読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
すべて技術製品投資株式市場その他

読書

2026-11-0585 投稿

更新

Professional Services Platform Certinia: Enterprise AI Agents Move from Orchestration to Context-Aware Delivery

クイックテイク
2026-11-05 05:55 GMT+8

What architectural problems must be solved for enterprise AI agents to truly take root in business processes? Professional services software company Certinia provided a specific answer in its coverage of the Dreamforce event in November 2026. Scott Hebner, Senior AI Analyst at theCUBE Research, identified a key shift from an industry perspective, while Raju Malhotra, Chief Product and Technology Officer of Certinia, described the product implementation path for this shift. The roles of the two differ: one provides external analytical judgment, while the other offers vendor product statements.

Hebner believes that the significance of Veda lies not only in introducing a suite of AI agents and intelligent workflow actions, but in demonstrating a possible direction for enterprise AI: AI collaborating with humans in critical workflows, helping people make better decisions, and earning trust by being grounded in business context, governance, and real operational outcomes. He specifically pointed out that this represents a move from orchestration to context-aware delivery—Certinia’s platform pushes Veda beyond simple task chaining toward governed actions that understand business context.

The System of Action described by Certinia adopts a three-layer architecture: the first layer aggregates structured records (project data, billable hours, utilization) and unstructured project data; the second layer leverages the company’s over 17 years of service operations experience to assist decision-making; the third layer enables agents to execute governed actions within existing workflows. The Moonnox platform, acquired in July 2026, added a context layer to Veda, capturing unstructured information such as emails, chats, and calls that actually occur during delivery into a context graph through knowledge graphs, persistent project memory, and service-specific agents.

Malhotra stated that Veda expanded from 10 agent packages at its April launch to 24, and will exceed 40 by year-end. The System of Action released in September expanded the library of intelligent actions from 64 to 135. These figures represent progress reported by the vendor and have not been independently verified.

The reference value of this architectural approach lies in breaking down the vague concept of enterprise AI agents into checkable components—where context comes from, how rules constrain actions, and what decision-making power humans retain. Robert Cesafsky, Chief Operating Officer of Certinia, described Moonnox as an intake machine for structured data, emphasizing its function of importing information generated during the delivery process into the context graph.

Boundaries need to be clarified: the entire report comes from a paid media partnership series between SiliconANGLE and Certinia, and Hebner’s comments come from the research department of the same media family. All statements regarding Veda’s capabilities and customer value are attributed to the vendor and do not constitute independent verification.

ソース:SiliconANGLE

リサーチ

Cisco Network Platform Executive: AI Agents Executing Network Changes Must Be Deterministic, Not Improvisational

トピック · 思科AI网络确定性框架クイックテイク
2026-11-03 23:04 GMT+8

Network operations are facing a scale problem: the volume of alerts has exceeded human processing capacity. Joe Vaccaro, Senior Vice President and General Manager at Cisco's Network Platform and ThousandEyes, admitted on the Targeting AI podcast that if humans alone were responsible for resolving these alerts, companies simply could not hire enough people. This pressure is driving enterprises to increasingly trust AI agents to execute network changes directly, bypassing the need for step-by-step manual approval.

To address this, Cisco designed a four-stage framework: sensing issues, diagnosing them using cross-domain telemetry, planning remediation, and executing changes. While AI is utilized across all four stages, the execution stage faces a hard constraint—networks require deterministic results, not probabilistic ones. Vaccaro explained that when reaching the execution stage, where changes are actually applied across the network, deterministic execution is essential. Cisco's approach is to use predefined workflows at this stage rather than allowing AI agents to improvise. This means that while AI can decide what needs to be done, how it is done must follow a predefined, fixed process.

ソース:AI Business

リサーチ

Together AI Unveils 'Polymorphic Inference' Framework, Claims Compiler Already Optimizing Production Models

トピック · Together AI重大
2026-11-04

AI infrastructure company Together AI published a blog post on November 4, 2026, formally introducing a new paradigm called "Polymorphic Inference" and announcing the deployment of its tool, Together Compiler, in parts of its production environment. The tool uses AI agents to automatically optimize inference performance, aiming to address the challenges posed by the rapid iteration of open-source model architectures, traffic patterns, and hardware platforms.

The research team at Together AI argues that the traditional "Static Inference" model relies on manually tuned kernels for specific configurations, which struggles to keep up with the current pace of open-source model releases, increasing architectural diversity, and frequent hardware updates. According to the company, 80% of traffic on OpenRouter goes to models released in the last three months, and the top ten most popular open-source models all use different architectures. The core of Polymorphic Inference is enabling systems to dynamically adapt to these changes rather than writing new static kernels for every scenario.

As a first step toward this goal, Together AI launched Together Compiler. Described as a "recursive self-improvement" (RSI) system, it is an AI-agent-driven performance optimization loop. The company claims that many models currently running through its endpoints are fully optimized by AI agents, calling it "the world's first production-scale recursive self-improvement (RSI) inference system." Together Compiler reframes the traditional compiler translation problem as an agentic search problem, using a swarm of agents paired with a high-fidelity persistent engine to automatically identify key optimization patterns and continuously adjust code to fit new traffic shapes and workloads.

It is important to note that the claims regarding being the "world's first" and the status of production deployment are self-reported by the vendor, with no independent benchmark data or specific performance improvement percentages provided. Together Compiler is currently available only to limited beta users. For readers tracking AI infrastructure efficiency, this move indicates that leading service providers are attempting to reduce reliance on manual kernel optimization through automation, though actual effectiveness remains subject to future independent verification.

ソース:together.ai

リサーチ

Render Removes 30-Day Limit on AI Sandboxes, Adds State-Preserving Durable Workflows

トピック · Renderクイックテイク
2026-11-05 02:04 GMT+8

Cloud infrastructure provider Render announced on November 4, 2026, that it has removed the previous 30-day continuous runtime limit on AI sandboxes within its paid workspace plans. This update allows Render's environments to run indefinitely, suspending when idle and resuming upon new commands while keeping files, memory, and running processes intact.

The launch also introduces "Durable Workflows," enabling developers to write long-running jobs in Python or TypeScript. If a process crashes, the workflow can resume on a new machine using saved step results, ensuring that successful model calls, payments, and database writes are not repeated. Additionally, new Shared Filesystems allow sandboxes and workflow tasks to access the same persistent files.

Render founder and CEO Anurag Goel stated that the design aims to differentiate from competitors like Vercel. According to The New Stack, Vercel currently caps uninterrupted individual sandbox sessions at 24 hours, and restarting them causes running processes to begin anew. Goel noted that maintaining state continuity is critical for long-duration agent tasks such as multi-day codebase migrations or full application testing.

Despite the new features, some capabilities remain in early stages. Credential brokering is initially available only to selected customers, private networking for sandboxes is still under development, and the mechanism for handling concurrent writes to the same file has not yet been finalized. Vercel had not provided a statement by the time of publication.

ソース:The New Stack

リサーチ

Independent Developer Tw93: In the AI Era, What Is Truly Reusable Across Platforms Is Not Code, But Rules

トピック · 跨端开发规则复用重大
2026-11-04 20:25 GMT+8

Tw93 is an author at Shao Shu Pai (sspai), having developed several desktop tools including Mole, Kaku, Pake, and MiaoYan. In November 2026, he published a long-form article detailing his entire process of developing Windows software on a Mac. His core conclusion is that in today's era where AI writing code is no longer a bottleneck, what truly needs to be reused in cross-platform development is not code, but rules.

This judgment stems from a specific engineering fact. The Mac version of Mole was written in Swift and SwiftUI, comprising over 100,000 lines of code. Tw93 initially intended to reuse as much of this code as possible for the Windows version. However, after analyzing it with AI, he found that less than 20% of the code could function independently of macOS system frameworks like AppKit and SwiftUI. The remaining logic heavily relied on macOS-unique path rules such as ~/Library, making it largely unusable. Consequently, the Windows version was written almost entirely from scratch.

Since the code itself could not be reused, how did he maintain consistency between the two platforms? Tw93's approach was to shift the synchronization layer upward to design rules and decision records. He took the design tokens—colors, font sizes, spacing, and corner radii—written in Swift for the Mac version and used a script to directly generate them as XAML resources for Windows. Copywriting in 11 languages was converted from the Mac's multi-language file format to the Windows format. Even the runner animation in the menu bar was synchronized using the same script. On the functional level, he maintained a SYNC.md file as a reconciliation ledger. For every new feature added or security issue fixed in the Mac version, the Windows side clearly marked which items were implemented, which were intentionally different, which were inapplicable, and why those judgments were made. Each synchronization also recorded the corresponding Mac commit to prevent the two versions from diverging if the AI went off track.

He summarized this experience into a judgment: As times change, so-called code abstraction becomes less important, whereas rule abstraction becomes more critical. Design, copywriting, and various rules established after encountering pitfalls constitute the 'capability abstraction layer' that AI is truly adept at. Underlying code can be written separately for each platform without forcing a compatibility layer to smooth things out, which would only add complexity. Instead, what can be reused are the designs, copywriting, and rules.

The feasibility of this method relies on the premise that AI has drastically reduced the efficiency costs of native development. Tw93's workflow kept all code and AI context on the Mac, using Claude Code for coding. Windows was used solely for compilation and verification, running via Parallels Desktop virtual machine, with the repository mounted as a shared folder. For scenarios where the VM could not accurately test (such as high-DPI screens, battery usage, and SmartScreen blocking), he verified them on a physical ThinkPad X1 Carbon. This machine was set to never sleep, allowing the agent to run regression tests 24 hours a day. Ultimately, he wrote over 80,000 lines of Windows code and created more than 4,000 tests in one month.

ソース:少数派 - 高品质数字消费指南

リサーチ

Blockchain Analytics Firm Chainalysis Discloses Breach via 'Offensive Agentic Techniques', Suspends Investigation Tools

トピック · Chainalysis重大
2026-11-05 08:00 GMT+8

US-based blockchain analytics firm Chainalysis disclosed a security incident on October 30, stating that external attackers utilized what the company calls "Offensive Agentic Techniques" to gain unauthorized access to internal systems and tools. This access resulted in the exposure of some customer user identifiers, names, email addresses, and organizational role information. Chainalysis serves government law enforcement agencies, financial institutions, and cryptocurrency exchanges, assisting in tracking transactions to investigate fraud, ransomware, and money laundering.

According to iThome, in an update posted on November 2, Chainalysis announced it had proactively suspended certain product services, including its blockchain investigation tool Reactor and data service product Data Solutions, pending further third-party security assessments. However, recognizing the importance of transaction monitoring products for customers detecting money laundering activities, Chainalysis decided to maintain these compliance services while continuing to monitor systems for anomalies. The company has not yet published the number of affected customers or a timeline for restoring the suspended services.

Security media outlet Bleeping Computer noted that "Agentic" typically refers to AI systems executing multi-step tasks with limited human intervention, though Chainalysis has not disclosed the specific tools or methods used by the attackers. Help Net Security analyzed that the leaked names and job titles could enable attackers to identify personnel at specific organizations and launch targeted social engineering attacks. This incident highlights the emerging risk posed by AI-driven automated attack vectors to critical financial infrastructure.

ソース:iThome

リサーチ

Wiz Investigation: Attackers Stole AI Platform Keys, Made Millions of Bedrock API Calls in Hours

トピック · Wiz重大
2026-11-05 08:00 GMT+8

Cloud security company Wiz recently released an investigation report indicating that during several cybersecurity incidents targeting AI development and deployment platforms in October, attackers successfully accessed sensitive credentials stored in customer environment variables. These environment variables often contain API keys for AI services, allowing attackers to use them for unauthorized calls once the platform was compromised.

Two cases of enterprise AI service key abuse tracked by Wiz showed that within less than 30 minutes of the platform provider reporting credential access, attackers began testing key validity, followed by large-scale calls to AWS's AI model service, Amazon Bedrock, days later. In one case, attackers used stolen keys to make approximately 800,000 API calls within 30 hours; in another, about 1.7 million calls were made within 13 hours. Researchers also linked some stolen keys to resale activities involving groups named momo and Xingjiabi.

The incidents involved two major platforms. Taiwan-based cloud deployment platform Zeabur was breached in mid-October, with attackers reading over 3.21 million environment variables and about 320,000 account records. Zeabur subsequently announced it would cease operations by December 31. Another cloud deployment platform, Northflank, also experienced a security incident in late October, with users reporting unauthorized use of their AI service API keys stored on the platform.

Northflank stated on November 2 that preliminary investigations suggest the attack activity may have focused on sensitive information stored in its control plane, with no signs yet of attackers accessing customers' running workloads or related data. The company has revoked some access tokens, terminated related sessions, and required all customers to immediately rotate sensitive credentials, including AI service API keys, cloud service credentials, private keys, and database passwords.

ソース:iThome

リサーチ

'Godfather of AI' Hinton Warns: Tech Giants Ignore Risks for Profit; Catastrophe May Trigger Regulation

トピック · 辛顿AI安全警告重大
2026-11-05 00:53 GMT+8

On November 4, 2026, TechRadar reported that Nobel Prize-winning computer scientist Geoffrey Hinton, widely considered one of the 'Godfathers of AI', expressed deep concerns about the future of artificial intelligence in an interview with The Guardian. While acknowledging AI's potential to create a better future, he warned that failure to manage risks could lead to a huge gap between rich and poor, massive unemployment, and society being torn apart.

Hinton specifically criticized the competitive dynamics within the AI industry. He stated that key figures such as Meta CEO Mark Zuckerberg, Tesla CEO Elon Musk, and OpenAI CEO Sam Altman understand the risks involved but prioritize profit over safety. "I think Zuckerberg and Musk and Sam Altman, they understand the risks, and they don’t give a shit. They’d rather have the money. That’s my belief," Hinton said.

Regarding regulation, Hinton proposed a troubling view: it may take a serious AI-related catastrophe—such as an attack on global financial systems or massive power brownouts—to prompt governments to implement effective oversight. He compared this potential turning point to the partial nuclear meltdown at the Three Mile Island power station in 1979, which led to significant reforms in US nuclear regulation.

Despite his pessimism, Hinton clarified that he is not an 'AI doomer'. "My gut is that we’re still in control," he added. "We might well be able to do something. It’ll only work if we put a big effort into it, but we might well put a big effort into it." He called for international collaboration to better understand and mitigate AI risks.

ソース:TechRadar

リサーチ

Sierra Club warns US data center expansion could drive 42% surge in power-sector emissions by 2030

トピック · 塞拉俱乐部数据中心警告重大
2026-11-05 06:10 GMT+8

According to TechRadar, the US environmental organization Sierra Club has released a new assessment warning that planned data centers supporting artificial intelligence infrastructure could lead to a sharp rise in carbon dioxide emissions from the American power sector. The group estimates that CO2 emissions linked to this expansion could exceed 460 million metric tons by 2030, a figure equivalent to Mexico's total annual carbon dioxide emissions.

The report indicates that such growth would push US power-sector emissions 42% above 2025 levels, effectively reversing fifteen years of progress in greenhouse gas reduction. Currently, approximately 40% of American data centers rely on gas turbines for electricity, while renewable energy additions are struggling to keep pace with rapidly increasing consumption.

Beyond environmental concerns, supply chain bottlenecks add further uncertainty. Gas turbine prices have risen by 50%, and manufacturers face production backlogs extending several years, with some equipment reservations reportedly scheduled for delivery in 2032. These constraints may encourage developers to retain fossil fuel generation rather than wait for cleaner alternatives. Additionally, financial uncertainty among hyperscalers regarding investment returns could lead to postponed or abandoned projects, leaving the ultimate environmental impact variable.

ソース:TechRadar

リサーチ

South Korean Camera Module Maker Camus Begins Mass Production for Samsung’s ‘Bespoke AI Family Hub’ Refrigerators

クイックテイク
2026-11-04 13:45 GMT+8

South Korean camera module manufacturer Camus announced on November 4 that it has begun supplying camera modules for all models of Samsung Electronics’ “Bespoke AI Family Hub” refrigerator series. This marks the company’s first mass-production revenue generated in the home appliance sector.

Previously known primarily as a supplier of camera modules for Samsung smartphones, this move signifies Camus expanding its visual hardware capabilities from mobile devices into consumer appliances. The company stated plans to broaden its supply scope to other home appliances through further cooperation with Samsung’s Digital Appliances (DA) division.

Beyond home appliances, Camus is actively expanding into non-mobile sectors such as robotics and security. Recently, the company entered the camera supply chain for Hyundai Motor Group’s humanoid robot “Atlas” and is pursuing supplies for facial recognition cameras in the security field.

ソース:미래를 보는 창 - 전자신문

リサーチ

NVIDIA Internal Practice: Five Key Lessons for AI Agent Operations of GPU Clusters

トピック · 英伟达AI代理运维实践重大
2026-11-05 06:00 GMT+8

As GPU cluster scale expands, operations span multiple layers including compute, network, hardware, and scheduling. When training task performance degrades, SREs often need to open different dashboards one by one to check scheduler status, GPU temperature, network latency, and application logs separately before locating the root cause. Kavya G Y, a product manager at NVIDIA, and Rohan Adwankar, an engineer responsible for agentic systems and observability, recorded in a technical blog how they used a set of AI agents to solve this repetitive labor problem.

This system is called the AI Factory Operations Agent, running on NVIDIA's internal DGX H100, DGX B200, and DGX GB300 production clusters. The core idea is not to use a single all-powerful agent to replace SREs, but to let a group of domain-specific agents perform their respective duties, communicating with each other through the Model Context Protocol (MCP). SREs can ask any agent questions in natural language, switch investigation directions, and every step of the operation has audit records.

NVIDIA deployed six domain agents. The Job Recovery Agent detects training job failures and goodput degradation, performing multi-layer root cause analysis across the scheduler, application, hardware, and network layers, and recommending recovery steps once the root cause is confirmed. The Hardware Agent triages hardware faults at the device level—PCIe errors, ECC errors, GPU issues, NVMe failures, and thermal anomalies—tracing symptoms down to specific components. The Observability Agent queries historical Prometheus metrics and Grafana Loki logs, dynamically generating dashboards that are displayed directly in the web interface. The Research Agent retrieves pre-loaded NVIDIA Mission Control manuals and DGX SuperPOD reference architecture documents, and also supports users uploading their own runbooks. The Kubernetes Agent provides read-only cluster state awareness, covering pods, services, deployments, GPU resources, node status, and scheduling status. The Cluster Management Agent interfaces with the NVIDIA Base Command Manager API through diagnostic MCP tools for cluster management advice and troubleshooting.

The center of the architecture is the user-side agent runtime provided by the NemoClaw blueprint, which receives SRE queries and routes them to the appropriate domain agents. OpenShell is responsible for governance and control. Each shell execution takes place in an independent Kubernetes sandbox rather than running inside the main gateway pod; the sandbox lifecycle does not modify existing services or production workloads. Security is layered: Kubernetes checks invoke a fixed version of the kubectl binary via the server-side run_kubectl tool using registered read-only credentials, and OpenShell itself does not receive kubectl or Kubernetes credentials. RBAC explicitly limits the scope of resources agents can view—the article shows actual permission verification results: it can get pods and configmaps, but cannot get secrets, create pods, or update pods.

From production usage, NVIDIA distilled five lessons. First, clear agent boundaries: overlapping responsibility ownership creates noise rather than signal, so each agent is only responsible for its own domain. Second, read-only first: accurate investigation capabilities build SRE trust before automation, allowing them to see what the agent can and cannot do first. Third, cross-layer correlation: the framework connects workload, hardware, and network signals to find root causes missed by isolated tools—this is exactly the core pain point of multi-dashboard troubleshooting. Fourth, precise MCP tool definitions: clear tool descriptions improve routing and reduce wrong tool selection. Fifth, transparent reasoning: SREs need to see tool calls, evidence, and conclusions before taking action.

ソース:NVIDIA Technical Blog

リサーチ

Uber says only 4 rides booked by AI agents during busiest week

トピック · AI代理消费级落地重大
2026-11-05 05:56 GMT+8

US ride-hailing and delivery giant Uber disclosed during its third-quarter earnings call on Wednesday that only four rides were booked through an AI agent during its busiest week in October, which coincided with Halloween.

Chief Financial Officer Balaji Krishnamurthy stated that the company handled approximately 320 million trips and orders across Uber and Uber Eats during that week, but only 500 Uber Eats orders were placed by an AI agent.

CEO Dara Khosrowshahi acknowledged that the traction from AI agents is "very, very early and very limited," though he remains bullish on the potential for agents to reach new users.

These figures stand in stark contrast to the growing hype surrounding personal AI agents capable of handling tasks like shopping and booking, suggesting that actual transactional adoption on major consumer platforms remains negligible.

ソース:TechCrunch

リサーチ
次へ →
ログイン / 新規登録
読書リサーチレーダー投資フレームワーク

He also adjusted the architecture to support this: Mole for Windows is divided into four layers. The bottom Core layer depends only on .NET and does not touch any Windows interfaces; cleanup rules and path security checks are placed here, using a simulated memory file system based on Windows path rules for testing, allowing it to run well on the Mac. Above that is the Platform layer, which houses registry access, file systems, and Win32 calls; with EnableWindowsTargeting turned on, this layer can also compile on the Mac. Next is the ViewModel layer, which corresponds one-to-one with the Mac version. Only the top WinUI interface layer must be compiled on Windows. The lower the layer, the less it touches Windows, and the more code can be verified on the Mac.

However, it should be noted that Tw93 declared a direct conflict of interest with the products mentioned in the article at the beginning. His conclusions naturally lean toward validating his choice of native development. Furthermore, the premise of writing 80,000 lines of code in a month is that he is proficient in both platforms, has years of client-side development experience, and deeply uses AI coding tools. For teams lacking these conditions, or for scenarios where business logic is highly shared (such as Kotlin Multiplatform solutions), cross-platform frameworks remain a more pragmatic choice. His judgment is more applicable to independent developers or small teams pursuing the best experience on each platform with AI assistance.

NVIDIA continues to refine this framework by measuring root cause analysis accuracy, tracking SRE agreement, and expanding the agent ecosystem. For large-scale AI factories, the recommended deployment model is to place agent orchestration and reasoning on a dedicated GPU-accelerated control plane isolated from the production cluster, while the production cluster only provides telemetry data, does not consume training GPUs, and does not interfere with workloads. Internal deployments run model inference on dedicated GPU capacity, using existing control-plane CPU Kubernetes admin space to host runtime services. Smaller scales can use RTX PRO Workstations or DGX Spark.

In multi-tenant scenarios, scoped agents, tools, sandboxes, and Kubernetes permissions control the cluster data, tools, and nodes accessible to each tenant. Tenant SREs can only see their own namespaces, jobs, and nodes, while central operations retain broader cluster visibility. Each tenant can use its own domain agents and dedicated inference hardware. OpenShell enforces these boundaries by running each tenant's agents and tools in isolated sandboxes with kernel-level restrictions.

This orchestration framework is currently released as experimental open source software on GitHub; APIs, configuration, and deployment requirements may change between versions. NVIDIA customers can contact their account managers to obtain internally developed cluster operations agents.