2025-11-20 Tech Briefing | A new approach to building RAG applications

Thursday

Welcome to this edition, featuring 30 articles.

📚 Contents/📝 TL;DR

📚 Contents

📝 TL;DR

Primary

Accelerating Generative AI Use Cases with Amazon Bedrock and Oracle Database

A new approach to building RAG applications

Accelerating Generative AI Use Cases with Amazon Bedrock and Oracle Database

A new approach to building RAG applications

Building Scalable IPv4 Addressing with AWS NAT Gateway, Amazon VPC IPAM Policies, and Prefix Lists in Regional Availability Mode

AWS has introduced NAT Gateway and IPAM new features to simplify IPv4 management.

Building Scalable IPv4 Addressing with AWS NAT Gateway, Amazon VPC IPAM Policies, and Prefix Lists in Regional Availability Mode

AWS has introduced NAT Gateway and IPAM new features to simplify IPv4 management.

How Tencent Music Reduced Kafka Costs by Over 50% Using AutoMQ

Tencent Music reduced Kafka costs by over 50% using AutoMQ

How Tencent Music Reduced Kafka Costs by Over 50% Using AutoMQ

Tencent Music reduced Kafka costs by over 50% using AutoMQ

Stop Overusing Base64—Blob Is the Right Approach for Frontend Unloading

Overusing Base64 is a misconception; Blob is a powerful tool for handling binary data on the frontend.

Stop Overusing Base64—Blob Is the Right Approach for Frontend Unloading

Overusing Base64 is a misconception; Blob is a powerful tool for handling binary data on the frontend.

Improving API Discoverability with the New Amazon API Gateway Portal

AWS introduces a new API Gateway management portal.

Improving API Discoverability with the New Amazon API Gateway Portal

AWS introduces a new API Gateway management portal.

Building and Applying Large Model Systems That Integrate Risk Control Knowledge

Tencent Tianyu shares a financial risk control large model to address challenges in small-sample modeling.

Building and Applying Large Model Systems That Integrate Risk Control Knowledge

Tencent Tianyu shares a financial risk control large model to address challenges in small-sample modeling.

Building an AI Gateway with Amazon API Gateway

Best practices for building an AI gateway

Building an AI Gateway with Amazon API Gateway

Best practices for building an AI gateway

Moving from C++ to Rust? What ClickHouse Has to Say

ClickHouse's practical experience with C++ and Rust hybrid development reveals pain points but remains optimistic.

Moving from C++ to Rust? What ClickHouse Has to Say

ClickHouse's practical experience with C++ and Rust hybrid development reveals pain points but remains optimistic.

WebAssembly 3.0 Released, 64-bit Memory and Garbage Collection Included, But Don't Get Too Excited – Component Model Still Not Done!

Wasm 3.0 is released, but the component model is not yet complete.

WebAssembly 3.0 Released, 64-bit Memory and Garbage Collection Included, But Don't Get Too Excited – Component Model Still Not Done!

Wasm 3.0 is released, but the component model is not yet complete.

CRRC Collaborates with PaddlePaddle Wenxin: AI Empowers High-Speed Rail Development, Establishing a New Benchmark for Intelligent Manufacturing

CRRC and PaddlePaddle Wenxin Collaborate to Empower High-Speed Rail with AI

CRRC Collaborates with PaddlePaddle Wenxin: AI Empowers High-Speed Rail Development, Establishing a New Benchmark for Intelligent Manufacturing

CRRC and PaddlePaddle Wenxin Collaborate to Empower High-Speed Rail with AI

Simplify Access to External Services Using AWS IAM Outbound Federation

New way for secure access to external services

Simplify Access to External Services Using AWS IAM Outbound Federation

New way for secure access to external services

Meta Replaces Native WhatsApp With a Poor Web App: Windows Users Deserve to Compromise?

WhatsApp desktop version returns to Web packaging, performance regresses.

Meta Replaces Native WhatsApp With a Poor Web App: Windows Users Deserve to Compromise?

WhatsApp desktop version returns to Web packaging, performance regresses.

Google Preview of Code Wiki: Can You Trust AI to Document Your Codebase?

A preview of Google's AI-based code documentation tool, with doubts about its reliability.

Google Preview of Code Wiki: Can You Trust AI to Document Your Codebase?

A preview of Google's AI-based code documentation tool, with doubts about its reliability.

Essential

Cloudflare's Worst Outage in Six Years: A Rust Code Line Causes Global Traffic Collapse! ChatGPT and Claude Go Offline

Cloudflare's outage caused by a configuration change disrupts global internet traffic

Cloudflare's Worst Outage in Six Years: A Rust Code Line Causes Global Traffic Collapse! ChatGPT and Claude Go Offline

Cloudflare's outage caused by a configuration change disrupts global internet traffic

LLM is Boring, Zuckerberg's Decision is Weak, LeCun Leaves to Found AMI

LeCun leaves Meta to start his own company, betting on the world model approach.

LLM is Boring, Zuckerberg's Decision is Weak, LeCun Leaves to Found AMI

LeCun leaves Meta to start his own company, betting on the world model approach.

What is Platform Engineering?

Platform engineering represents the evolution of DevOps, aimed at enhancing the developer experience with a product-oriented mindset.

What is Platform Engineering?

Platform engineering represents the evolution of DevOps, aimed at enhancing the developer experience with a product-oriented mindset.

Stop Blowing Performance Optimization: Your App is Lagging Due to Poor Product Design 🤷‍♂️

Poor performance stems from bad design; cutting requirements is the real optimization.

Stop Blowing Performance Optimization: Your App is Lagging Due to Poor Product Design 🤷‍♂️

Poor performance stems from bad design; cutting requirements is the real optimization.

Reducing False Positives in Retrieval-Augmented Generation (RAG) Semantic Cache: A Banking Case Study

Banking RAG semantic cache false positive rate drops from 99% to 3.8%

Reducing False Positives in Retrieval-Augmented Generation (RAG) Semantic Cache: A Banking Case Study

Banking RAG semantic cache false positive rate drops from 99% to 3.8%

Building Reactive APIs with Amazon API Gateway Response Streams

API Gateway now supports response streams, enhancing API real-time performance.

Building Reactive APIs with Amazon API Gateway Response Streams

API Gateway now supports response streams, enhancing API real-time performance.

From Detection to Generalized Perception: Building the Foundation for Spatial Intelligence

Deep dive into visual generalized perception technology, laying the foundation for spatial intelligence.

From Detection to Generalized Perception: Building the Foundation for Spatial Intelligence

Deep dive into visual generalized perception technology, laying the foundation for spatial intelligence.

React Foundation Leader Talks About the Next Steps for the Framework

React Foundation is established, planning community, technology, and commercial development.

React Foundation Leader Talks About the Next Steps for the Framework

React Foundation is established, planning community, technology, and commercial development.

On the Eve of Agent Scaling, AI Infra Challenges Are Fully Exposed

AI Agent deployment exposes AI Infra challenges, requiring re-evaluation of value and cost control.

On the Eve of Agent Scaling, AI Infra Challenges Are Fully Exposed

AI Agent deployment exposes AI Infra challenges, requiring re-evaluation of value and cost control.

2025 Tencent Cloud Big Data - Annual Selected Technical Practice Guide (Part 1)

Tencent Cloud releases the 2025 Big Data Technical Practice Guide

2025 Tencent Cloud Big Data - Annual Selected Technical Practice Guide (Part 1)

Tencent Cloud releases the 2025 Big Data Technical Practice Guide

Understanding Specification-Driven Development: Kiro, spec-kit, and Tessl

Exploration and Analysis of SDD Tools

Understanding Specification-Driven Development: Kiro, spec-kit, and Tessl

Exploration and Analysis of SDD Tools

Advanced Notice: Amazon S3 Will Default Disable SSE-C Encryption for All New and Some Existing Buckets Starting April 2026

AWS S3 will disable SSE-C encryption by default.

Advanced Notice: Amazon S3 Will Default Disable SSE-C Encryption for All New and Some Existing Buckets Starting April 2026

AWS S3 will disable SSE-C encryption by default.

General

InfraTalk Live Session 1: Full Version｜Agent Wave: Evolution of AI Infra and Directional Insights

Discussion on AI Infra development

InfraTalk Live Session 1: Full Version｜Agent Wave: Evolution of AI Infra and Directional Insights

Discussion on AI Infra development

Fragments 11 November: Observations from Europe

Martin's observations from his trip to Europe and summary of AI trends.

Fragments 11 November: Observations from Europe

Martin's observations from his trip to Europe and summary of AI trends.

Answering Developers' Questions: HarmonyOS Technical Problem Analysis, Episode 7

HarmonyOS Developer Common Questions and Answers, Episode 7

Answering Developers' Questions: HarmonyOS Technical Problem Analysis, Episode 7

HarmonyOS Developer Common Questions and Answers, Episode 7

New Journey, Aim for the Crown｜Cloud Valley Cup 2025 AI Application Innovation and Entrepreneurship Competition Semi-Final Successfully Held

2025 Cloud Valley Cup AI Competition Semi-Final Concluded, Ten Teams Advance to Final

New Journey, Aim for the Crown｜Cloud Valley Cup 2025 AI Application Innovation and Entrepreneurship Competition Semi-Final Successfully Held

2025 Cloud Valley Cup AI Competition Semi-Final Concluded, Ten Teams Advance to Final

The Three Stars of Orion: Alnitak, Alnilam, and Mintaka

Introduction to the three bright blue supergiants in Orion's Belt.

The Three Stars of Orion: Alnitak, Alnilam, and Mintaka

Introduction to the three bright blue supergiants in Orion's Belt.

Primary

Accelerating Generative AI Use Cases with Amazon Bedrock and Oracle Database

AWS - Databases•Published on November 19

Insight•AI and Frontier Technologies•Score: 9/10

Oracle AI DatabaseRAGVector SearchBedrockLangChain

Source

◆

Building Scalable IPv4 Addressing with AWS NAT Gateway, Amazon VPC IPAM Policies, and Prefix Lists in Regional Availability Mode

AWS 网络•Published on November 19

Insight•Cloud Computing and Backend•Score: 9/10

AWSNAT GatewayIPAMIPv4 ManagementNetwork Automation

🔬

Tech Insights

1. Regional Availability Mode NAT Gateway: Previously, each availability zone required its own NAT Gateway, but now only one per region is needed, with automatic scaling. This is a blessing for operations engineers, eliminating a lot of repetitive configuration and maintenance work, and reducing the complexity of multi-AZ architectures. 2. IPAM Policy Centralized Control: By using IPAM policies, NAT Gateway can be forced to use specific IP pool addresses, whether AWS-owned or BYOIPv4. This ensures consistency and compliance in IP allocation. For large organizations, this solves the chronic issues of disorganized IP address allocation and difficulty in tracking. 3. IPAM Prefix List Automation: This is the standout feature! It automatically maintains a list of all outbound IP prefixes and can share it with partners through RAM. This means external whitelist management is fully automated, eliminating the need for manual notifications when IPs change and avoiding many incidents of service disruption caused by IP changes.

Source

◆

How Tencent Music Reduced Kafka Costs by Over 50% Using AutoMQ

InfoQ - Cloud•Published on November 20

Insight•Cloud Computing and Backend•Score: 9/10

AutoMQKafkacost optimizationcloud-nativestorage and compute separation

Source

◆

Stop Overusing Base64—Blob Is the Right Approach for Frontend Unloading

Juejin Frontend Hottest•Published on November 18

Insight•Frontend Development•Score: 8/10

BlobBase64Frontend PerformanceMemory ManagementURL.createObjectURL

Source

◆

Improving API Discoverability with the New Amazon API Gateway Portal

AWS 计算•Published on November 20

Insight•Cloud Computing and Backend•Score: 8/10

API Gateway PortalAPI ManagementAWSAPI DiscoveryDeveloper Experience

Source

◆

Building and Applying Large Model Systems That Integrate Risk Control Knowledge

InfoQ - AI & LLM•Published on November 20

Insight•AI and Frontier Technologies•Score: 8/10

Financial Risk ControlLarge ModelsSmall-Sample LearningTencent TianyuDiscriminative AI

📈

Perspectives

In my view, financial risk control is one of the vertical fields where AI large models can generate practical commercial value the fastest, as the value of risk identification is directly reflected in avoiding financial losses. Tencent Tianyu's 'service provider technology perspective' and 'MaaS implementation approach' represent a typical 'To B' (business-to-business) solution strategy. By adopting cloud-native technologies and a 'model marketplace', it reduces the barriers for customers and solves the issue of data staying within the organization. Looking ahead to November 2025, financial institutions will invest more in large models, but they will be seeking 'out-of-the-box' or 'low-cost customization' solutions. Only by encapsulating core capabilities and offering modeling capabilities for small or even zero samples can solutions like Tianyu truly resonate with customers. However, it's important to avoid overselling 'large models' as a panacea; the ultimate test will be their actual business effectiveness.

Source

◆

Building an AI Gateway with Amazon API Gateway

AWS - Architecture•Published on November 19

Insight•AI and Frontier Technologies•Score: 8/10

Amazon BedrockAPI GatewayAI GatewayJWT ValidationLambda

Source

◆

Moving from C++ to Rust? What ClickHouse Has to Say

InfoQ - Backend•Published on November 20

Insight•Cloud Computing & Backend•Score: 8/10

ClickHouseC++Rustlanguage migrationmemory safetyhybrid programming

🔬

Tech Insights

ClickHouse opted for a progressive migration approach rather than an aggressive 'full Rust rewrite,' starting with integrating into the build system and then introducing modules incrementally. This is a more pragmatic and less risky method in real-world engineering. It acknowledges Rust's advantages in memory safety, thread safety, and ecosystem libraries (such as Delta Lake format library delta-kernel-rs), especially its rapid support for emerging data standards, which C++ struggles to match. It clearly states that while C++ is 'technologically mature, widely recognized, and easy to hire,' its security issues and massive build systems are significant trade-offs. The article's candid stance is commendable. The conclusion is that Rust is a great language, but it's not a universal solution, particularly when mixed with C++—it introduces new challenges such as build reproducibility, memory allocation, runtime panics, and dependency management.

⚠️

Notable

The article mentions that Rust libraries 'often overuse panic,' leading to server applications crashing suddenly. In my view, this is a major flaw. For backend systems, stability is the lifeblood; frequent panics could result in high costs for troubleshooting and resolution in production. The complexity of mixed C++ and Rust compilation, including the build system, cross-dependencies, and composability issues, is truly an 'engineering nightmare.' Introducing a 'safer' language may end up increasing engineering complexity. Rust's number of transitive dependencies (672) far exceeds C++ (156), although Milovidov jokes 'not as much as NPM.' For systems that prioritize extreme performance and control over dependencies, this significantly increases build time and security risks. The article does not mention the steep learning curve of Rust or the training costs for existing C++ teams to transition, which are important challenges in real-world projects.

📈

Perspectives

In 2025, the call for Rust as the 'next-generation systems programming language' is growing louder, but ClickHouse's experience tells us that 'ideals are full, reality is bony.' Especially for 'monolithic' systems with significant legacy code, smooth migration is a long and arduous process. For domestic backend technology stack choices, we should recognize Rust's potential and advantages but also be cautious of its 'implementation barriers and potential risks.' In core business areas, blindly chasing new trends may not be cost-effective. This practical experience offers us insight: instead of completely overhauling the system, start by introducing Rust in non-core modules or new features, take small steps, and gradually accumulate experience and clarify boundaries. Architectural evolution is an art of balance, not just a simple language replacement.

Source

◆

WebAssembly 3.0 Released, 64-bit Memory and Garbage Collection Included, But Don't Get Too Excited – Component Model Still Not Done!

InfoQ - Frontend•Published on November 20

Insight•Cloud Computing and Backend•Score: 8/10

WebAssembly 3.0Component ModelWASIGarbage Collection64-bit Memory

Source

◆

CRRC Collaborates with PaddlePaddle Wenxin: AI Empowers High-Speed Rail Development, Establishing a New Benchmark for Intelligent Manufacturing

InfoQ - AI & LLM•Published on November 20

News•AI and Frontier Technologies•Score: 7/10

CRRCIndustrial AIPaddlePaddle WenxinAerodynamic SimulationIntelligent Operations

🔬

Tech Insights

The dramatic leap in aerodynamic simulation efficiency: traditional days-level computation cycles were reduced to single GPU seconds-level, achieving over 30 times efficiency improvements, driving the transition from experience-driven to data-driven design in shape engineering, which is a typical success case of AI in industrial design. The rapid construction and deployment of virtual sensor models: from years to months, with an increase in fault detection accuracy of 10%, solving the high cost and long cycle issues of traditional sensor deployment, providing new ideas for rail transit safety operations. The construction of the 'Zhuo Lun' industry large model system by CRRC is the first industry large model of a central state-owned enterprise (SOE) in the equipment manufacturing sector to be approved by the National Internet Information Office, reflecting the national team's layout and determination in industrial AI.

📈

Perspectives

In my view, the 'National Team + Leading AI Platform' collaboration model is one of the main paths for the practical application of industrial AI in China. As a giant in the equipment manufacturing industry, CRRC's AI practice has strong demonstration effects. By November 2025, industrial large models have moved from concept to implementation, but the real challenge lies in the deep integration of 'know-how' with AI technology. CRRC's integration of AI into core business scenarios such as aerodynamic simulation and intelligent operations is the correct direction. This case also reminds us that in many traditional capital-intensive industries, the value of AI is not in replacing human labor, but in enhancing efficiency, optimizing decision-making, and ensuring safety. However, we should be cautious, as many so-called 'industrial large models' are merely traditional algorithms wrapped in a 'large model' shell. The key is to assess actual business impact and innovation.

Source

◆

Simplify Access to External Services Using AWS IAM Outbound Federation

AWS 新闻•Published on November 19

Insight•Cloud Computing and Backend•Score: 7/10

IAM Outbound FederationJWTSTSSecurity AuthenticationLong-term Credentials

Source

◆

Meta Replaces Native WhatsApp With a Poor Web App: Windows Users Deserve to Compromise?

InfoQ - Frontend•Published on November 20

News•Frontend Development•Score: 7/10

WhatsAppWebView2ElectronUWPNative Application

Source

◆

Google Preview of Code Wiki: Can You Trust AI to Document Your Codebase?

InfoQ - AI & LLM•Published on November 20

Insight•AI and Emerging Technologies•Score: 6/10

Code WikiAI documentationCode generationGemini CLIReliability

⚠️

Notable

Although AI-generated documentation sounds advanced, there are significant doubts about its understanding of context. For example, when Gemini was asked about PostgreSQL support, it showed potential inaccuracies. Many developers have criticized the documentation for lacking context, being either too vague or too detailed, or containing unsubstantiated claims, which can be problematic. In reality, it might be even worse than having no documentation at all, as users would have to spend time verifying the accuracy of the information. This so-called 'Code Wiki' doesn't truly meet the definition of a wiki, as it lacks collaborative and open editing features. The automatic updates, while appealing, could lead to confusion if they are too frequent and of poor quality, increasing developers' workload.

Source

Essential

Cloudflare's Worst Outage in Six Years: A Rust Code Line Causes Global Traffic Collapse! ChatGPT and Claude Go Offline

InfoQ - AI & LLM•Published on November 20

News•Cloud Computing and Backend•Score: 10/10

CloudflareoutageRustconfiguration errorsystem coupling

Source

◆

LLM is Boring, Zuckerberg's Decision is Weak, LeCun Leaves to Found AMI

InfoQ - AI & LLM•Published on November 20

News•AI and Frontier Technologies•Score: 10/10

LeCunAMIWorld ModelLLM BubbleFAIR

Source

◆

What is Platform Engineering?

CNCF•Published on November 19

Insight•Software Architecture and Engineering Practice•Score: 9/10

Platform EngineeringDevExDevOpsSelf-ServiceCognitive Burden

Source

◆

Stop Blowing Performance Optimization: Your App is Lagging Due to Poor Product Design 🤷‍♂️

Juejin Frontend Hottest•Published on November 19

Insight•Software Architecture and Engineering Practices•Score: 9/10

Performance OptimizationProduct DesignFrontend DevelopmentCognitive LoadRequirement Review

Source

◆

Reducing False Positives in Retrieval-Augmented Generation (RAG) Semantic Cache: A Banking Case Study

InfoQ - AI & LLM•Published on November 20

Insight•AI and Frontier Technologies•Score: 9/10

RAGSemantic CacheFalse PositivesDual EncoderBest Candidate Principle

Source

◆

Building Reactive APIs with Amazon API Gateway Response Streams

AWS 计算•Published on November 19

Insight•Cloud Computing and Backend•Score: 9/10

API Gatewayresponse streamLLM applicationreal-time performanceTTFB

🔬

Tech Insights

1. Core Pain Point Solved: Previously, one of the biggest drawbacks of API Gateway was its rigid request-response model, which made handling large response bodies and long-running operations require various workarounds. Now, direct support for streaming resolves issues like TTFB, 10MB response body limits, and 29-second timeouts, even extending to 15 minutes. This is a lifesaver for applications like AI Agents and chatbots that require real-time feedback. 2. Diverse Integration Options: Whether using Lambda, HTTP Proxy, or private integration, clear configuration and code examples are provided. Especially for Lambda, it directly offers streamifyResponse as a higher-order function, showing AWS is making an effort to improve user experience. 3. Not a Universal Solution: The article also mentions that advanced features like VTL transformation and integration response caching are not supported in response stream mode. This aligns with expectations, as streaming and caching/transformation logic inherently conflict. Developers need to weigh these trade-offs.

📈

Perspectives

1. Domestic Market Impact: In China, real-time communication and AI applications are already highly competitive. Although AWS's feature fills the gap, for teams accustomed to more flexible and China-centric solutions like Tencent Cloud API Gateway or Alibaba Cloud API Gateway, the appeal may not be as strong. After all, domestic users have extremely low tolerance for latency and experience, with many teams already setting up Nginx + SSE or using Kafka/RabbitMQ as message queues to handle real-time needs. 2. New Opportunities for AI Applications: For pure AWS technology stacks, especially those targeting overseas markets or highly regulated AI enterprises, this feature definitely reduces development complexity, allowing them to focus more on business logic. However, if aiming for the domestic market, network optimization and multi-cloud strategies remain unavoidable topics. Don't expect a single API Gateway to solve all issues.

Source

◆

From Detection to Generalized Perception: Building the Foundation for Spatial Intelligence

InfoQ - AI & LLM•Published on November 20

Insight•AI and Frontier Technologies•Score: 9/10

Generalized PerceptionObject DetectionTransformerDINO-XSpatial Intelligence

📈

Perspectives

In my view, generalized perception capability is the core foundation for future embodied intelligence and industrial automation scenarios. Once mature, it will fundamentally change how robots interact and how factories conduct quality inspection. Currently, many domestic efforts on large models are just surface-level in terms of language and multimodal capabilities, but the true 'hardcore' perception ability is key to building AI in the physical world. Meta's SAM and IDEA's DINO series represent the leading explorations on this path. Given the current date (November 2025), if DINO-X models can improve on edge deployment and customized training, their commercial potential in smart manufacturing, autonomous driving, and smart security could be massive, even potentially creating new unicorns. Now it's a matter of who can turn these 'lab SOTA' into 'industry SOTA'.

Source

◆

React Foundation Leader Talks About the Next Steps for the Framework

InfoQ - Frontend•Published on November 20

News•Frontend Development•Score: 9/10

React FoundationReact NativeMetacommunity governanceframework development

🔬

Tech Insights

React's separation from Meta and its establishment as an independent foundation under the Linux Foundation is a significant move, indicating the maturity and independence of the React ecosystem, no longer entirely dependent on a single company. Meta remains the largest contributor and will hold a majority of board seats for the first 2.5 years, ensuring initial stability and direction for the foundation, while also highlighting Meta's continued strong influence over React. The foundation has clearly stated its commitment to investing in the community and improving training, which means more standardized learning resources and community support will be available in the future, greatly benefiting our team in cultivating frontend talents. Giants like Amazon, Microsoft, and Vercel joining as commercial partners demonstrate React's widespread recognition and influence in the industry, and also provide financial backing for its future development.

⚠️

Notable

While the foundation talks about 'open governance,' Meta's absolute control over the first two and a half years raises questions about the level of openness. After all, the one who invests the most has the most say, a truth that has always been the case. One of the foundation's core goals is to 'get more people to know React and re-attract some developers who have distanced themselves,' which inadvertently exposes some past issues and controversies in React's community and development experience. The foundation needs to address these. Honestly, 'selling merchandise' can also be a revenue source for the foundation, which, in my view, has a 'fan circle' trend. While it can generate income, the foundation's core value and efforts should be more focused on substantive technical and community building.

📈

Perspectives

The establishment of the React Foundation is a significant milestone in the development of frontend frameworks. In today's increasingly competitive landscape among frameworks like Vue, Angular, and Svelte, React aims to further solidify its position as the dominant ecosystem through the foundation model. Considering the timing of November 2025, when the AI large model wave is in full swing, the frontend development field is also evolving rapidly. The challenge for the foundation is how to better integrate React with AI and enhance development efficiency. This is a critical challenge for its future development. For the domestic market, React's position remains strong, but the foundation needs to continue thinking and investing in local community and training resources, as well as reducing the learning curve to enable more developers to quickly adopt and benefit from React. This relates to the cost-benefit considerations for our team in recruitment and technology stack selection.

Source

◆

On the Eve of Agent Scaling, AI Infra Challenges Are Fully Exposed

InfoQ - Cloud•Published on November 20

Insight•Cloud Computing and Backend•Score: 9/10

AI AgentAI InfraCost ControlHeterogeneous Computing PowerCloud ComputingScaling Deployment

🔬

Tech Insights

1. Value Reassessment and Six Capabilities: Moving from focusing solely on computing power to paying attention to business efficiency, cost, security, and deployment potential marks the maturity of AI Infra. The six modules proposed by IDC serve as an important reference framework for building AI Infra. 2. Cost Control as a Systems Engineering Approach: The practical concept of 'running the minimum business closed-loop at the lowest cost and shortest cycle' proposed by Luo Longsheng is highly relevant. Token consumption is a major component of AI applications, and the article mentions context engineering, tiered model routing, and memory system design, which are essential for our team to manage meticulously in real projects. 3. Openness and Ecosystem: Huang Yang's view on 'single-cloud binding' is reasonable. At this point in 2025, if cloud vendors still think of 'binding' customers tightly, their vision is limited. Future competition will truly hinge on openness and ecosystem integration capability.

Source

◆

2025 Tencent Cloud Big Data - Annual Selected Technical Practice Guide (Part 1)

InfoQ - Big Data•Published on November 19

Insight•Cloud Computing and Backend•Score: 8/10

Data+AI IntegrationTencent Cloud Big DataEMRWeDataTCHouse

Source

◆

Understanding Specification-Driven Development: Kiro, spec-kit, and Tessl

InfoQ - AI & LLM•Published on November 20

Insight•AI and Frontier Technologies•Score: 8/10

specification-driven developmentSDDAI programmingtool comparisoncode generation

Source

◆

Advanced Notice: Amazon S3 Will Default Disable SSE-C Encryption for All New and Some Existing Buckets Starting April 2026

AWS 存储•Published on November 19

News•Cloud Computing & Backend•Score: 7/10

AWS S3SSE-CSSE-KMSEncryption PolicyCloud Security

🔬

Tech Insights

1. Clear policy shift direction: AWS explicitly states that SSE-KMS surpasses SSE-C in terms of security, management granularity, and audit capabilities. This adjustment is a step toward pushing users to adopt more modern and secure encryption practices. It's a correct direction, as security is never a trivial matter. 2. Sufficient transition period: The change takes effect in April 2026, which is a long time from now (November 2025), giving our team ample time to investigate and make necessary modifications. This reflects AWS's responsibility as a platform provider, avoiding sudden disruptive changes. 3. Emphasize advantages of cloud-native encryption: The article also highlights the strengths of AWS KMS, particularly in key lifecycle management, IAM policy integration, and CloudTrail auditing. These capabilities are crucial for building highly compliant systems.

⚠️

Notable

1. Although SSE-C's 'no key storage' feature may have its 'unique' value in certain extreme scenarios (e.g., when users want complete control over keys and distrust any cloud service provider storing keys), AWS's move essentially signals the end of SSE-C's mainstream usage. For those few users who are still using SSE-C and are 'stubborn', they will need to manually enable it, increasing configuration complexity. 2. The 'simplification' of S3 encryption options is essentially the S3 product team making 'subtractions' to reduce maintenance of outdated features. This is understandable, but for some systems with historical dependencies, it remains a 'minor hassle' requiring human effort to adapt. 3. Customers need to spend time auditing and investigating. While tools are recommended, for large enterprises, comprehensive auditing is still a significant workload.

Source

General

InfraTalk Live Session 1: Full Version｜Agent Wave: Evolution of AI Infra and Directional Insights

InfoQ - Cloud•Published on November 20

News•AI and Frontier Technologies•Score: 6/10

AI InfraAgentCloud InfrastructureTechnical EvolutionIndustrial Value

Source

◆

Fragments 11 November: Observations from Europe

Martin Fowler•Published on November 19

News•AI Engineering•Score: 5/10

Technology RadarAI AgentsData Mesh

Source

◆

Answering Developers' Questions: HarmonyOS Technical Problem Analysis, Episode 7

InfoQ - Frontend•Published on November 20

Insight•Frontend Development•Score: 5/10

HarmonyOSHSP modulefile storageportrait/landscape adaptationPreferences

Source

◆

New Journey, Aim for the Crown｜Cloud Valley Cup 2025 AI Application Innovation and Entrepreneurship Competition Semi-Final Successfully Held

InfoQ - AI & LLM•Published on November 20

News•AI and Frontier Technologies•Score: 5/10

AI CompetitionEntrepreneurship and InnovationHangzhouPolicy SupportNew Quality Productivity

Source

◆

The Three Stars of Orion: Alnitak, Alnilam, and Mintaka

NASA APOD•Published on November 20

News•Comprehensive News•Score: 1/10

OrionNebulaStarsUniverseAstronomy