What does an AI-native application look like?
**I. What are AI-native applications?**
1. **Definition and characteristics**:
– AI-native applications refer to those where AI is at the core of the application experience rather than an auxiliary function. Characteristics include being built on basic AI capabilities, breaking through traditional limitations, having continuous improvement capabilities, and possessing a certain degree of proprietary AI technology.
– It doesn’t mean an application must have GenAI functions from the very beginning. Many companies can evolve from cloud-native to AI-native.
– “AI-native” is a transitional label that will fade as AI becomes a standard feature.
2. **Development status and prospects**:
– The AI-native赛道 attracts a large amount of investment, and the number of applications with significant revenue growth is increasing. It is expected that more applications will reach a high-income scale in the future.
– From Perplexity to Writer, AI-native applications are redefining solutions in various vertical fields.
**II. Five-dimensional framework for evaluating AI-native applications**
1. **Design: Core competitiveness**:
– Create new interaction paradigms:
– Chat and search interfaces become dominant, functions are unlocked, multimodal breakthroughs occur, and evolution towards collaborative canvases and automated tools takes place.
– Accelerate feedback loops:
– Various feedback mechanisms such as likes, dislikes, voting, manual review, and innovative monitoring. Intelligent integration of feedback can achieve faster iterations.
– Build AI-native systems:
– Demonstrate mature system thinking, balance general and proprietary capabilities, and incorporate explainability, such as showing the relationship between input and output and citing sources.
– Application cases: Perplexity.ai and ChatGPT Search enhance the search experience; Hebbia and Reliant AI refine interaction design; Cognition provides professional development tools; Rilla and Bland.ai provide industry solutions.
2. **Data: Key asset**:
– Strengthen end-to-end data management:
– Including data procurement and planning, quality, governance, security, etc. Multimodal data processing capabilities are crucial.
– Activate dormant data:
– Including underutilized data and data not captured by the system, bringing advantages such as smoother interaction and faster access to optimal content.
– Create proprietary data assets:
– Such as multimodal interaction data, metadata, pattern recognition data, etc., which can expand competitive advantages.
– Application cases: Glean, Writer, Jeeva.ai.
3. **Domain Expertise: Accelerator for AI-native applications**:
– Convert domain knowledge into AI workflows:
– GenAI can digitize end-user activities such as more accurate conversation transcription, more comprehensive research summaries, and more accurate relationship understanding. It can also convert advanced user patterns into reusable patterns.
– Scale real-time insights:
– AI-native applications can obtain massive data insights in near real time, speeding up the processing of information for customers in multiple fields and endowing users with superhuman capabilities.
– Fusion of global and local knowledge:
– Integrate global, domain-specific, and organization-specific knowledge to optimize output and automate workflows.
– Application cases: Abridge, EliseAI, Supio, Magic School.
4. **Dynamism: Adaptive engine for AI-native applications**:
– Product experience optimization:
– From a single model to orchestrating the interaction of multiple models, focus on infrastructure flexibility, giving rise to model routers. In the future, advanced AI functions will be more adaptive.
– GenAI customer journey:
– Such as customized marketing and personalized shopping, achieving multi-level ultra-personalized experiences.
– Application cases: HeyGen, Mercor, Evolv AI.
5. **Distribution: Pricing strategy for AI-native applications**:
– Flexible pricing strategies:
– Including embedding GenAI functions for free, creating advanced versions, launching independent applications, and testing consumption-based and results-based charging models.
– Application cases: Pricing methods of companies such as Salesforce, Zendesk, and Sierra.
**III. The future of AI-native applications**
The real breakthrough lies in the innovative integration of the five-dimensional framework to create a unified interface and multimodal applications, and adopt a flexible metered charging model. In the future, improvements are needed at all levels of the technology stack. Although a new generation of models may bring changes, research on reasoning is also very important. With the progress of models and technologies, we will enter a new era of application innovation, but there may also be situations such as slow large-scale deployment and failure to reshape workflows. Enterprises that can quickly integrate new technologies will become an important force.
