The Advanced Roadmap for AI Product Managers
-
AI product managers differ from regular product managers not only in their grasp of AI algorithms but, more crucially, in their AI-driven thinking.
Artificial intelligence product design should aim for extreme simplicity in operation, but the simplicity of the front end reflects the complexity of the back end. The more complex the system, the more intelligent it can be.
Similarly, the development of artificial intelligence relies on the collective advancement of the industrial ecosystem. Upstream chips provide computational power, midstream AI vendors focus on developing algorithms and models, and downstream application fields offer real-world scenarios for implementation.
The AI industry chain can be structurally divided into the foundation layer (computing infrastructure), the technology layer (software algorithms and platforms), and the application layer (industry applications and products).
This includes computing hardware (AI chips), computing system technologies (cloud computing, big data, and 5G communications), and data (data collection, labeling, and analysis).
The foundation layer is primarily hardware-centric, including GPUs/FPGAs for performance acceleration, neural network chips, sensors, and middleware, which are prerequisites for supporting AI applications. These hardware components provide the computational power for AI operations, currently dominated by international IT giants.
The technology layer encompasses algorithm theories (machine learning algorithms, brain-inspired algorithms), development platforms (open-source frameworks, technology open platforms), and application technologies (computer vision, natural language understanding, and human-computer interaction).
Currently, domestic AI technology platforms mainly focus on computer vision, speech recognition, and language processing at the application level. Domestic technology-layer companies are growing rapidly, with representative firms including iFlytek, DeepGlint, AISpeech, Horizon Robotics, SenseTime, Yonghong Tech, Megvii, and Unisound.
The application layer primarily integrates the foundation and technology layers with traditional industries to enable diverse scenario applications.
With AI's technological breakthroughs in speech, semantics, computer vision, and other fields, its application across various industrial scenarios is accelerating. This includes industry solutions ("AI+") and typical products (robots, smart speakers, smart cars, drones, etc.).
Unlike the internet's development, AI emphasizes the integration of hardware and software for practical implementation. Therefore, I’ve outlined general AI technologies and related platforms. Only by combining underlying hardware and software with suitable algorithms can intelligent products be developed.
Currently, domestic AI-related companies are positioned in one or more areas of the following landscape.
The advancement of general AI technologies depends on technological maturity and business penetration.
AI capabilities hinge on two factors: technological maturity and penetration into specific business scenarios.
The accuracy of computer vision, speech recognition, and natural language understanding applications lies in the construction of knowledge graphs and machine learning capabilities. AI technology is shifting from single-point applications to holistic solutions, with companies focusing on integrated technological development. AI progress relies on data accumulation, as firms optimize algorithms through scenario penetration to build industry barriers.
AI is transforming hardware devices, and its future market potential depends on the synergy between AI technology and hardware-based application functions. AI is enabling the Internet of Things (IoT) and seamless online-offline data exchange. Innovations in user-device interaction, such as visual, speech, and semantic AI, depend on their ability to interpret contextual data. The broad market potential of AI in hardware requires integration with core hardware functionalities.
At the hardware level, chips are critical for ensuring algorithm performance and computational power. The success of chips depends on their technical capabilities, which vary based on deployment location and tasks.
Cloud chips are typically used for data training, handling massive datasets and requiring strong parallel processing capabilities. Edge chips focus on data inference, where power efficiency is key. Brain-inspired chips, breaking the von Neumann architecture, mimic brain structures to improve computational efficiency and reduce power consumption, representing the long-term trend in AI chips.
The development of visual sensors, unlike software systems with marginal effects, focuses on overcoming cost barriers. LiDAR is crucial for autonomous driving, with the industry concentrating on reducing production costs. Solid-state LiDAR meeting automotive standards should be a strategic priority.
Domestic research on millimeter-wave radar is in its early stages, with 24GHz and 77GHz models currently available. The 77GHz radar offers better accuracy and penetration, poised to dominate the market, making cost reduction a strategic focus. Cameras, integrated with computer vision, enable machine intelligence in security, autonomous driving, and smart TVs, with scene-specific analysis and decision-making as key capabilities.
Analyzing the AI industry structure reveals the need for AI product managers at every node.
Based on company size and individual technical skills, AI product managers can be categorized into four quadrants.
Generally, tech trends evolve through three stages from emergence to decline.
AI technology is still in its first stage, meaning individual technical skills play a larger role in product development. Only large enterprises can afford the financial investment in R&D, while smaller firms lack the resources.
I classify AI product managers into four quadrants:
These four types form a pyramid, with breakthrough AI product managers being rare pioneers driving societal progress. This role requires patience and resilience to withstand failures.
Innovative product managers are those who can implement cutting-edge technology, mastering core tech while identifying innovative opportunities and viable business models. Such opportunities are scarce but highly rewarding.
Application-oriented product managers are the foot soldiers of AI democratization, building user-friendly infrastructure and laying the groundwork for AI adoption.
Widespread AI product managers are found in small and medium-sized enterprises across cities, acting as missionaries injecting AI capabilities into various industries. They are frontline experts deeply familiar with markets and scenarios.
From industry to roles, AI product managers are distributed across positions as follows:
AI product management is still nascent and will become more specialized over time. Focusing on a niche now can lead to expertise in the future.
To enhance skills, targeted learning is essential for efficiency.
Clarify your position, industry, and role in the value chain. AI product managers must be versatile specialists.
Common pitfalls for AI product managers include:
They must possess technical understanding, domain expertise, and a methodology for AI product implementation, balancing technical and demand boundaries.