🚀 The Rise of Cloud-Native AI.

🧠 The New Backbone of Enterprise IT

AI is no longer a futuristic concept it’s a core component of modern application design. Here’s why:

📊 Key Trends Driving AI Adoption

These stats underscore a seismic shift: AI applications are now central to enterprise architecture.

🔄 The Symbiosis of Applications and Infrastructure

AI demands a new breed of infrastructure. Let’s break down the feedback loop:

🧱 How Applications Shape Infrastructure

• Real-time inference requires millisecond-level elasticity.
• Long-session state retention demands persistent compute instances.
• Multimodal interactions push for asynchronous, high-throughput communication.
• Security and cost-efficiency redefine resource pooling.

⚙️ How Infrastructure Fuels Innovation

When infrastructure offers tools like serverless GPU, AI gateways, and observability stacks, developers can focus on creativity not complexity.

🔁 Iterative Resonance: A Spiral of Co-Evolution

Just like microservices and containers revolutionized cloud-native design, AI applications and infrastructure now evolve in tandem only with greater scale, complexity, and unpredictability.

⚠️ Challenges in Building AI-Native Systems

Despite the momentum, enterprises face three major hurdles:

1. New skill stack for agent development 🧑‍💻
2. Seamless integration with legacy systems 🏛️
3. Stable, secure, scalable operations 🔐

To overcome these, we need a purpose-built AI-native infrastructure.

🧩 Introducing AgentRun: The AI-Native Technology Stack

AgentRun is our blueprint for AI-native architecture, comprising eight core components:

- AI Runtime
- AI Gateway
- AI MQ (Message Queue)
- AI Memory
- AI Observability
- AI DevOps
- AI Evaluation
- AI Marketplace

This modular stack helps enterprises accelerate Proof of Concept (PoC) to production without reinventing the wheel.

⚡ Accelerating Agent Development with Cloud-Native Tools

🧬 Function Computing: The Ideal Runtime for AI Agents

Agent workloads are unpredictable, multi-tenant, and security-sensitive. Enter Function Computing (FaaS):

• Millisecond elasticity for bursty traffic
• Session affinity for long conversations
• Security isolation via containerized sandboxes

Function Computing breaks the traditional stateless FaaS mold by supporting persistent sessions up to 8 hours, enabling rich multi-turn dialogues.

🛠️ Built-In Capabilities

• Multi-language support: Python, Node.js, Java, Shell
• Execution latency <100ms
• Sandboxes: Code, Browser, RL, Compute
• Serverless GPU with memory snapshots
• Fine-grained GPU slicing (1/N card virtualization)

Used by leading clients like Bailian, Modao, and Tongyi Qianwen, Function Computing is the go-to runtime for scalable AI innovation.

📡 RocketMQ for AI: Messaging Reimagined

As agent interactions grow complex, asynchronous communication becomes essential. That’s where RocketMQ for AI shines:

🌟 Key Innovations

• LiteTopic: Lightweight, session-specific topics for context retention
• Millions of topics per cluster with zero performance loss
• Large message support for AIGC payloads (images, long prompts)
• Strict message ordering for coherent LLM outputs

RocketMQ ensures agents can resume sessions, scale throughput, and maintain conversational integrity.

🔗 Bridging Intelligent Agents with Legacy Systems

🧠 Challenge 1: Connecting Old Systems to New Intelligence

Legacy systems often use REST/HTTP protocols and lack agent compatibility. Rebuilding them is costly. The solution? Alibaba Cloud Native API Gateway:

🧰 Features

• Protocol adaptation (SSE, WebSocket)
• Built-in observability and security
• Hardware acceleration (TLS, Gzip)
• Plugin ecosystem for hot updates

This gateway enables low-cost, high-impact AI upgrades.

🧪 Challenge 2: Rapid MCP Server Deployment

MCP (Model Context Protocol) is the emerging standard for agent tools. Building MCP servers can be complex unless you use Function Computing and AI Gateway:

• Zero-maintenance runtime with millisecond elasticity
• Multi-language support for fast development
• Unified management of registration, throttling, and observability
• Zero-code HTTP-to-MCP transitions

Perfect for enterprises creating private AI marketplaces.

🛡️ Ensuring Stable Operations for AI Agents

🌐 AI Gateway: Traffic Control for Intelligent Applications

AI traffic is unique streaming, high-latency, long-lived. The AI Gateway addresses this with:

• Multi-model routing and failover
• Security integrations (WAF, API keys)
• Protocol unification (MCP, A2A, model calls)

Available in both Higress (open source) and commercial forms.

🔍 AI Observability: Full-Stack Insight for AI Systems

AI apps face challenges in latency, cost, and output quality. Observability tools help mitigate these:

📈 Capabilities

• End-to-end tracing of requests and dialogues
• Real-time monitoring of latency, throughput, token usage
• Automated evaluation of hallucinations, bias, and safety
• Pre-integrated dashboards with PAI, Bailian, and container services

Bonus: An AIOps assistant powered by multi-agent architecture autonomously diagnoses and resolves issues.

🧪 AI DevOps: Closing the Loop

To truly embed AI into enterprise workflows, DevOps must evolve. Alibaba Cloud’s AI DevOps connects:

• Coding → Building → Monitoring
• Integrated with Lingma, AIOps, and more
• Auto-generates diagnostic reports post-release

This ensures continuous delivery of intelligent applications.

🌍 Open Source: The Heartbeat of Cloud-Native Innovation

While we’ve spotlighted commercial products, our commitment to open source remains unwavering. Key contributions include:

This dual-track evolution commercial + open source ensures enterprise-grade reliability and community-driven innovation.

🎯 Building the Future, One Agent at a Time

AI-native architecture isn’t just a trend it’s the foundation of tomorrow’s enterprise software. With tools like AgentRun, Function Computing, and AI Gateway, businesses can confidently build, integrate, and scale intelligent applications.

Ready to explore? Dive into our console and experience the future firsthand.