The age of agentic artificial intelligence (AI) arrived in the fall of 2025, and 2026 may determine who leads it. Agentic AI systems have crossed a critical threshold. Anthropic’s Claude Code achieved autonomous coding capabilities that compress development cycles by orders of magnitude. In the words of a senior Google engineer, Claude Code “generated what we built last year in an hour.” Meta’s acquisition of Chinese startup Manus following its relocation to Singapore reveals how corporations are navigating the competitive landscape in this new agentic AI paradigm.
Anthropic documented how Chinese hackers are already automating cyberattacks with AI agents, the White House is racing to harness them for scientific breakthroughs through the Genesis Mission, and Chinese tech companies like ByteDance are beating many American firms to market with applications of agentic AI like agentic-integrated smartphones. These parallel developments signal a critical inflection point: the nations and companies that master AI agents and social disruption first will reshape global commerce, security, and governance for decades to come.
These are not isolated incidents but harbingers of a fundamental shift in how nations will compete for technological supremacy and the power to define how AI integrates into the fabric of modern society.
The Rise of Agentic AI
The historic roots of agentic AI date back to Alan Turing, who saw a future where autonomous agents would execute complex tasks, make decisions, and coordinate activities without constant human oversight. Already today, the first stage is centered on reliable AI agents designed to perform defined tasks, such as coding, which will then scale into an agentic AI framework enabling multi-agent collaboration across complex enterprise ecosystems.
The business use cases are exploding, and consumers are slated to tap into personal agents in 2026. McKinsey projects that the genesis of large language models (LLMs) and generative AI is giving way to an agentic era, which will unlock considerable productivity gains. Already, AI agents like Claude Code are completing coding tasks at rates that beat humans. Despite breakthroughs and forecasts, the timeline for mass adoption is cloudy.
Sam Altman predicted that agents would join the workforce in 2025, while OpenAI co-founder Andrej Karpathy initially rejected existing agents as “agentic slop,” but his experience with Claude Code changed his opinion. Nevertheless, a Gartner report estimates that 40 percent of today’s agents will not survive to 2027 and instead will be replaced by superior agent options in an overall market correction. The coming generation of agents will usher in experimentation on a grand scale that embeds agents in reimagined business, government, and national security workflows. Significant technical hurdles still exist to realizing the vision for mass adoption.
Technical Hurdles and the Race for AI Dominance
Agentic AI faces formidable engineering challenges—from hallucinations to compute constraints before mass adoption. But this will not stop US companies from innovating to stay ahead of competitors.
Scaling multi-agent systems encounters fundamental technical barriers: bandwidth limitations overwhelm network infrastructure when thousands of agents exchange information in real-time, synchronization challenges intensify as agents struggle to maintain consistent states across dynamic environments, and cascading failures threaten system stability when individual agent malfunctions propagate through interconnected networks.
US AI labs and startups are racing to lead in AI agents. OpenAI, Anthropic, Amazon, Google, Meta, and xAI are vying with a host of startups to scale agentic AI. Google’s Gemini and Microsoft’s Copilot are embedding agentic capabilities into cloud, productivity, and developer platforms. Domestically, companies are deploying platforms such as Salesforce’s Agentforce, United Parcel Service’s (UPS) Orion, and visionary projects such as Amazon’s Agentic Store to revolutionize commerce. To support mass adoption, Nvidia’s AI factories are laying the infrastructure foundation for the compute layer necessary to run agentic AI.
Other companies, such as Airbnb, are also adopting China’s open-source models for agentic applications. Airbnb CEO Brian Chesky announced that his company relies on Alibaba’s Qwen for its customer service agent, praising its high quality and being “fast and cheap.” Some tech observers speculated that startup Cognition’s coding agent SWE-1.5 was based on Zhipu AI’s GLM-4.6 model, and Cursor’s coding tool appears to be built on Chinese models.
China’s Aggressive Push into Agentic AI
This adoption reflects China’s thriving AI ecosystem, which has aggressively entered this space and arguably, before Claude Code, led in agentic performance. Chinese AI leaders are likely to continue to rush AI agents to market in 2026 to keep pace with the United States and, most importantly, with Chinese competitors. ByteDance, Alibaba, DeepSeek, and more startups debuted or announced forthcoming AI agents in 2026. And as Airbnb’s experience shows, Chinese companies are a force to be reckoned with.
Chinese companies’ entry into agentic AI will only pick up steam from an impressive start. Manus snatched headlines in March 2025 when it released its first agent, and evaluations of Moonshot’s Kimi K2 model lauded its performance in agentic reasoning. Startups like Manus, Moonshot, Zhipu AI, and DeepSeek now compete with China’s tech giants Alibaba, Tencent, and ByteDance. More strategically, many Chinese agentic models are open-source, making them easily accessible to developing nations that lack the resources for expensive US alternatives.
Beijing’s Global AI Governance Action Plan, which debuted at the World AI Conference and High-Level Meeting on Global AI Governance in July, illustrates the push to promote its AI offerings in the Global South. China’s initiatives generate another vector for Chinese influence to expand globally through the Brazil, Russia, India, China, South Africa (BRICS) coalition and Belt and Road Initiative, offering AI packages alongside infrastructure investments. This creates a compelling alternative to Western AI governance models. Such opportunities will be enticing for countries seeking to modernize with agentic AI.
The Future of Warfighting and Cyber War
The civilian applications of agentic AI naturally extend into national security domains, where the implications prove equally transformative and concerning. Autonomous agents can boost intelligence processing by synthesizing data streams from space-based assets, unmanned systems, and terrestrial sensors to generate real-time operational pictures. These mosaics accelerate decision-making at tactical and strategic levels while functioning as force multipliers for wargames and planning, drone swarms, and logistics coordination. Agentic tools, nevertheless, create new vulnerabilities that malicious cyber actors will exploit.
Anthropic’s revelations about Chinese cyber operations demonstrated how AI agents will accelerate and automate cyberattacks, and not just for powerful nations. The challenge lies in the democratization of cyberattack capabilities with agentic AI. This will reduce technical barriers that may enable previously constrained actors to conduct operations at an unprecedented scale. States and non-state actors previously unable to conduct advanced cyber warfare may soon project power far beyond their traditional capabilities. The cybersecurity risks are particularly acute as both defensive and offensive operations adopt agentic capabilities.
These agentic systems represent novel attack surfaces that malicious actors can exploit for data poisoning and theft, enterprise-wide network attacks, or coordinated infrastructure disruption. Companies like CrowdStrike deploy defensive AI agents to counter these threats in an accelerating race between autonomous attackers and defenders. Public-private red-team testing of agentic AI, like that conducted by OpenAI and the US Center for AI Standards and Innovation, may reduce the risk of malicious cyber actors using agentic AI platforms for offensive attacks.
Prudent Policy for the Agentic Era
These developments present several critical challenges for the United States that require immediate policy attention. American AI companies must develop cost-competitive alternatives to Chinese open-source models while maintaining technological superiority. As Meta’s acquisition of Manus also indicates, this and subsequent administrations must convince Chinese firms that it is better to do business with the United States than live under the Chinese Communist Party’s shadow. Tailoring outbound investment restrictions and export controls are vital levers to incentivize Chinese startups to relocate out of China.
The US government will also have important roles in using policy, especially for compute at home and abroad. The building of new data centers will enable the United States to prepare for the compute necessary for agentic AI. Computational power will be critical for implementing the Genesis Mission’s vision to supercharge new scientific discoveries at national laboratories.
Without using regulation as a cudgel, the US government must deftly use its authorities to shape agentic systems that are secure-by-design. The AI Action Plan rightly calls for secure-by-design AI for critical infrastructure and national security applications. The White House and interagency will need to ensure its AI agents retain security first and foremost in the design stage, not as an afterthought, for agentic deployment.
Export controls on chips will remain a sound policy in the age of agentic AI. Agents and agentic AI will be ravenous for compute. DeepSeek’s innovation to find efficiencies through inference, and reportedly by borrowing heavily from US models through distillation, could occur again with agentic AI. No such solution exists for now. Sustaining export controls and spreading the tech stack to friendly nations will remain smart policy.
More broadly, the United States must strengthen international cooperation with allies and partners to promote an AI ecosystem that can outcompete China’s hypercompetitive AI sector. This means working through existing alliances on AI governance and standards while creating new multilateral frameworks, such as Pax Silica, that can compete with Chinese initiatives in the Global South. American policymakers must recognize that this competition extends far beyond technology transfer or economic rivalry. Competition to lead in agentic AI is fundamentally about whose vision of progress and governance will shape human societies in the 21st century.
The agentic era has begun, and these systems will soon be woven into the foundation of how governments operate, factories produce, and militaries fight. The question is not whether agentic AI will transform state capacity and warfighting. That transformation is already underway. The question is which national vision will achieve global primacy, and whether the United States can adapt quickly enough to shape rather than simply respond to this new reality. America’s response to this challenge will determine not just technological leadership but the fundamental character of governance in the age of AI.
About the Author: Brandon Williams
Brandon Williams writes on US-China technology competition and strategic policy with articles in The Washington Quarterly (a summer 2025 article “Winning the Defining Contest: The US-China AI Race”), Time, Lawfare, and The Washington Post. Brandon’s research and writings specialize in Artificial Intelligence, quantum computing and sensing, high-performance computing, and cybersecurity. He was a senior fellow at the Center for Global Security Research at Lawrence Livermore National Laboratory. Brandon holds a PhD in History from UC Berkeley.
Disclaimer: The statements expressed in this article represent the author’s opinion and do not represent the Department of Defense, Lawrence Livermore National Laboratory, or the US Government.
Image: SuPatMaN/shutterstock
