Some thoughts triggered by the closed-door meeting of Sequoia AI Summit

In the current technological upsurge, artificial intelligence has become the core force leading the transformation. The closed-door meeting of the third Sequoia Capital AI Summit brought together 150 top AI founders from around the world. After 6 hours of in-depth discussions, many cutting-edge views and consensus emerged. This summit not only revealed the latest trends in the field of AI, but also provided us with a new perspective on how AI can reshape business models, technical architectures, and future economic forms.

Key points of the summit: Reshaping AI business logic

AI: From selling tools to selling benefits

The traditional software sales model, especially in the B2B field, has always been function-oriented. Enterprises purchase software to improve efficiency, automate processes, etc. The value of software is reflected in its diversity of functions and ease of use. However, Sequoia Capital partner Pat Grady clearly stated at the summit that the business logic of AI is undergoing a fundamental change, that is, from "selling tools" to "selling benefits."

This shift means that customers are no longer willing to pay only for the software's functionality, but are more concerned about whether the software can directly bring them measurable and tangible benefits. For example, in the past, companies purchased CRM software to better manage customer information and assist in sales processes; but now, the value of AI-driven CRM agents lies in their ability to directly help complete a certain number of customer conversions, thereby generating revenue for the company. This outcome-based pricing model requires that AI products must be able to deeply integrate into customers' business processes, accurately solve business pain points, and ultimately measure their value based on business outcomes.

This new logic places higher demands on the capabilities of AI products. Products must not only have powerful functions, but also have a deep understanding of customers' business needs and be able to deliver results stably in complex business environments. This forces AI companies to pay more attention to the actual application effects of products, rather than just the technical advancement.

The rise of operating system AI

The summit also emphasized the shift in the subject of AI, from "being called" to "active scheduling". OpenAI CEO Sam Altman presented a clear timetable: AI agents will start working in 2025, AI will discover new knowledge in 2026, and AI will enter the physical world to create value in 2027. This shows that AI is evolving towards the level of operating systems.

In the cloud era, Microsoft's Windows is the core of the operating system; in the mobile era, iOS and Android dominate. The operating system in the AI ​​era is no longer a traditional installed software, but a task scheduling system. It can remember users, understand user intentions, and take actions on behalf of users. For example, the "Agent Inbox" proposed by LangChain has become the entrance to trigger the collaborative work of many agents, replacing the traditional chat box and becoming a system bus-like existence.

This operating system-style AI will redefine the way users interact with software. Users no longer need to actively operate tools, but simply issue instructions to let AI agents complete a series of complex tasks. This not only improves efficiency, but also changes the distribution logic of software. Whoever can become the first recipient of user intent can master the system's scheduling rights and then control resource allocation. For AI companies, this means that they need to build smarter and more proactive systems to seize this entry-level opportunity.

The formation of intelligent economy

Sequoia partner Konstantine proposed the concept of "agent economy", that is, the future AI is not only a model to be called, but also an economic participant that can act, make decisions and cooperate. Agents have three major elements: persistent identity, action ability and trust collaboration. They can form a network system and exchange value with each other.

For example, Anthropic's Claude Code is no longer just a code generator. It can actively submit PRs, evaluate code quality, and coordinate other agents to complete tasks together. This makes the agent no longer a simple tool, but more like an engineering role with output responsibilities. As AI changes from a "answering tool" to an "autonomous agent", collaboration becomes a key capability, and new organizational structures are gradually taking shape. The role of humans has also changed from "controller" to "orchestrator", designing the responsibilities, interfaces and trust boundaries of agents, and building an economic network of human-agent symbiosis.

AI subverts traditional software B2B sales model

For a long time, the traditional software B2B sales model has often focused on promoting product functions and emphasizing the convenience that software tools can bring to enterprises. However, the rise of AI is completely rewriting this situation. Today, the value measurement standard of AI applications has shifted from simply providing functions to whether it can actually meet the real needs of customers and bring them quantifiable results .

Application players such as OpenAI, Ramp, and Sierra are no longer obsessed with competing for the "most advanced AI models", but are fully focused on "who can deliver results". This change means that when companies choose AI applications, they are no longer just concerned with the product's feature list, but are more concerned with whether it can play a role in the actual business process, complete the full process task delivery from the starting point to the end point, and bring clear value enhancement to the company, such as cost reduction, efficiency improvement, or accelerated business growth .

This change has brought new challenges to the R&D and sales of AI products. The R&D team needs to deeply understand the business pain points of customers in different industries, build end-to-end solutions based on customer needs . The sales team also needs to change its mindset from selling tools to demonstrating results , and prove to customers the actual benefits that AI applications can bring to them through actual cases and data .

The advantages of end-to-end iterative commercial AI models are highlighted

In the process of business development of enterprises, end-to-end iterative commercial AI models are showing unique adaptability. This type of model has many significant advantages: its cost barrier is low , and there is no need to invest huge amounts of money in model construction and maintenance; it is small in size and has higher flexibility in data storage and transmission; the training difficulty is relatively low , and enterprises do not need to rely on top technical teams and massive computing resources; it has low requirements for computing power scale and can run well under the existing hardware conditions of the enterprise; the performance requirements are also more in line with the actual business scenarios of the enterprise , and it does not pursue excessive high performance and lead to resource waste.

Take some small and medium-sized enterprises as an example. In the process of business expansion, they often face the dilemma of limited budget and insufficient technical capabilities. The end-to-end gradually iterative business AI model enables these enterprises to introduce AI technology at a lower cost and conduct personalized model training and optimization according to their own business characteristics. As the business continues to develop and change, the model can be continuously iterated and flexibly adapted to new needs without placing a heavy burden on the enterprise. This flexibility and adaptability make the end-to-end iterative business AI model an ideal choice for enterprises to apply AI technology in the rapid development of their business.

MCP Development: Opportunities and Challenges of AI Middle Platform

As a key development direction of the AI ​​middle platform, the Model Context Protocol (MCP) has demonstrated unique advantages in scheduling AI capabilities. It is like a "universal socket", providing a unified standard for the connection between AI models and external tools, and realizing the convenience of "one-time packaging, global availability". Different large models can easily call various external data sources and tools through the MCP protocol, greatly improving the development efficiency and intelligence level of AI applications. In the e-commerce industry, developers can quickly build intelligent e-commerce applications with the help of e-commerce transaction MCP, search MCP, etc. released by Baidu, and realize efficient integration of functions such as product search and transaction processing; in the field of content creation, through the MCP protocol, AI applications can easily call text generation, image generation and other tools to realize the creation of multimodal content.

The application of MCP is not without barriers. It places high demands on the user's overall planning ability when applying it . Users need to have strong control capabilities and be able to accurately plan how to integrate different AI models with external tools to achieve the best application effect. Compatibility between different models is also a key issue. Since there are many different types of AI models on the market, they may differ in architecture, data format, operating logic, etc. How to ensure that these models can work together under the MCP framework to avoid conflicts and incompatibilities is a major challenge that users need to face. Only by overcoming these challenges and giving full play to the advantages of MCP can we truly build an efficient and stable AI middle-end system .

Flywheel growth: AI-driven systems engineering

The flywheel growth concept provides a new perspective for the application of AI in enterprises. It emphasizes that this is not just the growth of the number of users in the traditional sense, but also a systematic project . The core lies in whether the entire process of driving results can be driven by AI , and whether each link in this process is equipped with an independent professional AI model.

Take an e-commerce company as an example. From user traffic introduction, product recommendation, transaction facilitation to after-sales service, each link can be optimized with the help of AI technology. Through accurate user portraits and personalized recommendation AI models, the conversion rate of product recommendations can be improved; using intelligent customer service AI models, the efficiency and quality of after-sales service can be improved. When the AI ​​models of these links work together to form an organic whole, closed-loop services for users in all scenarios can be achieved. Users have a better experience throughout the shopping process, which in turn increases purchase frequency and loyalty, bringing more business results to the company. In this model, the big model or Agent is no longer just an isolated technical application, but is closely integrated with the industry value, achieving a true "double landing", pushing the company into a virtuous cycle of flywheel growth.

Agent explosive growth and future trends

At present, Agents are showing an explosive growth trend. Various intelligent entities are springing up and are widely used in different fields. With the passage of time and the development of the market, Agents will inevitably experience a process of homogeneous development and mergers driven by market demand. At present, many Agents on the market have certain similarities in functions and application scenarios. As competition intensifies, those Agents that lack unique advantages and core competitiveness will gradually be eliminated by the market. Market demand will become an important force to promote the development and integration of Agents. The needs of enterprises and users for Agent functions, performance, security , etc. will prompt high-quality Agents to continue to evolve and integrate. Some Agents that focus on specific industries may form more powerful industry solutions through integration and cooperation to meet the market's demand for one-stop services . This process will make the Agent market more mature and concentrated, and eventually form a small number of Agent ecosystems with extensive influence and strong competitiveness.

Towards the era of physical AI: the evolution of intelligent ecological networks

The future intelligent ecological network, namely the AI ​​network, is showing a clear trend from generative data to physical real-time data , which will lead us into the era of physical AI. In this new era, the interaction between various intelligent agents will become richer and more complex, including not only online virtual agents, but also offline robots, self-driving cars, drones and other physical intelligent devices. It is important to realize the interaction between intelligent agents, including the ability of global perception, cognition and real-time reasoning and decision-making .

Through Google Earth and Street View projects, Google has transformed geographic elements and buildings in the physical world into three-dimensional digital models, providing important data support for the application of AI in urban planning, autonomous driving and other fields; Nvidia and SoftBank have jointly promoted AI-RAN (radio access network) solutions to help Japan build a powerful AI infrastructure and improve the connection efficiency between the physical and digital worlds; Tesla’s "world model" concept is used to build and understand a high-precision simulation system of the real world, enabling self-driving cars to better perceive and make decisions; SpaceX's Starlink provides high-speed Internet access through thousands of low-orbit satellites covering the world, integrating physical space into the Internet digital network, and providing guarantees for data transmission of various intelligent entities.

In the era of physical AI, through high-precision sensors and IoT technology, intelligent agents can perceive various information of the physical world in real time , and use high-speed communication networks to achieve rapid data transmission, and then use powerful cloud computing and edge computing capabilities for data processing and intelligent analysis, thereby achieving efficient collaboration and interaction between various intelligent agents. Robots and automation equipment in smart factories can be intelligently scheduled and collaborated based on real-time production data; self-driving cars can communicate with transportation infrastructure and other vehicles in real time to optimize driving routes and improve traffic efficiency; drones can perform precise tasks in complex physical environments based on real-time perceived data. The construction of this intelligent ecological network will completely change the way we interact with the physical world and bring unprecedented changes to social development.