
AI-generated summary
Salesforce is signaling a shift in the enterprise software market from the traditional SaaS model of fixed UI operations to a new architecture that allows AI agents to directly access data and workflows.
The answer is Salesforce in Claude, a plugin with 37 skills for sales.
Using Salesforce in Claude, salespeople can inquire about the status of deals and pipelines, update records, and consider next actions from the Claude interaction screen. The 37 skills include pre-negotiation preparation, confirming the health of deals, and pipeline inspections. When set up for the first time, it reads the salesperson's work status from systems connected to Claude, such as Salesforce and Slack, and automatically configures a dashboard that shows the accounts and pipelines in charge.
The foundation is "Headless 360." It is an architecture that exposes Salesforce data, workflows, business logic, and permissions in a form that AI can directly call, and connects via the MCP (Model Context Protocol), a connection standard that connects AI models and external systems. It is said that no separate system integration work is required.
Controls are based on existing authorities and rules.
In terms of control, we will adopt a design that passes all operations performed by Claude to the Salesforce side. Both answers and operations are performed within the scope of existing Salesforce permissions and business rules, allowing users to see only the information and perform only the operations they are authorized to do.
Once an administrator connects, authentication and privileges are centrally managed, eliminating the need to configure settings for each user or rebuild the privilege model. For operations that involve writing, Claude can select controls such as requesting confirmation from the user before sending an email outside the company. The two companies will jointly work on a protection function called "Enterprise Frontier Safeguards," which will be rolled out in stages starting in fall 2026, and will combine "zero data retention," in which no data is retained, with detection of serious abuse.
The partnership also extends to incorporating Claude into the Salesforce product suite. Claude can be used as an inference model for Agentforce's inference engine ``Atlas Reasoning Engine,'' and runs ``Agentforce Vibes'' and ``Agentforce Coworker'' as standard.
Through Amazon Bedrock, it can also be used within Salesforce's security boundary, the Salesforce Trust Boundary. Claude is the standard model in Slack and supports assistant functions such as Slackbot. According to Salesforce, Slackbot, which incorporates Claude, has generated an annualized productivity improvement effect of 8.1 million hours within the company, more than doubling from the previous quarter.
With this announcement, Salesforce took a stand on the "death of SaaS" controversy. For decades, enterprise software has relied on users interacting with fixed UIs, but now that agents have direct access to data, workflows, and rules, software has become the system behind any interface. We take the position that even if the role of SaaS as a screen is reduced, the data and business rules behind it will continue to have value.
Salesforce in Claude is currently available to some pilot customers, and an open beta version is scheduled to begin in September 2026. Skills for departments other than sales are planned to be added from the second half of 2026.
AI outlook — possibilities, not facts
Open beta version to start in September 2026
Very likely · Within months
Addition of skills for departments other than sales from the second half of 2026
Likely · Within months

In response to the increasing number of 110 calls received, the National Police Agency has decided to develop an AI system that automatically converts call audio into text and analyzes the level of urgency. The aim is to quickly and accurately record the contents of reports and eliminate variations in interviews by responding police officers.

Samsung Electronics' Galaxy Z Fold8 and Huawei's HUAWEI Pura X Max are both horizontal folding smartphones, but they have different design philosophies. Samsung emphasizes thinness and lightness (201g, 9.7mm), while Huawei emphasizes the camera (3 eyes), battery (5300mAh), charging performance (wired 66W/wireless 50W), and waterproof performance (IP58/IP59). The weight difference of 28g makes a difference that can be felt in daily use.

Yuko Kawakami of IBM Japan explained that the company will promote collaboration between humans and digital workers (AI agents) starting in 2023, with over 4,000 digital workers working in over 450 project teams. It claims to have created a productivity improvement effect of $4.5 billion in 2025 by redesigning work flows and deploying it company-wide. He pointed out that in order to link the results of AI utilization to the management level, it is necessary to redesign both the business level and the infrastructure level.

The interactive distribution platform "Kustamie" utilizes AWS serverless architecture and generative AI to achieve low-latency distribution and real-time two-way communication for 60,000 people. A beta version is scheduled to start in the fall of 2026.
NEC announced on the 4th that it has started a demonstration experiment of a payment service using facial recognition at the employee cafeteria of SoftBank's headquarters. Users register their face photo and credit card in advance, and once their face is recognized by the in-store terminal, payment is completed in a few seconds. The demonstration period will run until November 30th, and is expected to involve more than 1,000 payments. The aim is to improve the purchasing experience of users by eliminating the complexity of cash register operations associated with the spread of cashless payments.
On September 6th, OpenAI simultaneously released the Research Acceleration Report and the essay "An Alien Mind" by Chief Scientist Jakub Pachoki. The report quantified the research efficiency of AI agents at 3.1 agent days compared to 1 human day, and Pachoki acknowledged that reliance on CoT monitoring is decreasing and predicted that future AI advances will be constrained by the reliability of monitoring. Additionally, incidents such as the suspension of container services due to a research infrastructure breach in July and the tightening of security restrictions on the Astra model were disclosed in detail.