Public Opinion on AI Governance Issues in the Age of Machine Development

Two astronomical images that garnered significant attention last week, representing considerable scientific value, are featured weekly in the program Olhar Espacial.

Asteroid Torifun

The first image is an unprecedented record of the asteroid Torifun. At the beginning of the month, this object was flown by the Japanese robotic probe Hayabusa2. This probe previously visited and collected samples from another asteroid, Ryugu, in 2018.

For ground-based telescopes, Torifun appeared as an elongated asteroid; however, the passage of Hayabusa2 revealed that it is actually a contact binary object formed by the gentle collision of two separate asteroids. In addition to the two lobes, Torifun has numerous large rock blocks and shows no distinct craters, suggesting its surface is a collection of debris.

Hayabusa2 is currently continuing its journey through the Solar System and is scheduled to encounter another asteroid, 1998 KY26, only in 2031.

Exoplanet in Beta Pictoris

The second image is a series of recordings showing the exoplanet Beta Pictoris b over more than ten years. This planet was discovered through direct imaging taken with the VLT – Very Large Telescope – in Chile, as well as with the James Webb Space Telescope.

Beta Pictoris b is the third planet discovered orbiting Beta Pictoris, which is the second brightest star in the constellation of the Painter, located 63.4 light-years away. Although the star should be at the center of the images, it was masked or removed during digital processing.

In early images, the exoplanet was obscured by the glare of another object, Beta Pictoris d. Identifying Beta Pictoris b, which is one of the faintest exoplanets discovered via direct observation, required additional digital processing to remove the brighter planet.

For most small and medium-sized enterprises (MSMEs), adopting artificial intelligence is not following a trend, but a strategic advantage that helps solve daily business challenges with limited resources. During the Google Gemini MSME Sparks 2026 session, Mai Kim, an Applied AI Specialist and Google Cloud JAPAC Customer Engineer, demonstrated how companies can create intelligent customer service agents in less than an hour using Gemini Enterprise for customer experience, all without writing a single line of code.

Applying AI Agents in Business

These AI agents are capable of automating responses to customer inquiries outside of business hours, tracking orders, managing returns, or qualifying potential leads. Thus, small businesses can overcome structural and organizational barriers by providing enterprise-level customer service experiences even with small teams.

The session reflected a broader trend in enterprise AI. According to Google Cloud, about 75% of its customers are already using AI products, and the company's models processed over 16 billion tokens per minute through customer APIs in 2026. This indicates that AI agents are moving beyond experimental status to become practical tools for businesses of any size.

A New Frontier in Customer Service

Mai Kim noted that in this AI-driven era, customers expect much more, moving from simple chatbots to deeply personalized service and purchasing experiences. She provided examples of agents that can analyze handwritten recipes and instantly add necessary ingredients to a customer's cart, or process text orders if the customer is late. Her goal is to ensure quick resolution of necessary queries.

Traditionally, meeting these expectations required significant investment in technology and customer support teams for MSMEs. Google Cloud's vision aims to eliminate this hurdle.

Demonstration of a Multimodal Agent

During the presentation, Kim showcased a multimodal AI agent capable of handling voice interactions, understanding customer intent, reasoning through complex dialogues, and retrieving up-to-date business information—all through a self-service interface. She reported that creating such an agent took her less than an hour, highlighting the progress in modern AI, which accelerates development through rapid configuration using natural language and visual tools.

Using the example of a growing D2C brand, this agent, created in 60 minutes, could handle inquiries, suggest complementary products, process order returns, and escalate complex questions to a human representative. The result was accelerated service, increased stability, and reduced operational costs—critical advantages for small businesses.

Gemini Enterprise for Customer Experience

The central element of the session was Google's Gemini Enterprise platform for customer experience. This unified platform integrates AI services for customer service, e-commerce, and search into a single ecosystem. Instead of viewing customer support as an isolated function, the platform connects all stages of the customer journey: from product discovery and purchase to after-sales support and performance analysis.

The suite offers six key capabilities, including Customer Experience Agent Studio, Agent Assist, Customer Experience Insights, ready-made agents for shopping and food ordering, and Vertex AI Search. This allows companies to avoid building every feature from scratch, instead utilizing pre-built AI components adapted to their specific products, customers, and workflows.

One of the platform's main strengths is its ability to balance automation and human intervention. While AI agents can autonomously handle routine tasks, the Agent Assist feature goes further by supporting live representatives during conversations, providing contextual answers. As Kim demonstrated, this feature delivers relevant knowledge and guidance in real time.

Furthermore, Customer Experience Insights helps businesses track key performance indicators, such as customer satisfaction, response quality, and agent performance, through an integrated dashboard, leading to more effective support with smaller teams.

Significance for Small Businesses

Small companies often hesitate to adopt AI due to budget constraints, lack of technical expertise, and implementation time. Kim emphasized that Gemini Enterprise is designed to remove these barriers, offering intuitive low-code and no-code tools that allow organizations to build high-quality AI agents without hiring specialized development teams.

Instead of spending months on development or hiring dedicated AI engineers, business teams can build, test, refine, and deploy conversational agents using visual workflows. This democratization of AI can be particularly valuable for MSMEs striving to compete with much larger enterprises that traditionally had greater access to technological resources. By automating repetitive customer interactions, companies can boost both operational efficiency and customer satisfaction.

Building an Agent Without Programming

A significant portion of the session was dedicated to Customer Experience Agent Studio—the visual development environment used for the demonstration. Users do not need to start from scratch; they can simply describe the purpose of their AI agent in one or two sentences. Then, Gemini automatically generates instructions, workflows, tools, and dialogue structure for the agent. Kim stated: 'We use AI to create an AI agent, not to write code.'

Developers can further customize the agent by defining its role, behavior, task flows, and decision-making logic, as well as adding optional knowledge sources to improve contextual understanding. The platform also supports numerous integrations, allowing agents to connect to Google Search, open APIs, Salesforce, ServiceNow, and custom functions.

Other features presented included cross-agent variables, customizable guardrails to block unwanted conversations, multilingual support, voice customization, model selection, and advanced dialogue parameters. To ensure production readiness, Agent Studio provides built-in evaluation tools. Companies can compare AI responses against expected outcomes using gold standard evaluations or simulate realistic customer interactions through scenarios, helping to identify edge cases before launch. These built-in testing tools allow teams to increase accuracy and reduce the risk of hallucinations or inconsistent responses.

AI for Small Teams

The session confirmed an important thesis for MSMEs: building AI-powered applications no longer requires large budgets, specialized engineering teams, or months of development. With platforms like Gemini Enterprise, companies are increasingly able to focus on defining customer problems rather than writing code. For MSMEs looking to improve customer interaction, automate routine support, and provide a faster, more personalized experience, AI agents are becoming a practical opportunity, not just an experimental technology. As the Google AI ecosystem evolves, opportunities for small businesses are becoming clearer: not just adopting AI, but using it as an amplifier that enables small teams to deliver enterprise-scale customer service experiences.

The Xiaohongshu Engine team presented the HELMSMAN system at the OSDI 2026 conference. This system is a cost-effective, high-performance solution for vector approximate nearest neighbor search, specifically designed for NVMe SSD arrays on fully flash memory.

Significant Infrastructure Optimization

HELMSMAN replaced Xiaohongshu's previous infrastructure, which consisted of approximately 35,000 CPU cores and 350 TB of DRAM across only about 40 fully flash servers. This achieved a reduction in hardware costs by over 90% while maintaining millisecond latency for company services such as search, recommendations, and advertising. These services process hundreds of billions of high-dimensional vectors with millions of queries per second.

Architectural Innovations

The core idea is that clustered ANNS methods are a more suitable architecture for modern high-speed SSD arrays compared to graph-based ANNS methods. Graph systems utilizing DRAM and SSD, such as DiskANN, suffer from sequential I/O dependency because reading from one SSD determines the next visited neighbor, which hinders SSD bandwidth saturation. Clustered systems, conversely, read candidate clusters in parallel, aligning with the parallel I/O capabilities of SSDs.

Overcoming Production Barriers

Traditional SPANN clustering systems faced three issues in production environments: insufficient throughput due to Linux I/O stack overhead, non-adaptive search strategies that either scanned too much or too little depending on query complexity, and slow single-processor index building that could not scale to hundreds of billions of vectors. HELMSMAN overcomes these obstacles through three innovations:

First, it features a specialized storage stack for ANNS that bypasses standard Linux I/O paths by managing NVMe queues directly in user space. This reduced software overhead, which previously consumed up to 58% of total latency. Second, a hierarchical learning mechanism for search pruning was implemented, which adaptively predicts the optimal number of clusters to scan based on query complexity, top-k size, and centroid distribution. Third, a distributed pipeline for index building with GPU acceleration was introduced, allowing indexes to be rebuilt at the scale of billions of vectors in just a few hours, matching the minute-level model updates required for recommendation and advertising systems.

Results and Industry Impact

Production tests demonstrated the system's efficiency. HELMSMAN achieves a 2–16 times improvement in throughput compared to existing DRAM-SSD systems, including DiskANN and SPANN, reaching up to 85% of pure in-memory deployment throughput while meeting SLA targets for average and tail latency within milliseconds. The ability to rebuild indexes at the scale of 10 billion vectors in hours makes HELMSMAN practical for production environments where embedding models and vector data are frequently updated. Xiaohongshu's deployment spans numerous business areas, including search, recommendations, advertising, content safety, and RAG pipelines.

The implications for cost and scalability extend beyond Xiaohongshu. Vector search infrastructure is one of the fastest-growing cost centers for internet platforms, as embedding retrieval is becoming universal for search, recommendations, and AI-powered features. Dependence on DRAM was a structural factor driving cost growth, given that vector scales double annually. HELMSMAN proves that high-performance ANNS on standard flash storage is viable in production even for the most latency-sensitive online serving scenarios, potentially changing the infrastructure economics for the entire industry. The research also raises questions about whether other memory-bound AI components can benefit similarly from co-designing the storage stack instead of simply expanding DRAM.