From "Pride of Domestic AI" to lightning layoffs and relocating to Singapore, a mere 130-day frenzy has made Manus the first star startup to "reach shore" in the Year of the Agent—securing a $75 million Series B funding round from Benchmark.
Almost simultaneously, the blade from Silicon Valley has already been unsheathed.
On July 18th, OpenAI launched ChatGPT Agent, sounding the horn for major model vendors to enter the AI Agent arena: with a user's simple command like "plan a wedding itinerary," the Agent automatically calls a browser to compare hotel prices, generates a travel guide with maps, recommends matching outfits, delivering in 25 minutes a task that traditionally took hours. It deeply integrates Operator's graphical operations with Deep Research's long-chain reasoning, completing the "think-execute" loop within a virtual computer, turning large language models into "digital labor" capable of searching, clicking buttons, filling forms, and writing code.
This seems to signal that the AI Agent startup arena is entering its second half, capital is entering, and model vendors are reaping the harvest. Successful startups that have broken through are beginning to be priced for acquisition, while companies that failed to attract attention are swept into the corner.
And all of this was actually foreseen early in the year.
The "Cultivation Game" of the Agent Year: Giants Cast Nets, Startups Fall Into the Trap
Those who have been following the AI Agent arena closely might still remember that earlier this year, Sam Altman and Sundar Pichai successively proclaimed the "Year of the Agent." This was a meticulously designed hunt—giants create concepts, startups sprint forward, capital enters to rapidly inflate valuations.
Now this game has finally reached its final step: giants harvest the market, capital cashes out profits.
The Giants' Open Scheme
Since ChatGPT ignited the AI arena in early 2024, this playbook has been repeating.
Whether it's open-sourcing models to create the "AI era," the launch of GPTs spurring the rise of prompt engineering and digital human arenas, or Sora's release opening up a brand-new text-to-video track. All such moves are standard operations by giants using their industry influence to gradually shape the AI ecosystem—just like Chinese giants in the mobile internet era, constantly creating new business models to lure startups into the field for trial and error and ecosystem shaping, finally harvesting the market through advantages in capital and user scale, feeding their own businesses, and gaining competitive edge against other giants.
In the AI era, this logic still applies.
The Life-or-Death Sprint for Startups
For most AI Agent startups, the so-called "Year of the Agent" is actually a pseudo-trend driven by external narratives.
Major model vendors tout "general agents will change everything" while releasing model capabilities via APIs. But for entrepreneurs, this means having to piece together a "decent-looking Agent" without real demand validation, with chaotic product structures, and under high model costs.
This forced startups' pace into an extreme state from the very beginning: either explode in popularity short-term, or exit rapidly. Manus, Kto, AI Pin, Rabbit R1—each project seemed to have been hyped, and each exposed core problems very quickly—no general Agent product has truly crossed the line from "Demo to daily use."
Often, so-called "user retention" and "high-frequency use cases" aren't real but data illusions built through community operations and early incentive mechanisms. The current AI user ecosystem has zero loyalty; users go with whoever is cheaper, whose output is more accurate. Once growth slows, or a trendier competitor appears, or models iterate, user behavior collapses immediately.
Capital Enters the Game
Capital's logic differs from product logic. They aren't betting on whether the market can actually produce a viable Agent product, but under the drive of traffic thinking, on who can become "the project that first steps onto the stage in the next wave of AI," thereby cashing out valuations before the giants harvest.
Manus is the lucky bet placed during this window: its product isn't stronger than other Agents in the industry, and even has significant flaws in functionality and usability. But it caught a rare combination punch window—patriotic narratives and giants' Agents not yet launched. Domestically, timing the DeepSeek moment, leveraging market frenzy for China's self-developed AI, igniting product traffic through scarcity marketing and patriotic narratives from community自媒体, and globally gaining attention by being among the first to launch a product following the "Year of the Agent" slogan.
From the start, Manus's product packaging, launch format, and overall tone had a strong Silicon Valley flavor—this wasn't a product for users, but promotion for US investors.
And what capital cares about is the exit path, not the product path. For them, Manus, which captured the first-mover position in Agent and garnered global traffic, was the prime investment target. Its eventual bet by Benchmark and choice to relocate to Singapore is a story of "successful exit." Even if the project later fails, it can be packaged as "the cost of market education."
Capital clearly understands that the essence of the general Agent arena is an extension of the battlefield of giants' model capabilities. This track will inevitably be covered by giants; their real goal is to dig the first bucket of gold and cash out profits as soon as possible.
The Real Competition in AI is Competition of Technical Routes: Visual Sandbox, Data Pipelines, and Ecosystem Integration
In the practical implementation of Agents, three major technical routes have gradually taken shape: OpenAI bets on the "visual sandbox," Anthropic focuses on "data semantic pipelines," and Google leverages ecosystem integration to build application closed loops. What they represent is not just differences in interaction forms, but different understandings of the path to implementing general intelligence.
OpenAI's Visual Interaction: The General Execution Entity Evolving in a Sandbox
- General Operation Model Based on Web DOM
OpenAI builds visual Agents through the Operator module, simulating user actions like clicking, inputting, and scrolling in a browser environment. Its advantage lies in not requiring custom adaptation, able to widely adapt to consumer-level web tasks like price comparison, ticket purchasing, and itinerary organization.
- Capability Boundaries Limited by the Sandbox Environment
For safety and controllability, this mode is strictly confined to a virtual environment. While avoiding system-level permission issues, it also means difficulty accessing local files, corporate intranets, and desktop-level tools, limiting its application depth in professional scenarios. Simultaneously, high computational cost and task latency remain unresolved issues.
Anthropic's Data Route: Long-Context Understanding and Vertical Application Deep Dive
- Complex Document Analysis Driven by Context Window
Claude, from 3.7 to 4.0, leveraging a context window as high as 200K and stable semantic retention, demonstrates advantages in handling structured and unstructured data in vertical industries like programming, healthcare, law, and pharmaceuticals. For example, in pharmaceutical patent analysis, it achieves high-accuracy document extraction and summarization.
- Emphasis on Both Compliance and Controllability
Anthropic builds a safety baseline for high-compliance industries through "Constitutional AI" and preset rules, earning a good reputation among financial and pharmaceutical clients. This path emphasizes "precision and safety first," currently a relatively stable force in enterprise application implementation.
- MCP Leverages External Ecosystem The landing point of MCP is to construct a new paradigm of "model as the main controller, software as the called unit." For Anthropic, this is a paradigm-shifting attempt no less significant than the plugin system. Claude is no longer just an add-on smart assistant for some SaaS, but attempts to become the main operating system for enterprise software stacks. This thinking naturally aligns with its focus on data compliance, task precision, and industry deep dive, forming a closed loop.
Google's Ecosystem Integration Strategy: Toolchain Micro-Integration and Cost Balancing
- Native Embedded Agent Experience
Compared to simulating user behavior, Google chooses to directly embed Agent functionality into tools like Workspace, forming "micro-Agent" chains. For example, extracting email content in Gmail to generate schedules, or linking Sheets with BigQuery for data analysis. This approach improves stability and efficiency, suitable for enterprise users' daily workflows.
- Leveraging Cloud Ecosystem for Cost Pressure
Google, with its massive cloud infrastructure, can offer Agent access capabilities at highly competitive prices (e.g., Gemini Flash's low-cost token pricing), creating significant market pressure on small and medium-sized entrepreneurs.
Behind the Route Divergence Lies Different Perceptions of the "Agent" Essence
- OpenAI views Agent as an extension of the "general execution model," focusing on how the model actively operates interfaces to complete tasks, adapting to current interaction methods by making the model mimic human behavior;
- Anthropic believes the core is data and semantic understanding, emphasizing stable capabilities in structured and compliance scenarios, attempting to connect data interactions between applications via the MCP open protocol, making the model the interaction hub at the data level;
- Google places more value on the practical integration of "AI + toolchain," first making Agent part of familiar tools, then gradually increasing intelligence, driving its own product ecosystem with models.
The three directions are not ranked, reflecting the natural extension of their respective technical foundations and product philosophies. In the short term, this technical heterogeneity will continue to coexist, but ultimately, whoever first bridges the closed loop between "generalization capability" and "application necessity" will truly determine the Agent landscape.
And Manus follows OpenAI's visual route, using KV-cache and file system compression to reduce costs and improve hit rates, essentially still engineering optimization wrapped around AI. But isn't this exactly what OpenAI can do and is doing?
The Fatal Weaknesses of General Agent Startups and Windsurf's Crazy 72 Hours
The Manus case exposes the fatal weaknesses of general Agent startups:
- Hollow Technology
Manus founder Ji Yichao admitted in a retrospective blog: "Choosing context engineering over self-developed large models," motivated by his first failed startup—his self-developed model was淘汰 by the released GPT-3. Manus actually uses KV cache to optimize costs, simulating "infinite memory" via file systems. Most general Agent startups use similar engineering tricks, reorganizing data to improve output efficiency. While this lightweight engineering approach can reduce costs, it heavily relies on underlying large models, unable to build its own moat, remaining vulnerable against giants.
- Cost Imbalance
Single-task costs reach $2 (5 times OpenAI's), computational consumption exceeds industry average by 500%, and this cost structure directly translates to degraded user experience. Manus adopts a credit-based consumption model: Pro membership costs $199/month, ideally allowing only 15 to 20 uses per day. Simple tasks don't need it, complex tasks are slow; with price叠加 response speed, the cost of using a general Agent is extremely high compared to directly using models.
- No Moat
General Agent services heavily depend on underlying large models (like Claude, GPT) for inference capabilities and update pace; computing power and capabilities are entirely entrusted to model API providers, unable to optimize independently脱离 their evolution path. This dependency决定了 Agent cannot form differentiated advantages through底层 technology accumulation. On the other hand, it also cannot accumulate high-value, transferable user data assets. User task records, semantic chains, even behavioral paths on an Agent are not fundamentally different from using any general large model platform; data cannot form专属 understanding models or preference profiles, cannot train proprietary systems, making it difficult to build compound-interest data assets.
In stark contrast to general Agents like Manus is the逆势 growth of a batch of vertical domain Agents:
- Hippocratic AI builds closed loops in medical dialogue and screening. According to official information, this year they increased colorectal cancer FIT screening participation rates to 2.6 times that of English-speaking patients through multi-channel interventions (including Spanish-speaking patients). Their clinical辅助 accuracy improvement is also significant, rising from around 80% early on to 99.38% in the latest version, with severe misdiagnosis rate dropping to 0. All this benefits from continuous accumulation of structured case data and feedback execution processes.
- PathChat for pathological imaging and its升级 version PathChat+ published papers in top journals like Nature, showing performance of "87% diagnostic accuracy" in image Q&A tasks. This is completely different from general visual Agents; its training and evaluation focus on a few high-value scenarios, possessing industry-level reference value.
- Genspark is praised by users in engineering practice for high execution efficiency and excellent cost-performance. Reddit users noted Genspark can complete more tasks than Manus in the same scenarios, with overall lower costs. Comparative estimates suggest Genspark user experience is "insanely fast," far surpassing the higher-priced Manus.
- Claude Code focuses on programming scenarios, supporting context持续 tracking, code debugging and explanation, highly praised by the developer community.
- Salesforce Agentforce embeds into the CRM ecosystem, naturally唤醒 AI capabilities in key workflows, seamlessly integrating into tools salespeople use daily.
Simultaneously, unsettling news emerged from Silicon Valley: Windsurf was rapidly dismembered within 72 hours.
OpenAI曾 offered $3 billion for acquisition, blocked by Microsoft leading to协议破裂; Google acquihired CEO Varun Mohan and other executives for $2.4 billion,同时 obtaining non-exclusive technology licensing; hours later, Cognition (team behind Devin) acquired the remaining team, IP, and brand, completing Windsurf's strategic重组 within 72 hours.
This incident reveals a dangerous signal for the industry: acquihire has become the mainstream clearing path in the Agent领域, the fusion of capital and technology is瓦解 startup trust mechanisms, CEOs can随时 jump ship due to "technology transferability," leaving VCs and employees in a bleak position. Casting a shadow over the general Agent startup arena.
For some time to come, general Agent startups may enter a more激烈 turbulent淘汰赛.
The AI Interaction Revolution: Survival Space Left for Startups
The cases of general Agents and Windsurf's拆分 illuminate the illusion of the general Agent model track, and also light up this track's true watershed—structure and closed loop.
After this fever, Agent startups will走向分化. Some will remain obsessed with "stronger, smarter Agents," trying to replicate the human brain's general path, but they will持续 be反噬 by computing power, costs, and error rates; others will return to rationality,重新 seeking answers from the original concept of Agent.
As early as 1995, in "Intelligent Agents: Theory and Practice," Nicholas R. Jennings and Michael Wooldridge systematically established the concept of AI Agent: Intelligent agents should possess the ability to perceive the environment, act autonomously, pursue goals, and emphasize that agent architecture must support this decision-execution system closed-loop mode.
I believe these four attributes correspond to the following engineering capabilities:
- Autonomy: Can operate independently,具备 multi-source heterogeneous data fusion and解析 capability.
- Reactivity: Perceives and responds to environmental changes, can respond to specific situations, remains silent otherwise.
- Pro-activeness: Takes proactive actions based on goals, has跨系统 perception and task planning capabilities.
- Social Ability: Capable of collaboration and communication, synchronous and asynchronous coordination with multiple models and tools.
From vertical Agent application cases, we can speculate that future AI Agents are not a "smarter" person, but a "more controllable" structure. They must possess the following capabilities: Structured Input/Output: Define task boundaries, control Agent capabilities;
Structured Behavioral Paths: Embed into workflows, provide stable process feedback;
Structured Feedback Data: Continuously iterate training,走出 closed-loop paths.
Past internet interaction modes will be broken. The logic of organizing information around an individual or organization as a支点, or information distribution centered on fixed processes or algorithms, will transform into a completely decentralized mode led by AI.
The second half of Agent startups is a competition of "structural capability." It's not about who understands large models deeper, but about who can use structural tasks to打通 feedback closed loops, compressing「model capabilities」into a system that is落地, controllable, and capable of growth.
Agent startups must build private data moats. Only by deeply integrating into business, accumulating独有的 data context and behavioral feedback, can they train Agents that others cannot imitate. This data isn't generated by scraping, but accumulated over time by embedding into processes, exchanging trust, efficiency, and feedback repeatedly through structural capabilities,逐步 accumulating organizational assets.
The final outcome of this training is like the difference between a chef's and a firefighter's reaction to fire: the chef会 flip the wok, while the firefighter会 pull the fire hose.
If general Agent companies hope that through model generalization capabilities, they will one day become levers撬动 social production, it's likely like fishing for the moon in water. Even AI models themselves cannot exhaustively list all roles in the world; what资格 do Agents built on top of models have to accomplish this grand ambition?
Creating真正的 Agents is必然 a long and thorny road. But as the success of Hippocratic AI启示, starting from a vertical细分 niche, redefining the vitality of "small and beautiful" within the cracks of giant ecosystems,才有机会诞生 AI Agents that truly solve problems.