In 2026, the office sector undoubtedly stands as the most vibrant battleground for AI. As AI models become increasingly versatile, competition among tech giants intensifies, morphing into a comprehensive "software and hardware" war across all fronts. Personal agents are surging in popularity, while AI hardware innovations are emerging endlessly, reshaping the digital office landscape.

From Anthropic's early-year launch of Claude Cowork, setting the pace for personal agents, to Meta, Alibaba, and ByteDance's frequent rollouts of smart hardware by year-end, leading companies are reconstructing digital office workflows with software agents while seizing physical world entry points with hardware terminals. The speed of product iteration even outpaces the refresh rates of workplace tasks.

As AI office tools officially transition from "conversational tools" to "task execution," the industry's core debate becomes clearer: which track—software-based personal agents or diversified AI hardware—holds greater long-term potential?

Making AI More 'Human-Like'

Over the past year, AI agents have collectively stepped out of chatboxes and into real-world business scenarios, with their numbers rapidly expanding. IDC predicts that the number of active agents will skyrocket from approximately 28.6 million in 2025 to 2.216 billion by 2030.

In the AI office sector, user interest has been ignited following last year's Manus and this year's early OpenClaw market education. The focus of competition among tech giants has shifted from vying for entry points to actual productivity, culminating in the recent emergence of "personal agents" capable of handling long-term tasks.

Anthropic, an early adopter of office agents, launched Claude Cowork in January 2026 for knowledge workers, enabling direct access to local files, coordinating sub-agents for parallel task processing, and autonomously executing complete workflows. In June, OpenAI integrated Codex into ChatGPT, introducing Agent Plugins and Sites to bundle agent capabilities for code, documents, and data processing into users' existing habits.

September marked a boom period for the global personal agent sector. Meta fired the first shot with Muse, which instantly became the year's phenomenal product. Within just 13 days, Meta topped the U.S. Apple App Store's free app chart, driving an 11.43% surge in Meta's stock price on the same day.

Muse broke away from traditional conversational AI's question-and-answer format, continuously remembering user information and understanding long-term intentions. At its launch event, Muse demonstrated a daily scenario: a user saves a recipe on Instagram, and Muse organizes a menu and shopping list based on family members' dietary restrictions, then proceeds with related tasks. Its popularity in North America validated the mass market acceptance of personal agent products among consumer audiences.

Manus followed with its personal agent application, Cue. Cue equipped each agent with a dedicated email, phone number, wallet, and computing terminal, granting it a complete digital identity. It could autonomously send and receive messages, make budget-approved payments, and independently complete tasks from start to finish. It could even answer calls on behalf of users and generate automatic call summaries, further expanding agents' responsibilities and application scenarios.

At OpenAI's "largest-ever" developer conference in late September, Sam Altman dedicated the finale to Dots. OpenAI described Dots as "intelligent agents capable of continuous responsibility," emphasizing "always-on" functionality and proposing a future direction where multiple Dots form teams for collaborative work.

The primary market also witnessed a surge in activity. That same month, Instinct, a startup with just 14 employees, secured $1 billion in financing, valuing it at over $10 billion. Instinct interacts via SMS and phone calls, currently offering invitation-only product experiences.

Overseas vendors collectively view personal agents as the next-generation traffic gateway after search engines and app stores, while domestic tech giants have also fully rolled out their strategies.

In late September, ByteDance was reported to accelerate its personal agent project, codenamed Spell, led by the Doubao mobile assistant team. Tencent Yuanbao already offers an AI-native desktop intelligent agent workstation, WorkBuddy, along with the "cloud-based personal assistant" LightVela and a WeChat AI assistant, "Xiaowei," currently in testing. Recent reports suggest an imminent launch of personal agent products.

Entering October, Ant Group's Lingguang initiated a new version beta test, focusing on enhancing personal task execution capabilities. Alibaba's Qianwen had previously announced accelerated development of Personal Agents, officially launching the personal assistant "Qianwen Go!" on the 13th and initiating a limited-time invitation test on the Qianwen App.

Among startups, Today AI, founded by former Doubao executive Qi Junyuan, opened for downloads on September 24, emphasizing long-term memory and background parallel execution, dubbed the "Chinese version of Muse."

With global tech giants and startups fully entering the fray, AI agents have officially transitioned from "conversational tools" to "task execution," shifting focus from "tasks" to "people."

Integrating AI Hardware into Daily Life

While the evolution of large models and agents has prompted even top scientists to issue warnings, in the real world, most personal terminals remain unchanged. If personal agents address task execution in the digital world, AI hardware serves as a crucial gateway between the physical and digital realms.

New forms of AI hardware abound, with AI glasses, earbuds, recording cards, necklaces, and rings constantly redefining perceptions of intelligence under the rallying cry of "reinventing all hardware with AI."

The core functionality of hardware products lies in serving as AI's "ears" and "eyes" for understanding the real world, providing agents with more complete user context and addressing their "perceptual shortcomings," thereby truly integrating technology into daily human life.

Overseas tech giants pursue a dual strategy of advancing both underlying computing power and terminal forms for AI hardware.

Intel launched the Core Ultra Series 3 processor at CES 2026, partnering with Microsoft to expand Copilot+ PCs to a broader consumer market. Microsoft collaborated with NVIDIA to release Windows laptops equipped with RTX Spark chips, capable of running high-performance AI models locally. Amazon introduced its first voice assistant-named Alexa series tablets, embedding AI assistants into hardware systems. Apple released HomeView, a smart home hub focusing on end-side AI experiences for home and light office scenarios.

Domestic tech giants, meanwhile, prioritize hardware implementation in office scenarios, collecting offline office data to address software agents' contextual shortcomings.

Take the fiercely competitive lightweight recording sector as an example. Plaud reignited the recording device war with its card-style AI summary product, Plaud Note. In September, Plaud announced a deep collaboration with Tencent WorkBuddy, interoperating their official MCP interfaces to seamlessly integrate into workflows, transforming single conversations from "a summary" into reusable work assets. A co-branded hardware product, Plaud Note Pro, is also planned.

Tech giants quickly followed suit. Alibaba DingTalk released its first AI recording hardware, A1, back in 2025. This September, Qianwen Office launched the upgraded QwenNote A2, emphasizing "burn-after-transcription" privacy features. iFLYTEK, a veteran in the voice sector, also entered the market with its AI recording card in September, leveraging its native voice technology advantages. ByteDance's Feishu partnered with Anker Innovations to launch the "AI Recording Bean" early in the year, weighing just 10 grams and adapting to various scenarios like daily interviews and meetings.

This year's standout in AI hardware is undoubtedly AI glasses, a focal point in the tech giants' arms race. Omdia data shows global AI glasses shipments climbing from 380,000 units in 2023 and 2.06 million units in 2024 to 8.7 million units in 2025, with an estimated breakthrough of 15 million units in 2026, making it one of the fastest-growing consumer electronics categories.

In the vertical sector, vendors like Rokid, Xreal, and Rayneo Innovation have sequentially ramped up shipments, each claiming top spot in shipment volume announcements. The collective entry of internet giants is now pushing this sector from geek circles to the mainstream market.

Alibaba stands out as the most aggressive domestic player. In March 2026, Qianwen officially launched its first AI glasses, G1, positioned as a "portable intelligent super assistant," fully integrated with the Qianwen large model and office ecosystem. In October, it released the next-generation Qianwen AI glasses, N1 and N1 Pro, with the N1 Pro supporting vital sign monitoring like heart rate and blood oxygen, further expanding perceptual dimensions. Shu Junliang, VP of Qianwen Office, stated, "Hardware serves as an entry point for incremental context collection, bridging the digital and physical worlds."

Xiaomi and Baidu have also joined the fray, with Xiaomi's AI glasses focusing on first-person filming and lifestyle recording, and Baidu emphasizing translation and portable AI assistance, both without screen displays. According to Tianyancha data, Tencent, through its WorkBuddy open platform, collaborated with Rokid to release the Leqi Rokid×WorkBuddy co-branded AI glasses.

This is just the tip of the iceberg in the hardware landscape. To seize the AI office sector, mine high-value data from offline scenarios, and complement physical world data inaccessible to pure large models, tech giants' hardware strategies now span all forms of audio-visual wearables, with the battle for entry points fully underway.

Short-Term Software Boom, Long-Term Hardware Dominance

Personal agents and AI hardware represent two stages of AI office evolution, but their growth potential, barrier height, and realization timelines differ significantly in terms of imagination hierarchy.

The short-term imagination for personal agents is undeniably strong, with the steepest growth trajectory.

Software's near-zero marginal cost and high distribution efficiency ensure personal agents scale far faster than hardware. From a user experience perspective, agent capabilities directly tie to underlying large models, with product evolution measured in weeks or even days.

In terms of commercialization, personal agents inherently possess a paid foundation. Instinct founder Noah Shinn revealed its annualized transaction volume has surpassed $1 billion, translating to annual revenues of $30-50 million based on a 3-5% commission rate.

Personal agents hold promise to become genuine "work-life companions." However, their development bottlenecks are clear, facing challenges in commercialization, social compliance, and trust.

First, greater capability demands stronger judgment. Balancing answering, thinking, and acting determines whether personal agents save time or create new work.

Moreover, mainstream domestic apps bundle social, payment, shopping, and travel functions into super apps, offering highly convenient operations within their independent ecosystems. WeChat, Taobao, Douyin, and Meituan restrict data interoperability and permission sharing, directly capping agents' capability ceilings.

This means personal agents cannot achieve true cross-platform, full-scenario execution, remaining confined to "auxiliary tool" status in terms of authority boundaries and autonomous execution depth, far from "independent task handling."

As for privacy concerns, Muse faced a setback in September when nearly 190,000 user conversation records were leaked, forcing it to urgently tighten data permissions and add full-process authorization prompts.

Compared to personal agents, AI hardware collects offline voice, scene, and vital sign data—exclusive resources unattainable by pure software agents—offering higher long-term value barriers.

However, AI hardware faces high R&D, supply chain, and channel costs, long iteration cycles, high "dust-gathering" rates, severe single-category homogenization, and limited ceilings. It must overcome user habit cultivation and supply chain maturity cycles to prove sufficient rigid value.

Models address "how context is understood," agents tackle "how context is executed," and hardware solves "where context originates." Together, they form an integrated intelligent ecosystem.

The AI-driven industrial wave has already reverberated across all sectors, leaving no one untouched.

From question-and-answer conversational tools to autonomous execution entities, from purely digital online scenarios to full coverage of virtual-real integrated office environments, this transformation is not only reconstructing the underlying logic of productivity but also redefining the collaborative boundaries between humans and intelligence.