In the past, the wave of "industry + internet" swept across various sectors, but the vast majority of attempts ultimately faded away, with only a few companies like Pinduoduo and SHEIN, which adopted the "internet+" model, becoming industry giants. Looking back on this journey, it's not hard to see that simply adding new technologies on top of old models is essentially "putting an engine on a horse-drawn carriage"—ultimately unable to adapt to the rapidly changing market.
The Lesson of the "Industry + Internet" Disillusionment: Why Did the Vast Majority Fail?
Looking back at the "industry + internet" boom around 2015, almost all traditional enterprises tried to transform themselves with internet technology. From apparel, e-commerce, healthcare, real estate to manufacturing, they invested heavily in building apps, creating communities, and conducting big data analysis.
However, in the end, the failures far outnumbered the successes.
For example:
- Redbaby Maternal and Child E-commerce: Attempted to rapidly expand its online business beyond physical stores but was forced to sell due to a lack of e-commerce DNA, logistics system, and product strength.
- Metersbonwe's 'Bangou.com': As a traditional apparel brand trying to launch an e-commerce platform, it deviated from user experience logic, invested heavily, and ultimately shut down.
- The failed e-commerce transformation of the 'Yishion' brand: Even after investing in building apps and CRM systems, the brand did not truly digitize; it was merely superficially online, eventually becoming marginalized.
- SF Express' 'Heike': An innovative attempt to integrate e-commerce experiences into offline convenience stores, costing a huge amount but with a business model unverified by the market, eventually leading to significant contraction.
- LeEco's Ecosystem: Built an integrated internet ecosystem of content, hardware, and cars, but collapsed due to broken capital chains and blind expansion.
The common characteristic of these cases is: using the internet to patch old processes rather than reconstructing business logic.
They appear "digitalized," but in reality, they only changed the interface; the underlying operating system remained traditional, hierarchical, and linear. This is similar to the current "AI+" behavior of many enterprises, trying to use AI to "accelerate" old tools rather than redesigning the tools and workflows themselves.
Now, "AI+" has become the new trend, yet most enterprises still fall into the misconception of "treating AI as a tool plugin." This mindset limits AI's potential and causes us to miss real opportunities for transformation. The key to breaking through this dilemma lies in how to design the relationship between AI and business—
AI should not be a "plugin in task nodes" but rather a "core node in workflows."
"Internet+": The Logic of Structural Reshapers' Comeback
Although most "industry + internet" projects ultimately failed, a few enterprises truly achieved exponential growth by "reconstructing business logic." Their common point is: they did not simply add an "internet shell" to old models but rebuilt the entire value chain from first principles.
Pinduoduo: Reconstructing the Distribution Logic of "People, Goods, and Places"
Pinduoduo did not simply make traditional e-commerce mobile; instead, it leveraged the WeChat social system to reconstruct the logic of "matching people with goods," transforming the originally centralized search logic into a decentralized traffic distribution method of "group buying + algorithm recommendations." This not only reduced customer acquisition costs but also opened up new growth points in the lower-tier market.
SHEIN: Integrating Supply Chain Data to Achieve Flexible and Fast Response
The core of SHEIN's success lies in its highly digitalized supply chain management system. It did not simply put clothes on an app to sell; instead, through a closed-loop process of front-end user data → back-end factory sampling → mid-platform intelligent production scheduling, it achieved "user-driven supply." This made it faster than Zara, with richer SKUs and lower costs.
ByteDance: Replacing Human Editors with Recommendation Algorithms
ByteDance did not "create a smarter portal"; instead, it directly eliminated the content editorial system of "portals" and changed the traffic distribution logic to "algorithm recommendations." This relies on a powerful user profiling system and real-time feedback mechanisms, completely transforming the information supply logic of the content industry.
The Limitations of Conversational AI: Only Understanding Two-Dimensional Information
Currently, the most popular conversational AI, such as ChatGPT, processes "two-dimensional information"—that is, input and output at the textual level. It excels at answering clear questions but lacks the ability to understand three-dimensional information in the real world.
Real-world information is multidimensional:
- Time Dimension: The sequence, frequency, and rhythm of events
- Space Dimension: Information scattered across different systems, files, and applications
- Structural Dimension: The hierarchical relationships and organizational methods within information
Conversational AI can only receive text information at single points and cannot automatically capture when, where, with whom, and what events occurred for the user. This limits the interaction to users having to accurately express their needs through text, which, for most people, is an advanced skill requiring long-term training.
The Key to Future AI: Integrated Understanding of Multi-Source Heterogeneous Data
Truly valuable AI should become an intelligent aggregator and interpreter of multi-source heterogeneous data. It must not only understand text but also comprehend users' behavioral trajectories and business environments:
- Automatically capture users' daily work data in OA systems, generating weekly reports without requiring active user input
- Link project progress, meeting minutes, and email exchanges to achieve real-time information integration
- Intelligently infer business changes based on historical and current data, dynamically adjusting workflows
This is true "three-dimensional information understanding" and the guarantee of AI-driven business agility.
AI Should Actively Perceive "Spatiotemporal Interaction Semantics"
In other words, AI should not passively wait for instructions but actively perceive:
| Dimension | Content | Example |
|---|---|---|
| Time | When did the user do what | Meeting in the morning, revising the plan in the afternoon |
| Location | In which system or environment the operation occurred | Editing documents in Notion, discussing in Slack |
| Object | Relevant personnel or tools | Syncing with designers, sending emails to clients |
| Event | Specific actions | Creating requirements, initiating reviews, submitting code |
Only based on these "spatiotemporal interaction semantics" can AI truly understand the business, automatically design workflows, and collaborate with users to complete complex tasks.
The Leap from "Prompt AI" to "Contextual AI"
The current mainstream AI belongs to the "Prompt AI" stage, relying on users' explicit questions and lacking contextual awareness of users' backgrounds and behaviors. The future trend is "Contextual AI"—
AI automatically perceives context based on user identity, behavior, and environment, generating intelligent suggestions and becoming the user's "second brain."
For example:
- Automatically generating weekly reports, summarizing meeting minutes, reminding of pending tasks
- Analyzing users' work trajectories, proactively recommending optimization solutions
- Dynamically adjusting business processes to quickly respond to market changes
Such AI not only reduces users' learning costs but also greatly improves work efficiency and business agility.
Google I/O 2025: A Frenzied Self-Revolution
At the Google I/O 2025 conference, Google comprehensively demonstrated its determination to transform into "AI-driven products." From the Gemini family to the deep integration of Workspace AI, Google officially shifted AI from a "plugin tool" to the core position of a "task engine."
Among them, Google highlighted several key transformations:
- Project Astra Real-Time Perception System: This is an AI assistant combining multi-source sensors like cameras, microphones, and positioning. It no longer waits for user input but actively perceives the user's physical environment, context, and task status in real time, providing voice/visual feedback proactively.
- AI Agents for Workflow: Duet AI in Google Workspace was upgraded to a "workflow coordinator" capable of scheduling resources across Gmail, Docs, Calendar, and Sheets. It can proactively help users generate meeting agendas, fill project plans, and send follow-up emails, rather than passively responding to requests.
- Veo Multimodal Video Understanding: Not just generating images, Google's Veo model also entered the multimodal field, understanding the context of images, text, and videos, promoting automation in content creation and enhancement of interactive scenarios.
These updates mark Google's move to position AI as a foundational component of the operating system, rather than an add-on at the application layer. This paradigm shift is precisely what this article emphasizes: the true form of "AI-driven workflows"—where AI actively understands scenarios, schedules processes, links data, and generates tasks.
Behind this lies an organizational revolution within Google.
As a company long dependent on search advertising revenue, Google's transformation this time directly confronts its "cash cow"—the traditional search business. If previous Bard or Gemini were more responses to ChatGPT, then the 2025 I/O conference truly showcased Google's ambition to "reconstruct everything with AI."
Although new search forms (such as multi-turn conversations, contextual Q&A, task delegation, etc.) might erode the original advertising entry points and click paths, Google still chose to advance comprehensively, demonstrating its strategic investment in the future "AI-driven ecosystem."
As Sundar Pichai said at the conference: "We're no longer building products. We're building agents."
This is not just an evolution in product form but a reconstruction of business models and company logic. Google's actions show that even if it means disrupting its old core, it will become the new foundation of the AI era.
This self-revolution is a landmark event in the trend of "AI-driven workflows."
The Conflict Between AI-Driven Operations and Bureaucracy: The Game Between Flattening and Hierarchy
However, it is worth noting that AI-driven business updates require rapid response and reduced management, which directly challenges hierarchical organizational structures.
Traditional organizations rely on clear job responsibilities, multi-level approval processes, and distinct divisions of labor to maintain stability. This structural design is precisely to "resist uncertainty" and "reduce communication costs."
The advantages of AI systems, however, lie in:
- Real-time perception, automatic judgment
- Rapid feedback, skipping intermediate layers
- Flat collaboration, cross-departmental linkage
In AI-driven work scenarios, everyone is an executor, and the system automatically distributes tasks and tracks progress based on data. At this point, the so-called clear responsibilities and strict hierarchy in traditional organizations become a joke.
This may be the underlying reason why many enterprises today cannot truly embrace AI: they are unwilling to give up "management authority."
Letting AI access all core data equates to stripping managers of their "data privileges." Once data permissions are centralized in AI, managers lose information barriers, and their legitimacy in exercising management authority within the organization will quickly weaken.
This structural tension is not just a technical issue but also a divergence in management philosophy:
- Traditional hierarchy emphasizes "control"
- AI-driven operations emphasize "response" and "autonomy"
If future organizations cannot adapt to this change and still try to frame AI as an "auxiliary tool" rather than a leading mechanism, the value of AI will be greatly diminished, potentially even causing internal conflicts and resistance.
If the current "AI+" trend remains stuck in "plugin thinking," it may repeat the mistakes of the "industry + internet" era. The true value of AI lies not in simple functional additions but in deeply integrating into workflows, achieving multidimensional information understanding, and providing proactive services.
Recognizing the structural tension between AI and organizations, reshaping information flows, decision chains, and execution paths, is the only way to truly unleash the productivity dividends of the intelligent era.