water-saving cooling technologies. For example, CDC Data Centers use 100% carbon-neutral electricity and use closed-loop cooling systems to reduce water consumption to near zero. NBN expansion underway We are upgrading the National Broadband Network (NBN) to provide fast and stable internet connectivity nationwide, and improving accessibility in regional and remote areas through low-orbit satellites and expanded mobile coverage.

Key Highlights 🎬 Visual Suite has been significantly upgraded Video Editor 2.0: Professional-Level Editing Without Watermarks Interactive forms: Data collection tools that can be embedded directly into your website. Canvas Code Data Integration: Connect with Sheets to Create Real-Time Interactive Apps Email Design: Create and export marketing emails directly from Canva. 🤖 The world's first Canva design model released Integrated AI that understands text, images, and layout at once Integrated with ChatGPT, Claude, and Gemini, so you can use it anywhere. AI-generated content right from the search bar: 3D elements, photos, and even videos. "Ask Canva AI": Get real-time design advice in the comments. 📈 Business Growth Solutions Canva Grow: From ad creation to performance analysis, all in one place. Brand System: AI automatically builds brand kits and maintains company-wide consistency. Canva Business Plan: Advanced Marketing Insights 💎 Affinity, permanently free Vector, pixel, and layout tools all in one app Professional design tools provided free for life Combine generative AI and manual work with Canva AI Studio. The age of imagination has arrived

Observations started immediately after the launch of ChatGPT in 2023. Since 2023, the Wharton Human-AI Institute and marketing consulting firm GBK Collective have been asking the same question every summer: "How is your company actually using generative AI?" This study, now in its third year, was conducted from June 26 to July 11, 2025, with decision-makers from 801 mid-sized and large U.S. companies with annual revenues of $50 million or more. Respondents held significant decision-making power in organizations with 1,000 or more employees: HR managers, IT executives, legal department heads, marketing and sales executives, operations managers, product development managers, purchasing managers, finance directors, and even top executives. What they revealed in this 15-minute online survey isn't just headlines; it's the real story on the ground. Key findings • It has become a tool used every day. Started measuring investment returns It is used to improve employee skills, but concerns about a decline in proficiency have also grown. Top management began to manage it directly. Budgets increase, but education decreases. AI entering everyday life In June 2023, six months after ChatGPT's launch, only 37% of executives surveyed said they used generative AI at least once a week. Most approached it with a mixture of curiosity and caution. A year later, in 2024, that number had skyrocketed to 72%. Usage had nearly doubled, and spending had increased by 130%. While the novelty had worn off, the response was still positive. As of 2025, 82% use it at least once a week. A more significant change is the daily usage rate: 46% use generative AI daily, a 17 percentage point increase from 29% in 2024. For IT departments, 68% use it daily. 52% of purchasing and procurement departments, 51% of HR, and 51% of finance and accounting departments use it daily. Marketing and sales use it relatively low at 34%. By company size, 87% of large companies with annual revenues of $2 billion or more use it at least once a week, a 22 percentage point increase from 65% in 2024. The gap with small and medium-sized businesses is narrowing. By industry, technology and communications (94%), finance (90%), and professional services (92%) lead the way. Manufacturing (80%) and retail (63%) lag behind. What is it used for? The most common use is data analysis, with 73% using it. Document and meeting summarization ranked second (70%), followed by document creation and editing (68%). Presentation and report writing, and idea brainstorming followed, with 68% and 66%, respectively. Marketing content creation (66%), customer service (66%), email writing (64%), internal help desk (64%), and sales collateral writing (64%) round out the top 10. Job-specific uses also emerge. IT scores 123 points higher in coding (compared to the overall average). HR scores 129 points higher in employee recruitment and onboarding, and legal scores 133 points higher in contract drafting.

"It would be a truly sad future if AI delegated all creative work to robots." Designer Pablo Stanley speaks frankly, but adds, "I don't think that will happen. It will actually be a more collaborative utopia." Pablo is the co-founder and designer of Lumi, an AI-powered image collection platform. Recently acquired by Udemy, Lumi is a collection of images created by thousands of AI creators who were tired of boring, repetitive stock images. He is also the co-founder of several AI-powered creative tools, including Blush and Mushu. But his journey began with a pure childhood fascination with robots. A child growing up in the age of robots "I grew up in a time when robots were the coolest thing ever. Movies, comics, and pop culture were filled with all kinds of versions of robots." Pablo takes up his pen and spends 90 seconds drawing on his childhood inspirations: the Jason Family robot, RoboCop, Blade Runner... The robots of that era weren't simply machines. They had personalities, and they asked themselves, "Am I alive?"—and humans pondered that question along with them. "A dystopian world where robots are not accepted... We're there now." The sketches he drew are reminiscent of the robots people imagined in the 1960s: square bodies, metallic textures, and rugged machines. His proposal is to return to the robots we imagined. "People who make decisions before even trying AI" Many creative professionals are still hesitant about AI. Pablo attributes this to fear of the unknown and the low quality of early AI results. "Just a year ago, we would have laughed at images of six-fingered figures or strange limbs. Such flaws made it easy to dismiss AI. But that's no longer the case." His advice to creators is clear: start with low-risk projects. "I usually tell people to start with something stupid—a low-risk, silly project, just a silly idea. Then you can test what you can actually do and start seeing how it can be applied to real-world work." Lessons Learned from the Browser Wars Everything moves quickly. New tools emerge, existing ones improve... How can you keep up with the times in your daily work? Pablo recalls the browser wars of the past. When one browser developed a feature, others copied and implemented it. Users benefited from this competition. "AI tools are now looking at each other. They're competing, and when a competitor comes up with a useful feature, they try to implement something similar. Now, they all have very similar capabilities." His conclusion: Pick one and stick with it. You'll still get good results, and you won't miss out on much.

The AI market landscape has changed in just six months. As of the end of 2023, OpenAI held a monopoly, holding half (50%) of the enterprise AI market. But as of 2025, something surprising has happened. Anthropic's Claude has taken the top spot with a 32% share, while OpenAI's share has nearly halved to 25%. What on earth happened? The story of Claude's capture of developers' hearts Code generation becomes AI's first killer app. "Currently, 100% of our production workloads run on premium models like Claude. We initially started with POCs using Llama and DeepSeek, but over time, they couldn't keep up with Claude's performance." This quote from a corporate developer illustrates the current situation well: Claude holds a 42% market share, particularly in code generation, more than double OpenAI's 21%. With the release of Claude Sonnet 3.5 in June 2024, the AI development ecosystem has completely changed. What started as a single market, GitHub Copilot, has grown into a massive $1.9 billion ecosystem, and a plethora of new categories have emerged, from AI IDEs like Cursor and Windsurf to app builders like Lovable and Bolt, and even enterprise coding agents like Claude Code. Pioneer of the agent era 2025 is being called the "Year of the Agent." Unlike existing AIs that simply provided answers, Claude has evolved into a true agent capable of thinking step-by-step, reasoning through problems, and leveraging external tools. By integrating various resources, including search, calculators, and coding environments through the Model Context Protocol (MCP), Claude has become a much more useful tool for real-world tasks. Amazing achievements in numbers Explosive growth indicators Revenue growth : $87 million in early 2024 → over $5 billion by August 2025 Enterprise customers : less than 1,000 two years ago → currently over 300,000 (300x increase) Funding : $13 billion Series F, valuation $183 billion API spending more than doubled from $3.5 billion to $8.4 billion in six months. Performance of actual companies NBIM (the world's largest sovereign wealth fund): 20% increase in productivity, 213,000 hours saved. Novo Nordisk : 99.9% reduction in clinical documentation time (10 weeks → 10 minutes) SK Telecom : 34% Improvement in Customer Service Quality for Millions of Users

How would Amazon respond if its generative AI lied? AI is a liar? 😱 While the generative AI craze is raging, there's one problem that every business is grappling with: AI lies too often . Creating information that doesn't exist 📝 Missing important information 🕳️ I have tons of ideas, but it's hard to know what's really effective 🤷‍♀️ So, most companies rely on human review or use separate testing tools. However, these methods are expensive and slow, making it difficult to verify all AI-generated content. Amazon's Innovative Solution: Catalog AI ✨ Amazon approached this problem in a completely different way: They created a system called "Catalog AI" that verifies AI with AI ! 📊 Amazing achievements Generate and test tens of millions of hypotheses annually (compared to just a few thousand in traditional systems) 8% of suggestions contribute to actual sales increase 80% of AI results pass quality verification (initially 20%) Amazon's 4-Step Quality Management System 🎯 1️⃣ Audit: Face the Reality We set a benchmark by comparing the AI-generated product pages with information we already know . This is the same principle as quality control in manufacturing. 2️⃣ Guardrails: Triple safety device Simple rules : Basic rules like "Weight must be followed by units (kg, lb)" Statistical Profile : Create guardrails with common table information, warning when out of range. AI checks AI : A system where the first AI creates content, and then the second AI reviews it. 💡 Tip : AIs trained with different data can check each other to catch more errors! 3️⃣ Effectiveness Testing: The Evolution of A/B Testing Automatically run all AI suggestions through A/B testing.

Over the past 21 years, Google.org has invested $6 billion globally, supporting 3,600 nonprofit organizations and 1,700 universities, expanding its reach to 160 countries. The US AI Opportunity Fund has invested $90 billion to empower one million people through AI education, providing comprehensive support ranging from job training to digital safety education. From AI flood prediction to fire detection satellites, Google has demonstrated concrete results in solving real-world social problems through science and technology, saving lives. New inequalities brought about by the technology gap With the advent of the AI era, the world stands at a new crossroads. According to the World Economic Forum, the top three skills most in demand by the American workforce by 2030 will be AI/big data, networking/cybersecurity, and technological literacy. But what's the reality? Only half of the world's workers have access to adequate AI training opportunities. Farmers in developing countries lose up to 50% of their crop yields due to lack of access to crop pest information, and hundreds of thousands of people die each year due to inadequate natural disaster prediction systems. Technology has advanced, but its benefits are not reaching those who really need them. Knowledge, Technology, and Learning: Leading the Way in AI Democratization Key Achievements: Providing computer science education to 280,000 students 4.1 million people completed training in high-paying jobs Support for the creation of 456,000 jobs Main Programs: AI Opportunity Fund : AI training for 1 million people in the US (worth $90 million) Goodwill Partnership : 400,000 People Participate in Digital Career Acceleration Program Blockly Accessibility Fund : Developing Coding Education Tools for Children with Visual and Motor Impairments Real-world example : Jesslyn Miller, owner of the fashion brand Unique & Lovely, received AI training from Goodwill and automated product description writing and social media content with GenAI. The result? She achieved 6,000 additional content impressions in just one month. Science Advancement Area: From Lab to Reality Key Achievements: Support for 27,000 student researchers 16,000 research awards provided AI adoption reduces average work time by two-thirds and reduces costs by 50%.

OpenAI has finally released the update developers have been waiting for. Codex has evolved from a simple coding tool to a true AI development partner. The newly released GPT-5-Codex is a model optimized for real-world software engineering workflows, capable of handling everything from code reviews to large-scale refactoring independently. The core of this update is clear: to provide a faster, more reliable, and ubiquitous development environment . 1. GPT-5-Codex: An AI engine specialized for practical development. Innovation as revealed by performance metrics The new GPT-5-Codex shows remarkable performance improvements over the original GPT-5: It achieved 74.5% accuracy on the SWE-bench Verified test , and significantly outperformed GPT-5's 33.9% accuracy on code refactoring tasks with 51.3% accuracy. The most impressive part is its adaptive thinking time . For simple requests, it provides a quick response using 93.7% fewer tokens, but for complex tasks, it thinks twice as long and produces thorough results. The reality of working independently for 7 hours straight During testing, GPT-5-Codex performed independent work for over seven hours, continuously implementing, fixing test failures, and iteratively improving, ultimately resulting in a successful implementation. This demonstrates a true development partner, going beyond simple code generation. 2. Upgraded Codex CLI: A New Standard in Open Source Development Start with installation The new Codex CLI has been completely rebuilt based on community feedback. Designed around the agentic coding workflow, it becomes a more powerful and reliable partner. Major improvements Image Support : Share screenshots, wireframes, and diagrams directly from the CLI to build shared context for design decisions. Track progress : Track progress on complex tasks with to-do lists, and connect to external systems via web searches and MCP. Improved Terminal UI : Tool calls and differences are better formatted and easier to follow. 3-step approval mode : Read-only : requires explicit permission Automatic : Full access within the workspace, requires external approval Full access : Can read files and execute network access commands. 3. IDE Extension: Full Integration of the Development Environment

3-line summary Wealthier countries and regions use AI more , with a strong correlation found: a 1% increase in GDP leads to a 0.7% increase in AI usage. Businesses are 77% more likely to favor automation than individuals, suggesting a major labor market shift is imminent. Automation rates surge from 27% to 39% in just nine months , with people handing over more responsibilities to AI. Problem Statement: The Emergence of New Technological Inequalities Anthropic's latest economic index report reveals a shocking reality: AI adoption is starkly divided by economic power and geography, potentially replicating the economic disparities previously caused by general-purpose technologies like electricity and the internal combustion engine. Looking at the current situation: Developed countries such as Israel, Singapore, and Australia overwhelmingly lead in AI usage per capita. Within the US, Washington, D.C., Utah, and California recorded the highest usage rates. Income level is a key factor in determining AI usage. Even more serious is the difference in usage patterns . High-income countries prefer collaborative approaches with AI, while low-income countries rely on simple automation. This suggests a gap in AI utilization capabilities itself. The Solution: Building an Inclusive AI Ecosystem 1. Accessibility Expansion Policy Mandatory AI Education Programs : Incorporating AI Literacy into All Curricula Infrastructure Investment : Expanding Internet connectivity and computing resources in low-income areas. Breaking Down Language Barriers : Developing AI Models That Consider Diverse Languages and Cultural Contexts 2. Bridging the gap between businesses and individuals Data showing that businesses are 77% more likely to embrace automation than individuals suggests a dramatic shift in the labor market. To address this: Expanding Retraining Programs : Developing New Job Skills Collaborating with AI Strengthening the Social Safety Net: A Support System for Job Displacement Due to Automation Guidelines for Ethical AI Adoption : Preventing Indiscriminate Automation in Businesses

Companies are seeing tangible results from their AI investments. A recent study by Google Cloud found that 88% of leading companies adopting AI agents are achieving a positive ROI from generative AI , significantly exceeding the overall enterprise average of 74%. Now that the era of AI agents has truly arrived, how can you maximize your investment? 7 Strategies for Maximizing AI Agent ROI 1. Secure C-level sponsors Organizations with comprehensive executive support are achieving 78% ROI from AI. The first condition for successful AI adoption is a clear vision and support from top management. Beyond simply securing approval for the technology, it's crucial to secure comprehensive understanding and support of how AI aligns with business goals. Eric Lambert, Vice President of Legal at Trimble, emphasized, "Leaders must first determine what ROI means. It goes beyond simple financial returns." The performance of AI investments should be measured across multiple dimensions, including improved efficiency and achievement of business objectives. 2. Establishing a data governance and security system 37% of companies cited data privacy and security as their top consideration when choosing an LLM provider . Building trustworthy AI systems requires robust data governance and an enterprise security framework. Natalie Bowman, Director of Data Management at Alaska Airlines, noted, "The biggest security concern for LLMs is the risk of malicious actors accessing the data, causing hallucinations, or altering the data." We must establish systematic security policies to ensure data integrity and prevent vicious cycles, and maintain a system that always allows for human intervention. 3. Start with the areas where you can make the biggest impact. The effects of AI were found to be greatest in the following order: productivity (70%), customer experience (63%), and business growth (56%). Not all AI projects deliver equal value. Focus on building AI agents that can automate repetitive tasks to achieve a clear ROI. Specifically, in the area of individual productivity, 39% of companies have already achieved ROI, and of those reporting productivity gains, 39% said employee productivity has at least doubled. 4. Granting AI agents appropriate tool access. For AI agents to perform practical tasks, they need access to internal enterprise systems like CRM or Drive . Building AI agents that go beyond simple chatbots and perform real-world tasks requires providing secure and controlled access to the necessary systems and data. "AI agents can be applied to a wide range of use cases, and every business has workflows where agentic AI can deliver meaningful value," said Fiona Tan, CTO of Wayfair. 5. Building a Scalable AI Rulebook As AI use increases, so do the risks, so it's important to establish clear, company-wide guidelines early . "AI technology is evolving rapidly. Even a year ago, very few people were talking about AI agents and agentic AI at the enterprise level," said Cristina Nitulescu, director of digital transformation at Bayer Consumer Health, emphasizing the importance of redesigning processes for the future. We need to establish clear standards and procedures for data security, intellectual property protection, and compliance now.

Key Summary • Big Tech Showdown : Google Gemini Challenges ChatGPT's Stronghold, Chasing Second Place on the Web and Half the Mobile Market • Emergence of Chinese AI Legion : 3 of the top 20 web apps and 22 of the top 50 mobile apps are Chinese and are dominating the global market. • Accelerating the Coding Revolution : Lovable Rises to 22nd Place, Entering the Era of AI-Powered App Creation • Ecosystem overhaul : New entrants reduced to 11, clearing the line between winners and losers. • All-Star Club : 14 companies make the top list for the fifth consecutive time, emerging as true contenders in the AI market. ChatGPT ChatGPT still holds the top spot across both web and mobile, maintaining its dominant position. However, Google Gemini is threatening to overtake ChatGPT, with a 12% share of web traffic and half the number of monthly active users (MAU) on mobile. In the Chinese market in particular, access restrictions are forcing ChatGPT to cede market share to local competitors. Google Gemini Google's ambitious Gemini achieved the most notable performance in this survey. It secured a solid second-place position, second only to ChatGPT, at second place on the web and second on mobile. Notably, it secured a 90% user base on Android devices, demonstrating the power of the Google ecosystem. In addition to Gemini, Google also placed four other products in the top 50 simultaneously: AI Studio (10th), NotebookLM (13th), and Google Labs (39th), demonstrating its strong presence in the AI market. China's AI Empire The entry of Chinese companies into the global AI market is notable. Among the top 20 web developers, three companies—Quark (9th), Doubao (12th), and Kimi (17th)—all operate Chinese websites, with over 75% of their traffic coming from China. Conversely, while 22 of the 50 mobile apps were developed in China, only three are actually primarily used in China, demonstrating a clear "develop in China, export globally" strategy. In particular, Chinese companies' advantage over Western counterparts in video generation is attributed to their larger research team and relatively lax intellectual property regulations. Vibe Coding The "Vibe Coding" trend, where AI directly creates apps, is experiencing explosive growth. Lovable has soared to 22nd place on the Brinklist, and Replit has also entered the main list. According to credit card panel data, the US user cohort of a major Vibe Coding platform has demonstrated over 100% revenue retention for several months after signing up, demonstrating a sustainable business model. This means that even if users churn, overall revenue grows month-over-month due to increased usage by remaining users. Grok Grok, the AI assistant from X (formerly Twitter), has seen remarkable growth, especially on mobile. From a "cold start" in late 2024, it now boasts over 20 million monthly active users. In July, the launch of the Grok 4 model (July 9) and the introduction of AI companion avatars (July 14) led to a nearly 40% user increase. The animated avatar "Ani," which includes NSFW options, proved particularly popular upon its launch. DeepSeek DeepSeek, once a leading player in the Chinese AI craze, is experiencing a steep decline. Its web usage has fallen by more than 40% since its peak in February 2025, and its mobile usage has also declined by 22%. This suggests that, despite its initial buzz, it is struggling to maintain sustained user retention. All-Star Club The 14 "all-star" companies that have made the Web Top 50 for five consecutive years have established themselves as true winners in the AI consumer market. They cover a wide range of areas, including general assistants (ChatGPT, Perplexity, Poe), companions (Character AI), image generation (Midjourney, Leonardo), video and image editing (Veed, Cutout), voice generation (Eleven Labs), productivity tools (Photoroom, Gamma, Quillbot), and model hosting (Civitai, HuggingFace). Interestingly, only five of the 14 companies have their own foundation models, suggesting that the API-based model is more effective. They hail from five countries: the US, the UK, Australia, China, and France, and, with the exception of Midjourney, all have received venture funding.

Key Features Unidentified AI model quietly appeared in LMArena without an official announcement. Excellent image editing capabilities - Accurately handles complex multi-step editing commands Natural language-based editing - Edit using natural language commands like "Change to a dark suit" or "Soften the lighting." Maintain consistency - Naturally preserve the lighting, perspective, and composition of the original image. Preservation of character identity - accurately maintaining the same character's characteristics across multiple scenes. High-quality results - professional-level image creation and editing quality Supports a wide range of styles - from photorealistic to character art Core Features Text-to-Image Generation - Create images in various styles with text descriptions. Local image editing - maintain the overall structure while modifying only specific areas. Style conversion - Convert to various art styles such as realistic, watercolor, oil painting, etc. Multi-turn interactive editing - create perfect results with step-by-step refinements. Background Replacement and Removal - Natural Background Change Work Object Removal and Correction - Remove unnecessary elements and automatically correct them. How to use Step 1: Approach Access the LMArena site (lmarena.ai) https://lmarena.ai/ ? Use the "Generate Image" or "Image Edit Arena" function Repeat until Nano-Banana is randomly assigned Step 2: Upload your image