AI Integration

C Cloudflare has introduced a new tool called the Answer Engine Optimization (AEO) Visibility Dashboard. This dashboard is part of Cloudflare’s AEO Suite and allows website owners to see how their business is being recommended by AI assistants when customers ask relevant questions.

The AEO Visibility Dashboard joins Agent Readiness, another tool in the suite that checks if AI systems can find and read a site in the first place. Together, these tools provide marketers with the data they need to understand and optimize their content for AI-driven discovery channels.

AI assistants have become a primary means of consumer discovery, but brands had no way to measure how well they were performing. There was no signal to tell a brand whether it was recommended, briefly mentioned, or skipped entirely when an AI assistant provided a name in response to a customer’s query.

For too long, businesses relied on sending test prompts to AI chatbots and sampling the responses. However, this approach has its drawbacks – it’s prone to inconsistency if not paired with other data signals, and it can only provide insights on a small scale.

Cloudflare sits at the network layer between AI platforms and websites, giving it a unique advantage. By observing actual crawl and referral activity directly across millions of sites, Cloudflare creates deeper and more transparent insights than sampling test prompts alone.

‘Being discoverable used to mean ranking on a page,’ said Stephanie Cohen, Chief Strategy Officer at Cloudflare. ‘That’s not enough anymore.’ The agentic era has arrived, where AI assistants are providing recommendations and answers to customer queries without most brands even knowing if they’re in the conversation.

Cloudflare observes actual signals: real crawl activity, real referrals, what AI systems are genuinely doing across millions of sites. This is what powers its tools – giving brands a clear view into how their content is being used and providing control over what they choose to share.

The AEO Visibility Dashboard provides four key metrics for marketers to understand and optimize their content: Citation Rate, Mention Rate, Prominence, and Share of Voice.

Citation Rate shows which AI platforms are pulling from a brand’s site as a trusted source. Marketers can direct budget toward testing and optimizing the content that actually drives AI recommendations – not just throwing money at sites that don’t work.

Mention Rate reveals whether AI assistants are naming a brand even when they don’t cite its site. A brand that gets mentioned but not cited has an authority problem, not an awareness issue. This distinction changes strategy entirely.

Prominence tracks how much of an AI answer is actually attributed to a brand and where in the response it appears. A passing mention at the end of a long answer isn’t the same as a strong recommendation – marketers get a number that reflects this difference.

Share of Voice shows how a brand stacks up against competitors across specific questions its customers are asking AI assistants, whether that position is improving or eroding over time.

The AEO Visibility Dashboard joins Agent Readiness in providing businesses with data analysis tools to optimize their content for AI-driven discovery channels. This represents a major shift forward in helping brands understand and work with the power of AI assistants.

Cloudflare built its AEO Suite on a foundational commitment: giving site owners visibility into how their content is being used, and control over what they choose to share. Whether it’s protecting and monetizing content or maximizing discovery in the agentic era, Cloudflare’s tools are designed to serve those interests directly.

Citation Rate helps marketers see which AI platforms rely on a brand’s trusted source material. They can then allocate their budget more effectively – optimizing the content that drives real results, rather than just throwing money at sites with no impact.

The AEO Visibility Dashboard and Agent Readiness work together to provide businesses with insights into how well they’re being recommended by AI assistants. This marks an important step forward in empowering brands to make data-driven decisions about their online presence.

Citation Rate is not the only metric provided by the AEO Visibility Dashboard. Mention Rate reveals whether a brand gets mentioned even when its site isn’t cited, which can have significant implications for strategy and budget allocation.

Prominence tracks how much of an AI response is attributed to a specific brand or website – providing marketers with valuable insights into what works and what doesn’t in the agentic era.

The AEO Visibility Dashboard also shows Share of Voice, which allows businesses to see how they stack up against competitors across specific questions their customers are asking AI assistants. This information can be used to inform content optimization decisions and maximize discovery potential.

Citation Rate is just one part of the equation – but it’s a crucial component in helping brands understand how well their content is being received by AI assistants.

GTT Communications, a leading provider of networking and security services to multinational organizations, has expanded its API access across GTT Envision. This move aims to simplify the integration of GTT’s capabilities into existing systems and workflows for enterprise and wholesale customers as well as Technology Services Distributor (TSD) partners.

The enhanced APIs adhere to REST standards and build on GTT’s existing API capabilities. They offer improved service assurance ticketing for all GTT products and services, along with two quoting APIs designed specifically for high-volume connectivity products such as broadband, 5G Fixed Wireless Access, and dedicated internet access (DIA).

These new APIs are tailored to meet the needs of enterprises, wholesale providers, and TSDs. They require automation and management tools that can run infrastructure operations at accelerated speeds to quickly adapt to changing business drivers and needs.

According to Tom Major, Senior Vice President Product Management at GTT, ‘Enterprises, wholesale providers, and TSDs need automation and management tools that help run their infrastructure operations at accelerated speeds.’ The enhanced GTT Envision APIs provide enterprise customers with simplified global incident and service request management. Wholesale providers and TSDs receive real-time access to an AI-enhanced quoting tool.

Gartner notes that network automation adoption in the enterprise lags behind server automation, as more than 60% of enterprise networking activities are performed manually. Limited network automation creates bottlenecks in provisioning and incident resolution while increasing human errors.

Drawing on over two decades of experience integrating and managing global networks across six continents, GTT designed these new API capabilities to support the automation of networking management across multiple vendors and service types.

The new GTT Envision APIs include a Service Assurance Ticketing API for simplified global incident and service request management. This allows customers to create, read, update, and monitor incident and service request tickets directly from within their existing systems. The API offers complete ticket visibility, built-in governance, and real-time response capabilities through webhooks.

The Service Assurance Ticketing API supports all GTT services, including managed networking, security, cloud, and professional services. It helps customers streamline their operations by automating incident and service request management across multiple vendors and service types.

GTT has also introduced two quoting APIs for TSDs and wholesale markets. These APIs enable instant quote creation and configuration while keeping information continuously synchronized between the partners’ systems and GTT Envision.

The GenAI-powered address validation from GTT EnvisionDX offers accurate and confirmed pricing across 11 million serviceable business addresses and site options from more than 3,000 connectivity partners. This feature is particularly beneficial for TSD quotes for broadband, 5G Fixed Wireless Access, and DIA as well as wholesale partner quotes for these services.

Developers can access comprehensive documentation, resources, and setup guides at the GTT Developers Hub to begin integrating GTT Envision capabilities with existing business system workflows. This move aims to simplify the integration process and make it easier for developers to leverage GTT’s AI-powered tools for their businesses.

A new feature has been added to Amazon Bedrock, a comprehensive data analysis tool that helps developers build and deploy AI models. The feature is called Web Search, and it allows foundation models to access current web knowledge in real-time. This means that when a model needs information about recent events or updates, it can now retrieve this information directly from the web without requiring manual intervention.

When building AI-powered applications, one of the biggest challenges developers face is grounding their models in up-to-date knowledge. Foundation models are trained on vast amounts of data but often lack current information, leading to inaccuracies and hallucinations. Web Search addresses this issue by providing a seamless way for models to access web-based knowledge without requiring manual integration or maintenance.

The traditional approach to connecting foundation models with web knowledge involves identifying, integrating, and maintaining third-party Web Search providers. This process can be time-consuming and introduces data residency risks and operational overheads. With the introduction of Web Search on Amazon Bedrock, developers no longer need to worry about these complexities, as it becomes a native capability of the platform.

Web Search is designed specifically for Amazon Bedrock model inference and offers several key differentiators. Firstly, its multi-source grounding approach combines a web index operated by Amazon with a built-in knowledge graph that anchors entities in a domain along with their connections. This enables Web Search to answer factual questions with strong confidence rather than relying on the model’s inference capabilities.

Another significant advantage of Web Search is its context-efficient retrieval mechanism. Rather than handing the model raw pages and expecting it to find relevant information, Web Search performs semantic snippet extraction – pulling passages from web pages that bear on the query and returning them in a form optimized for the model’s context window. This reduces the number of tokens spent on boilerplate and speeds up retrieval.

Enabling Web Search is straightforward and requires only three steps: configuring AWS credentials, pointing the OpenAI client at the bedrock-mantle endpoint, and adding the Web Search tool to the request. The Responses API supports built-in tools natively, so no function schema needs to be defined or a client-side loop managed.

Once enabled, Web Search becomes a single parameter in existing OpenAI-compatible API calls, removing the need for vendor onboarding, API keys, orchestration layers, and separate SDKs. This simplifies development and reduces operational overheads. Additionally, Web Search operates entirely within Amazon Bedrock’s infrastructure, supporting customers’ compliance requirements.

When using Web Search, developers can choose between two retrieval modes: indexed-web or live-web. Indexed-web is currently available, while live-web will be enabled in a future update. The default mode requires the bedrock-websearch:ExternalWebAccess permission, which grants access to Amazon’s pre-indexed web corpus.

The API returns grounded responses with structured citation annotations, including URL and page title for each referenced source. This enables developers to render inline footnotes or highlight exact spans supported by citations. Web Search is integrated with AWS CloudTrail out of the box, providing a comprehensive audit trail without exposing query text or raw page content retrieved during searches.

The integration with CloudTrail captures calling identity, timestamp, action, and account context for each request. This provides security and compliance teams with a clear view of who used the tool when, without revealing what end users searched for. The lack of exposure to sensitive information is particularly beneficial in environments where data residency requirements are stringent.

Web Search on Amazon Bedrock removes the undifferentiated heavy lifting associated with connecting foundation models to up-to-date web knowledge. It delivers context-efficient, multi-source grounded results with low-latency and simple enablement through a single API parameter – allowing developers to add web grounding without managing vendors or compliance reviews.

The feature is generally available in US regions (us-east-1, us-east-2, and us-west-2) for pricing details. To get started, see the Web Search documentation for complete API references and examples.

A major shift in how businesses manage their marketing efforts has occurred. EZ Texting, a trusted platform used by over 230,000 customers, has introduced its Model Context Protocol (MCP) servers. This innovation lets companies run entire SMS marketing campaigns through natural conversation with AI tools they already use.

The MCP servers allow business owners to interact with their AI assistants just like they would with a human colleague. They can tell Claude or ChatGPT to send a broadcast, check campaign performance, or look up contacts - and it happens instantly without needing separate apps or logging into dashboards.

EZ Texting’s integration of MCP servers is bridging the gap between AI assistants and small- to medium-sized businesses. Until now, these platforms have been mostly inaccessible to non-tech-savvy entrepreneurs who don’t have in-house developers on hand. With this launch, EZ Texting makes it possible for shop owners to use their existing tools without needing extensive technical expertise.

The shift towards relying on AI assistants as a primary means of getting things done has left many small businesses behind. However, with the introduction of MCP servers, these companies can now automate and streamline their marketing efforts using familiar interfaces like ChatGPT or Claude. This is a significant milestone in making SMS marketing more accessible to a broader range of users.

Biznesy can use EZ Texting’s MCP servers for various tasks, including sending texts without opening the app, getting instant answers about campaign performance, managing contacts through voice commands, automating workflows without coding, and staying on top of their account details. Every action runs through EZ Texting’s existing platform, ensuring that permissions, credits, and compliance rules remain unchanged.

The MCP servers connect to any MCP-compatible AI tool, including Cursor, allowing businesses to chain SMS actions together with other AI-connected tools in their stack without writing a line of code. This integration enables seamless automation and streamlines the marketing process for small- to medium-sized enterprises.

EZ Texting’s CEO Vijesh Mehta emphasizes that ‘AI is quickly becoming how people get things done,’ and his company wants to ensure that this shift doesn’t bypass small businesses. With MCP servers, business owners can ask their AI assistants to perform tasks without needing a developer or new tools - it happens exactly as fast as they think.

The EZ Texting MCP servers are available now at no additional cost beyond existing plans. Customers can get started by visiting answers.eztexting.com/s/article/mcp-getting-started, while developers can learn more about the technology and its applications on mcp.eztexting.com.

A growing trend in business is to automate processes using artificial intelligence (AI). While this technology can accelerate workflows and save time, it’s essential for organizations to reconsider their existing systems before implementing new tools. A recent company-wide meeting highlighted a crucial point: just because we have the ability to automate something doesn’t mean we should.

The CEO shared an anecdote about the introduction of electricity in factories built for steam power. By the end of the 19th century, many manufacturers relied on large central engines that powered machines through complex systems of belts, pulleys, and shafts. When electric motors became available, they replaced these central power sources without changing the factory’s layout or workflows. This example illustrates how new technology can be integrated into existing systems without challenging underlying assumptions.

This phenomenon is not unique to industrial settings; it also occurs in organizations adopting AI. Many companies start by implementing sensible and useful applications, such as summarizing meetings, drafting emails, preparing reports, organizing research, or creating first drafts of presentations. These tools can save significant time across everyday work, but they often amount to installing a better motor inside a factory whose basic design has not changed.

People rarely begin by questioning an entire system; we adapt new technology to the processes we already know. This is understandable, as it’s easier to build upon existing infrastructure rather than redesigning from scratch. However, this approach can lead to missed opportunities for improvement and innovation.

Across organizations, processes often start as thoughtful solutions to real problems but eventually become outdated or inefficient over time. Familiar examples include approval chains that have outlived their original purpose, information entered into multiple systems, or reports produced mainly because they’ve always been done this way. A document labeled ‘Final_v7_FINAL2_UseThisOne’ might be found on a shared drive, demonstrating how easily temporary workarounds become permanent infrastructure.

Before using AI to accelerate processes, leaders should ask fundamental questions: would we design this process in the same way if we were starting today? If information can be gathered and analyzed almost instantly, does a decision still need to move through the same chain? If a first draft takes minutes rather than hours, where should human expertise enter the process?

These are not just technical considerations but also questions about how work is organized. For People and HR leaders, they create a responsibility that goes beyond teaching employees how to use new tools. Training matters, but adoption alone will not redesign work. Organizations need to examine how responsibilities are divided, where decisions are made, and where human judgment creates the most value.

Efficiency can hide deeper design problems. Cutting time required to complete tasks does not necessarily make surrounding processes more valuable. A faster approval chain may still contain unnecessary approvals, a better reporting tool might produce reports no one uses, or an AI-generated first draft could save employees’ hours while leaving unclear who is expected to challenge, improve, or take responsibility for the final decision.

Many processes exist for good reasons, including accountability, safety, regulatory requirements, and human judgment. Revisiting them does not mean discarding those protections; it means giving leaders and employees permission to say ‘Maybe the process isn’t sacred’ and then test which parts remain essential and which survive mainly because they are familiar.

History rarely rewards organizations simply for adopting new technology; it rewards those willing to let that technology challenge assumptions once made perfect sense. As AI becomes more capable and deeply embedded in everyday work, organizations will benefit most by using it to question the design of work itself. They will decide process by process which parts of yesterday’s factory are still worth keeping.

Sivan Avihud, Vice President and Global Head of People at ERGO NEXT Insurance, emphasizes that adopting AI tools for businesses is not just about automating tasks but also about reevaluating existing processes to ensure they remain relevant in today’s landscape. By doing so, organizations can unlock the true potential of AI technology.

Dynamics 365’s latest offering, the Customer Experience MCP Server for Service, is now generally available. This server enables organizations to access customer service data and tools from AI experiences that are compatible with Microsoft’s Model Context Protocol (MCP). By leveraging this standardized approach, businesses can streamline their operations without having to develop separate integrations for individual AI applications.

The MCP Server implements the Model Context Protocol, an emerging open standard for connecting AI applications to enterprise tools and data. This means organizations can make Customer Service capabilities available to compatible clients through a single interface, rather than building multiple custom connections.

Manage connections through the Microsoft-hosted Agent 365 Tooling Gateway, which handles authentication to your Dataverse environment. Access aligns with existing Dataverse roles and permissions, ensuring that established governance and access controls are maintained while supporting requirements for visibility, security, and governance.

The server works seamlessly with a broad range of hosts, including Microsoft 365 Copilot, Visual Studio Code, GitHub Copilot CLI, and other MCP-compatible clients. This flexibility allows organizations to make Customer Service capabilities available across different AI experiences without having to rebuild separate integrations for each interface.

Experience the benefits firsthand in ChatGPT: with the Dynamics 365 Customer Experience MCP Server for Service, teams can access trusted data, surface relevant context, and support decision-making within existing workflows. Similarly, GitHub Copilot CLI users can leverage the server’s capabilities to streamline their development processes.

At general availability, the server ships with more than 90 service-oriented tools spanning many day-to-day workflows in a customer service organization. These capabilities cover common areas of service work, including case management, customer context, knowledge, email and activities, intelligence and recommendations, data and search, and environment and administration.

The breadth of these capabilities means the server can support real, end-to-end workflows. From answering questions to taking actions necessary to get work done, teams can rely on the Dynamics 365 Customer Experience MCP Server for Service as a comprehensive platform for customer service operations.

The server is built to be extended, with admin and maker tools that help organizations shape the experience around their business requirements. This extensibility allows businesses to connect external MCP servers, bring in Copilot Studio agents, and tailor the experience to their teams’ needs without requiring significant changes or updates.

Administrators can get started by connecting the server to supported clients through the Agent 365 Tooling Gateway. Step-by-step setup instructions, supported clients, and authentication details are available in the Microsoft Learn documentation.

A recent security review using Large Language Models (LLMs) has demonstrated that thorough code analysis can be achieved at a significantly lower cost than previously thought. The review, conducted on GlobaLeaks, a mature whistleblowing platform with a history of rigorous auditing, identified 29 confirmed vulnerabilities and 12 denial-of-service issues in just one pass through the codebase.

The most striking aspect of this exercise is its affordability. With an estimated total cost of around $3,140 for API calls, the review was able to pinpoint weaknesses that would have traditionally required weeks of specialist work and a substantial budget. This shift in economics changes the landscape for defenders, making it possible for more entities to engage in thorough code analysis.

The distinction between who can afford this kind of analysis is no longer as clear-cut as it once was. Historically, only a small group with specialized skills could undertake such an exercise, pricing out casual adversaries and leaving them at a disadvantage. However, the advent of commercial LLMs has altered this dynamic, making thorough code reading more accessible to a wider range of actors.

The review itself was not conducted on neglected software but rather during an intensive hardening cycle when maintainers had made 183 commits in just one month. This is significant because findings uncovered at such moments carry greater weight than those discovered in stale or abandoned codebases, providing a clearer signal about the system’s overall security posture.

The distribution of costs across different LLM models was also noteworthy. One high-reasoning model accounted for approximately 61.9% of total spend despite processing only around 90 million tokens out of over 1.24 billion used in the campaign. Cheaper models, on the other hand, handled most of the broad reading volume at a significantly lower cost.

This disparity highlights that deeper reasoning capabilities come with a higher price tag but no longer act as an insurmountable barrier to entry. The capability is real and, by any motivated adversary’s standards, inexpensive according to GlobaLeaks. This assertion underscores the potential for LLMs to streamline code analysis without replacing expert judgment.

The review produced 110 triaged records in total: 29 confirmed vulnerabilities, 12 denial-of-service findings, 42 hardening recommendations, and 27 retained non-findings kept for transparency. The inclusion of these latter items is crucial as it demonstrates not only what was found but also what was considered and later set aside. This approach presents a healthier view of LLM-assisted research than assuming every model output holds equal weight.

Some of the most significant findings were surprisingly mundane, involving session-to-account takeover paths, whistleblower anonymity risks, tenant-boundary weaknesses, missing audit trails for sensitive actions, and availability problems triggered by a single unauthenticated user. These issues are precisely what make the result uncomfortable: the value of LLM-assisted analysis lies not in discovering exotic bugs but in making broad, systematic reading affordable enough to be repeated at scale.

The report is careful not to overstate the capabilities of machine learning models. Every candidate produced by the models was treated as a hypothesis until human reviewers validated it through code tracing and assessed its practical impact. The LLM reduced the cost of looking but did not replace expert judgment, emphasizing that thorough analysis still requires human oversight.

The takeaway for teams building or defending critical software is clear: no longer can they assume that thorough code reading is too expensive for most adversaries due to commercial LLMs changing this equation. A practical response would be continuous hardening, disciplined review, and the assumption that the next entity reading the code may indeed be cheaper, faster, and more patient than the last.

Evaluating the quality of agents is crucial during development, but it’s equally important to measure their performance after launch. The Agent Platform has introduced agent and model evaluations as a generally available feature, allowing users to assess and compare agents at both stages on one engine with consistent metrics.

When using consistent quality scoring for local experiments and live traffic, any drift in production points to an issue with the agent rather than its measurement method. This ensures that issues are identified promptly and addressed accordingly.

The primary unit of work in evaluation is the experiment, which consists of a dataset of eval cases and a set of metrics and scores to run and review as you iterate. The evaluation service comes equipped with a flexible UI for defining metrics, kicking off new runs, and reviewing results down to individual failures, where users can access the agent’s full trace and session log.

Experiments can be run locally for rapid iterations or on existing sessions and traces if an agent is already deployed on Agent Platform with telemetry enabled. The system also supports issue clustering, grouping eval failures into interpretable clusters against a user-defined taxonomy of failure reasons.

For large evaluation jobs, the service uses pre-built taxonomies such as adaptive rubrics that cover common ways agents go wrong. More than 20 pre-built metrics are included in the service, with computation-based metrics scoring deterministically against ground-truth references like ROUGE for summarization and BLEU for translation.

Beyond these metrics, an advanced LLM-judge metric workflow co-developed with Google DeepMind’s research partners creates case-specific pass/fail tests from eval cases, developer instructions, and tool declarations. This adaptive rubric grades traces against the generated rubrics, providing verdicts and rationales per rubric.

The evaluation service includes a range of metrics for assessing agent performance, including Task Success, Tool Use Quality, Safety, Trajectory Quality, Final Response Quality, Hallucination, Grounding, image and video quality. Natural-language guidelines steer rubric generation toward specified criteria, with resulting rubrics reviewable and reusable across agents.

When pre-built metrics don’t fit a specific use case, users can bring their own code-based or LLM-as-a-judge metrics. Code-based metrics are Python functions covering exact text matches and JSON-shape checks, while LLM-as-a-judge metrics carry custom criteria, rating scales, and judge models.

Regardless of how metrics are defined, they land in the same versioned registry running unchanged across offline experiments and online monitors. The system is actively expanding coverage to more specialized tasks and input modalities, with a welcome for user feedback.

The evaluation service meets users where they build agents, being part of the Agent Platform SDK with operations available over REST API for languages and pipelines not reached by the SDK. For command-line work, agents-cli makes eval a first-class command alongside deployment and telemetry inspection tools (ADK-Python agents today).

Users can even run evaluation loops from their coding agent using reusable skills that walk Claude Code and similar tools through the full agent-quality flywheel. For teams building on ADK, evaluation is built into the framework – define eval sets and run them against your agent locally as you develop, including inside pytest for CI.

The Evals Worksheet web UI provides a grid to review runs without writing code. The system also integrates with Cloud Trace, allowing users to evaluate existing traces directly with one click in the traces UI. Better still is setting up online monitors to grade live traffic as it arrives, using built-in sampling and targeted filters to manage costs.

Every evaluation needs test cases to run against, which can be generated by the service’s case generator. This tool seeds synthetic eval cases from agent instructions and tools, reducing manual effort and capping coverage at what users thought they could type out. For ADK agents, it pairs with a user simulator to reach scenarios hard to produce manually.

The environment simulator stands in for systems an agent calls, allowing users to test how the agent handles failing or slow backends without touching production. Agent and model evaluations are now generally available, with supported regions and enterprise security features listed in the regions and security features table.

Supabase has made its benchmark and framework, Supabase Evals, available as an open-source tool. This release allows developers to test the performance of various coding agents on real-world tasks, providing a comprehensive evaluation of their capabilities.

Evals is designed to assess how well AI agents can build, debug, and maintain applications using Supabase’s services. The benchmark runs against actual containerized stacks, not mock environments, ensuring that results are accurate and reliable.

The framework scores the performance of coding agents like Claude Code, Codex, and OpenCode on a range of tasks, including building schemas, debugging Edge Functions, and fixing broken RLS policies. These scenarios cover various dimensions, such as products (database, auth, storage), topics (RLS, security, migrations), and stages (build, deploy, investigate).

Supabase defined the smallest scenario set that touches each dimension once, grounded in support tickets, bug reports, and GitHub issues. The benchmark scenarios are published for public evaluation, while regression scenarios cover known failure modes and refresh daily.

The framework boots a hosted-like stack and a local CLI project in containers, allowing agents to call the actual MCP server and CLI. A platform-lite runtime exposes a Management API-compatible surface backed by @supabase/lite. Scoring combines deterministic checks with LLM-as-a-judge, giving one retry before grading.

Each eval directory contains PROMPT.md (task plus frontmatter), EVAL.ts (the scorer), and optional remote/ and local/ starting states. Shipping a local/ workspace or declaring interface: cli boots a Docker sandbox with the real CLI installed.

The results show that agents pass most scenarios without loaded skills, but skills significantly improve performance for smaller models. Opus 5 and Kimi K3 scored 100% unaided in the Build stage, while Sonnet 5 rose from 78% to 100%, GPT-5.6 Sol from 89% to 100%, and GPT-5.4 mini from 78% to 89% with skills loaded.

However, three weaknesses surfaced: agents hand-write migrations instead of using declarative schemas, prompting a skill guidance update; agents verify auth by hand rather than reaching for @supabase/server, prompting a package selection guide; and docs usage varies sharply between coding agents, even when skills are loaded.

The release of Supabase Evals marks an important step in evaluating the performance of AI agents on real-world tasks. By providing a comprehensive benchmark, developers can better understand their capabilities and limitations.

MIND has introduced its MIND AI DLP Agents, a set of tools designed to automate various data loss prevention (DLP) tasks. These agents focus on classification, investigation, policy management, remediation, and exception handling, aiming to streamline the work of security teams in this area.

The increasing speed at which sensitive data is created and moved has put pressure on organizations’ ability to govern it manually. Many DLP programs struggle with building classifiers, investigating issues, tuning policies, monitoring exceptions, and remediating exposure, tasks that consume a significant amount of time and resources.

According to Eran Barak, CEO of MIND, ‘AI has fundamentally changed how data is created and moves, but it hasn’t changed how most organizations secure it.’ He emphasizes the need for DLP programs to keep pace with AI’s speed by leveraging automation tools like the MIND AI DLP Agents.

These agents are designed to work alongside security teams as a force multiplier. They automate tasks that have traditionally been time-consuming and labor-intensive, freeing up staff to focus on more strategic activities such as reducing risk and enabling business innovation.

The MIND AI DLP Agents specialize in five key areas: custom classifier creation, policy development, issue investigation, rapid response, and reason review. Each agent is designed to address specific pain points identified by customers, including the need for efficient classification, accurate policy management, and effective incident handling.

One of the most significant features of these agents is their ability to work with natural language instructions through an interface called Model Context Protocol (MCP). This allows security teams to assign tasks in plain language without needing to navigate multiple consoles or perform repetitive operational tasks.

Organizations using MIND’s AI DLP Agents have reported a range of benefits, including an 80% reduction in DLP program effort and near-zero false positives. They also report spending significantly less time investigating incidents, with some organizations achieving a 50 percent decrease in investigation time per incident.

The solution is designed to be highly scalable without disrupting ongoing operations. Customers can deploy the MIND AI DLP Agents quickly, taking specific action to lower data security risk within hours and operating at scale without operational disruption.

The launch of JetHost’s AI Connector marks a major milestone in the evolution of web hosting management. This innovative integration enables businesses and developers to securely manage their hosting infrastructure through natural language conversations with popular AI assistants like ChatGPT, Claude, Cursor, Lovable, and other MCP-compatible platforms.

The Model Context Protocol (MCP) is at the heart of this new technology, allowing customers to connect their JetHost accounts directly with these AI assistants. This secure integration bridges the gap between traditional hosting management systems and AI tools for businesses, providing a faster and more intuitive way to manage infrastructure while maintaining complete control over security and permissions.

JetHost’s launch represents a significant step forward in web hosting management as organizations increasingly look for practical ways to connect AI with their existing software. The company’s AI Connector enables AI assistants to securely retrieve information and perform approved hosting-related tasks, reducing repetitive administrative work and improving operational efficiency.

Rather than logging into multiple systems or navigating complex dashboards, users can simply ask questions like ‘Which domains are due for renewal?’ or ‘Deploy my latest GitHub project.’ The AI Connector converts these natural-language requests into actions within the customer’s own JetHost account, streamlining routine hosting administration tasks.

The Model Context Protocol (MCP) provides a unified framework that enables compatible assistants to access business systems in a secure and structured way. This open standard allows AI assistants to securely communicate with external software platforms through standardized connectors, rather than building separate integrations for every individual platform.

JetHost’s AI Connector has been developed for a broad range of users, including software companies, SaaS platforms, digital agencies, developers, e-commerce businesses, IT teams, startups, and growing enterprises. For developers, the connector simplifies routine hosting administration tasks while for agencies it streamlines the management of multiple client environments.

JetHost AI Connector removes much of the technical complexity traditionally associated with hosting administration by allowing business owners to retrieve information using simple natural-language requests. This marks a significant shift in how businesses interact with their digital infrastructure as artificial intelligence becomes an increasingly integral part of everyday operations.

Rather than just being a support feature, JetHost views artificial intelligence as an operational layer that fundamentally changes the way businesses manage their hosting environments. The AI Connector enhances existing platforms by providing a faster and more natural interface that complements traditional management tools.

The company’s vision is to make hosting management feel effortless, like having a conversation with someone who knows exactly what you need. This goal reflects JetHost’s commitment to security, transparency, and customer control, with the AI Connector designed to maintain complete control over sensitive infrastructure.

Security remains one of the most important considerations when integrating AI into operational workflows, which is why JetHost has built multiple layers of protection into the AI Connector. These features include OAuth 2.1 authentication, PKCE verification, two-factor authentication support, account-scoped permissions, read-only access by default, short-lived access tokens, explicit approval for account-changing actions, and instant permission revocation.

Customers remain in complete control over which permissions are granted to an AI assistant and can modify or revoke access at any time. This security-first architecture allows organizations to adopt AI-powered hosting management without compromising control over sensitive infrastructure.

The JetHost AI Connector has been developed with a focus on scalability, making it suitable for software companies, SaaS platforms, digital agencies, developers, e-commerce businesses, IT teams, startups, and growing enterprises. For developers, the connector streamlines routine hosting administration tasks while for agencies it simplifies multiple client environment management.

JetHost plans to continue expanding the Connector’s capabilities while maintaining a strong commitment to security, transparency, and customer control. The company believes that AI will become an operational layer that fundamentally changes how businesses interact with their digital infrastructure rather than just replacing existing hosting platforms.

The JetHost AI Connector is powered by the Model Context Protocol (MCP), which provides a unified framework for compatible assistants to access business systems in a secure way. This open standard allows AI assistants to securely communicate with external software platforms through standardized connectors, reducing the need for separate integrations.

JetHost’s CEO & Co-Founder Metodi Drenovski stated that ‘businesses shouldn’t have to spend valuable time navigating multiple dashboards for routine tasks.’ The company aims to provide a faster and more intuitive way to manage hosting infrastructure while maintaining complete control over security and permissions, making it feel as natural and effortless as having a conversation.

The introduction of JetHost AI Connector marks an important step forward in the adoption of AI tools for businesses. This practical solution provides a secure connection between conversational AI assistants and real operational hosting environments through open standards and enterprise-grade security.

By using the JetHost AI Connector, business owners can focus on their core activities without getting bogged down by administrative tasks. The Connector makes it easy to manage hosting infrastructure by allowing users to retrieve information and complete common operational tasks using simple natural-language requests.

The company’s vision is reflected in its commitment to maintaining a seamless experience for customers while expanding the AI Connector’s capabilities. JetHost plans to continuously enhance the security features of the AI Connector, ensuring that organizations can adopt AI-powered hosting management without compromising control over sensitive infrastructure.

JetHost views artificial intelligence not just as an operational layer but also as an integral part of everyday operations in businesses. The company aims to make hosting management feel effortless and natural by providing a faster interface that complements traditional tools, rather than replacing them.

The JetHost AI Connector is designed to work seamlessly with popular AI assistants like ChatGPT, Claude, Cursor, Lovable, and other MCP-compatible platforms. This secure integration bridges the gap between traditional hosting management systems and AI tools for businesses, making it easier to manage infrastructure while maintaining control over security and permissions.

AWS has recently made available the public preview of its Amazon GuardDuty investigation agent, a cutting-edge AI-powered security tool designed to streamline threat triage and reduce the time spent on security investigations. The tool evaluates findings, correlates historical activity, and maps threat telemetry across AWS accounts and organizations, providing a comprehensive view of potential threats.

The company’s goal is to minimize the manual overhead required for security teams to investigate alerts by synthesizing metadata from security findings, 90-day activity logs, and affected resource topologies into structured analysis reports. This capability was first announced in June as part of AI-powered investigations, with a detailed walkthrough published under the investigation agent name in July.

Threat detection services like Amazon GuardDuty continuously stream alerts regarding suspicious network or runtime behavior. However, security teams often struggle with alert fatigue and the manual overhead required to correlate findings across fragmented accounts and logs. Clarke Rodgers, who works in the Office of the CISO at AWS, framed the product thesis directly on LinkedIn: ‘Security teams don’t have a detection problem. They have an investigation problem.’

The GuardDuty investigation agent addresses this bottleneck by evaluating telemetry on demand across three distinct scopes: Finding Analysis, Account Analysis, and Organization Analysis. Each scope provides a structured result set containing an overall risk rating (ranging from Info to Critical), a confidence score, classification against the MITRE ATT&CK matrix, and actionable CLI remediation steps.

Finding Analysis evaluates a specific 32-character GuardDuty finding ID during preview, supporting all Extended Threat Detection findings as well as select foundational, S3, and Runtime findings. Account Analysis assesses the current threat posture of an individual 12-digit AWS account, while Organization Analysis evaluates threat findings across up to 100 member accounts in an AWS Organization.

Sena Yakut, a cloud security architect and AWS Security Hero, drew the distinction between the agent and GuardDuty’s existing correlation capability. She noted that Extended Threat Detection connects related findings into an attack sequence, whereas Investigation analyzes affected resources, IAM activity, and surrounding context to generate a summary with recommended next steps.

Yakut sees the strongest fit for this tool in organizations without a large security function, where initial evidence-gathering consumes a disproportionate share of analyst time. She tested it against sample findings, which returned low risk ratings on test resources as expected. Her caveat is that AI should assist investigation, not replace human validation and decision-making before taking remediation actions.

Programmatically, developers and SecOps teams can trigger investigations using standard AWS SDKs, the AWS CLI via aws guardduty create-investigation, or via EventBridge rules to automate downstream response pipelines. This integration allows for automated threat triage, streamlining the process of identifying potential threats and reducing manual overhead.

The MCP integration is a key detail for teams already running agentic tooling. Through the AWS MCP Server, an engineer can trigger a threat investigation from Claude Desktop or a custom CLI agent runner using existing AWS Identity and Access Management (IAM) credentials. This places security investigation inside the same agent surface as code and infrastructure work.

The integration also inherits governance questions such as which principal actually ran the investigation, whether the agent’s context window now holds 90 days of correlated security telemetry, and how that transcript is retained. AWS’s own Loom reference platform exists precisely to answer these types of questions for agent deployments.

To address data residency and compliance concerns associated with Large Language Models (LLMs), AWS leverages its Cross-Region Inference Service (CRIS) powered by underlying Bedrock models. While compute inference might route to another region within the same geographic boundary to optimize resource availability, investigation data and generated reports remain stored within the originating home region.

The launch puts GuardDuty in a category entered first by other hyperscalers such as Microsoft’s Security Copilot for incident summarization and guided response across Defender and Sentinel. Google offers Gemini-assisted investigation inside Security Operations. What distinguishes the GuardDuty agent is its scope, which is bounded to GuardDuty’s own finding corpus and surrounding AWS telemetry rather than positioned as a general security assistant.

The GuardDuty investigation agent is currently available in public preview across 10 commercial AWS regions: US East (N. Virginia, Ohio), US West (Oregon), Canada (Central), Europe (Frankfurt, Ireland, London, Paris, Stockholm), and Asia Pacific (Tokyo). Usage is free during the preview period, subject to a throttle limit of 10 investigations per account per day, capped at a cumulative maximum of 100 per account during the preview phase. Failed investigations do not count toward either quota.

The limits on usage sit awkwardly beside the EventBridge integration. Ten investigations per account per day, capped at 100 for the whole preview, is a budget for manual triage rather than automated response pipelines. This constraint aligns with Yakut’s caveat that the preview is scoped for analysts evaluating output, not systems acting on it.

The GuardDuty investigation agent marks a significant step forward in automating threat triage and reducing the time spent on security investigations. By streamlining the process of identifying potential threats and providing actionable remediation steps, this tool has the potential to greatly improve the efficiency and effectiveness of security teams.

Siri, Apple’s virtual assistant, has undergone a significant upgrade in iOS 27. One notable aspect of the new version is its approach to interacting with users, which sets it apart from other AI assistants on the market.

For nearly two months, I’ve been testing Siri AI as part of the iOS 27 beta program. My experience so far suggests that Apple has made considerable strides in enhancing the capabilities of their assistant.

The new Siri is indeed more capable than its predecessor and delivers on its promise to be a reliable aid for users. However, it’s essential to note that if you require an extremely powerful AI tool, other options like ChatGPT, Claude, or Google Gemini might offer more advanced features.

One key difference I’ve observed between Siri and these third-party tools is the way they interact with users. Unlike some of its competitors, Siri doesn’t attempt to mimic human-like behavior or personality traits.

Siri’s responses are clear and concise, making it evident that it’s a tool designed to assist rather than engage in personal conversations. It also reminds users regularly that it can make mistakes due to its AI nature.

Other AI chatbots like ChatGPT, Claude, and Google Gemini tend to be more verbose and human-like in their interactions. This approach may appeal to some users who prefer a more personalized experience with their AI assistants.

I personally find the ‘human-like’ tendencies of these tools disingenuous. I believe that AI assistants should stay focused on providing assistance rather than pretending to be something they’re not.

The fact that Apple designed Siri in this way is refreshing, as it acknowledges the tool’s limitations and avoids creating unrealistic expectations about its capabilities.

Microsoft customers can now use IntelAgree’s contract management platform, which has just become available on the company’s commercial marketplace. This is a big deal for IntelAgree, as it marks a significant expansion of its reach into enterprise environments.

The AI technology behind this new platform helps legal and business teams analyze contracts more effectively. They can use it to identify potential risks, extract key terms, and automate related workflows. By integrating with Microsoft technologies, organizations can easily incorporate contract intelligence into their existing processes while getting the security benefits of using a Microsoft product in the cloud.

One major advantage for customers is that they can purchase IntelAgree through their existing Microsoft agreements. This means that purchases will be counted towards a customer’s Azure Consumption Commitment (MACC). As a result, teams can allocate budget they’ve already committed to contract management tasks, like automating workflows and streamlining processes.

IntelAgree has embedded Saige Assist, its purpose-built generative AI tool, into the platform. This means that answers from this tool are based on each customer’s specific procedures and standards, so recommendations always reflect their own policies and practices.

According to IntelAgree, participating in the Microsoft marketplace and co-sell program is designed to make software procurement and deployment easier for customers while enabling closer collaboration with Microsoft’s enterprise sales organization. The integration helps organizations improve contract visibility and reduce manual legal review through AI-assisted workflows – which can be especially helpful when dealing with complex contracts.

IntelAgree is available now on the Microsoft Marketplace, with integrations across the entire Microsoft ecosystem. This move will provide customers with a more effective solution for managing their contracts by combining the power of AI tools like IntelAgree’s contract management platform.

A new era in cybersecurity has dawned, one where autonomous systems can reason, adapt, and operate continuously. However, this shift also brings significant challenges, as attackers can now generate exploits faster, scale campaigns further, and operate with unprecedented efficiency. The traditional approaches to security are no longer sufficient, and a new Cyber Stack is needed to keep pace with the evolving threat landscape.

The physics of cybersecurity have changed dramatically in recent years. With the rise of AI-powered attacks, defenders must adapt their strategies to stay ahead of the curve. Microsoft has been working on Project Perception, a comprehensive agentic security system designed specifically for this new reality. At its core, Project Perception is built around three key principles: continuous perception, reasoning, and action.

Project Perception brings together signals, context, models, and specialized agents into a continuously learning system of defense. This integrated approach enables the Cyber Stack to reason across vast amounts of context, prioritize actions at machine speed, and take corrective measures in real-time. The system is designed to amplify defenders with better insights and more powerful ways to act.

The defining characteristic of Project Perception is its ability to continuously perceive risk across the entire digital estate. This involves a deep understanding of how an attacker sees the world, how a defender evaluates risk, and how protections are improved over time. To achieve this, Project Perception coordinates three classes of specialized agents: Red team agents identify potential paths to compromise before an attacker can exploit them.

Blue team agents investigate, reason over context, and determine what represents meaningful risk. Green team agents take corrective actions and strengthen defenses across the environment. Working together, these agents form a closed-loop system that continuously discovers, evaluates, and improves an organization’s security posture. This integrated approach enables Project Perception to stay ahead of emerging threats.

A key aspect of Project Perception is its ability to provide broad visibility across the digital estate. Microsoft brings together signals from various sources, including identities, endpoints, applications, data, clouds, and AI systems. This comprehensive view allows agents to reason over risk in real-time, prioritize actions, and make decisions that are informed by a deep understanding of the environment.

Project Perception’s multi-model architecture is another critical component of its design. Rather than relying on a single model, Project Perception combines frontier and specialized cyber models, optimizing for both quality and cost. This approach enables customers to benefit from advances in AI without being tied to any single model. The system continuously selects the capabilities best suited to the task, ensuring that each security workflow is optimized for effectiveness and economics.

The actuators within Project Perception are designed to connect insights to actions. Security teams do not need more information; they need better outcomes. Actuators enable agents to take corrective measures in real-time, reducing risk rather than simply identifying it. This integrated approach empowers defenders to strengthen security while remaining firmly in control.

Project Perception is built with safety first principles at its core. The system inherits the security, compliance, governance, and operational controls that Microsoft’s customers already rely on. This ensures that capabilities are delivered with the same rigor, accountability, and enterprise readiness that customers expect.

The future of security will be shaped by AI-powered attacks. Defenders need systems that can continuously perceive, reason, and act alongside them. Project Perception is a significant step towards building this future. By providing a comprehensive agentic security system, Microsoft aims to help organizations stay ahead of emerging threats and protect their digital estates with confidence.

Project Perception enters public preview on August 3, marking an important milestone in the development of this new Cyber Stack. As the threat landscape continues to evolve, it is clear that traditional approaches to security will no longer suffice. Project Perception offers a comprehensive solution for defenders seeking to stay ahead of AI-powered attacks and protect their organizations with confidence.

Microsoft’s commitment to building a secure future extends beyond Project Perception. The company has been working on various initiatives aimed at enhancing AI security through global red teaming, research-to-reality projects, and other efforts. These endeavors demonstrate Microsoft’s dedication to staying at the forefront of cybersecurity innovation and its willingness to collaborate with experts in the field.

Project Perception is a testament to Microsoft’s ongoing investment in AI-powered security solutions. By leveraging machine learning jobs and expertise from across the industry, Project Perception represents a significant step towards building more effective defenses against emerging threats.

Netflix has developed an in-house platform for serving Large Language Models (LLMs), which it calls Triton. This infrastructure is designed to handle the complexities of supporting different model sizes, hardware requirements, and rapidly evolving inference engines. The company’s account provides a detailed look at the production lessons learned from building this platform.

The platform builds on Netflix’s existing JVM-based serving layer, which continues to handle tasks such as routing, feature retrieval, candidate generation, post-processing, and logging. Smaller models can run in-process on CPUs, while larger requests are delegated to MSS, where Triton takes over model loading, batching, GPU scheduling, and multi-framework serving.

This approach allows the surrounding production workflow to remain consistent even as inference moves between local and remote hardware. Within the GPU path, Netflix selected vLLM for its operational fit and extensibility while retaining Triton’s model-management and scheduling responsibilities. Triton controls the serving environment around the model, whereas vLLM performs inference and provides extension mechanisms for custom behavior.

Netflix reports that mismatched Triton and vLLM versions can prevent deployments from loading, requiring compatible releases to be tested and pinned together. Custom models introduced another integration challenge. Hugging Face compatibility in vLLM was insufficient for some Netflix models, so the company used vLLM extension points for custom architectures and decoding behavior.

Netflix also compared two Triton packaging approaches: Triton’s Python backend and its vLLM backend. The company reports that the vLLM-backend approach allows models and frontends to evolve more independently than the Python-backend option. This choice affects how tightly a model is coupled to its serving environment rather than which engine performs inference.

The common serving interface did not eliminate differences between the underlying engines. Although Triton exposed an OpenAI-compatible API alongside KServe HTTP and gRPC frontends, Netflix still encountered gaps in how some features were handled across those integrations. Constrained decoding was one example of this issue.

Constrained decoding allows Netflix to force model responses into formats such as valid JSON by filtering the tokens the model may generate at each step. Because these rules depend on everything generated so far, the decoder must maintain state throughout the request. When vLLM pauses and later resumes a request to manage GPU resources, that state can fall out of sync with the token history.

Netflix added logic to detect this change and rebuild it before generation continues. Compatibility also affected deployment. Netflix pins tested Triton and vLLM versions together to prevent backend-loading failures, while Red-Black and Versioned deployment strategies handle changes at the model level. Versioned deployments keep old and new revisions available separately, allowing consumers to migrate after adapting to incompatible input or output schemas.

Uber has described a related approach at the application boundary. Its generative AI gateway presents an OpenAI-compatible interface across externally hosted and internally managed models, while centralising concerns including authentication, caching, observability, and routing. The implementation differs from Netflix’s serving platform, but both separate application integrations from the models, runtimes, and hosting environments behind them.

Netflix’s experience shows how a common serving interface can sit above several distinct layers. This type of architecture reflects a broader effort to give application teams a stable integration surface while model providers and serving runtimes continue to change. Netflix’s account also shows that the abstraction does not remove the underlying work: packaging, compatibility controls, constrained decoding, and deployment isolation still require engineering at each layer.

A report from Apple Insider suggests that the company is planning to introduce a new feature, similar to its existing Support Assistant tool, to the App Store. The Virtual Shopping Assistant would provide users with personalized chat experiences and relevant responses when shopping on the platform.

The existence of this potential feature was hinted at in Apple’s updated App Store privacy policy, which mentions data collection for virtual shopping assistants. However, there is no indication that the new AI-powered tool has been implemented yet, despite an update to the Apple Store app last week.

According to the policy, when a Virtual Shopping Assistant is available on the Apple Store app, it collects and stores user account information, device identifiers, carrier details, chat data, and location information if enabled. This data would be used to personalize the shopping experience and provide relevant responses.

The introduction of AI-powered shopping assistants aligns with recent research from PYMNTS Intelligence, which found that these tools have become a top priority for retailers in the current market. The report ‘Global Digital Shopping Index: The AI-Powered Shopper Has Arrived’ revealed that 37% of merchants plan to invest in AI assistants over the next three years.

This shift towards prioritizing AI-powered shopping assistants has coincided with a decrease in emphasis on other digital commerce features, such as cross-channel shopping and stored payment methods. Merchants are becoming more selective about where they allocate their development budgets and personnel.

The PYMNTS report also highlighted changes in consumer behavior, noting that nearly half of online shoppers use some type of AI during their most recent purchase. Consumers employ these tools to compare products, research purchases, and find product information before making a transaction.

Furthermore, the study found that 64% of consumers expect to use AI shopping agents within the next two years. This growing demand for AI-powered shopping experiences is driving retailers’ focus on developing more comprehensive data analysis tools to enhance their offerings.

A new tool has been made available to help users of the AI platform Claude better understand how artificial intelligence is being used in various industries and occupations. The Anthropic Economic Index connector allows anyone with access to Claude to explore data on AI usage directly, providing insights into which jobs are most affected by automation and what tasks are being taken over by machines.

The Anthropic Economic Index itself measures the actual use of AI in the economy, gathering information that has already proven useful for researchers, journalists, and policymakers. However, its creators want this valuable resource to be accessible not just to experts but also to anyone curious about how AI fits into their daily work or personal life. With the new connector, users can ask Claude questions like ‘Which occupations rely most heavily on AI?’ or ‘What are the common ways people in Colorado use Claude for?’

The data behind these answers comes directly from the Anthropic Economic Index, providing a comprehensive picture of how AI is being used across different sectors and regions. Users can start by asking broad questions about their industry or occupation, then drill down into specifics to get more detailed information. They can also ask Claude to show them the underlying data supporting any answer they receive.

It’s worth noting that the Anthropic Economic Index reflects patterns in Claude usage rather than providing a comprehensive view of the entire labor market. As users explore the data, Claude will remind them about these limitations and point them back to the source information for further context. The full datasets remain freely available on the website, alongside the new connector.

The Anthropic Economic Index connector is now live within claude.ai, requiring only a minute or so to set up. Users can find it in the connectors menu by searching for ‘Anthropic Economic Index’ and enabling it – this works with any conversation using any Claude model without needing additional installation. This tool marks an important step towards making complex data more accessible to those who need it most.

Users of claude.ai can now explore how AI is being used across various industries, occupations, and regions. The Anthropic Economic Index connector provides a valuable resource for anyone looking to understand the impact of automation on their field or daily life.

Japanese IT services company NTT DATA Group has successfully reduced the time it takes to analyze complex incidents from three days to just 30 minutes using its Codex AI tool. This achievement is a significant milestone in the company’s efforts to automate tasks and streamline work processes, marking a major shift towards creating greater value through artificial intelligence (AI).

Building on the widespread adoption of ChatGPT Enterprise across the organization, NTT DATA Group expanded Codex to approximately 9,000 employees, including both technical and non-technical roles. This move reflects the company’s evolving growth model from expanding revenue by adding headcount to creating greater value through AI.

As a global IT services company spanning consulting, systems development, and operations, NTT DATA Group is at the forefront of leveraging AI in its business environment. The company has been actively promoting the use of AI within its own organization as part of its ‘Client Zero’ approach, treating itself as the first customer to test AI solutions.

The success with Codex was preceded by the deployment of ChatGPT Enterprise across NTT DATA Group’s global operations following a strategic partnership with OpenAI in May 2025. This move enabled employees to build experience working with AI for research, writing, and content creation, laying the groundwork for delegating clearly defined tasks to Codex.

One notable example of Codex’s capabilities is its automation of complex incident analysis for critical systems. Previously requiring five experienced engineers and taking three days to complete, this task was completed in just 30 minutes using Codex. This achievement quickly gained attention from senior leaders and became an early proof point for what Codex could make possible.

Codex has demonstrated the potential to go beyond being a coding assistant by independently investigating, executing, testing, and revising based on instructions. Hiroaki Sato of NTT DATA Group’s AI Technology Department notes that ‘Codex has changed how people across the company think about AI.’ The idea that AI can take the lead in carrying out work has had an impact similar to the arrival of ChatGPT.

Sato emphasizes, ‘Codex is about more than helping developers write code faster. As ChatGPT Enterprise has become part of how our employees think, explore ideas, and solve problems, Codex opens the door to a new way of working.’ By enabling AI to research, organize, execute, and validate work, it empowers every employee to transform their role.

The potential demonstrated in incident analysis extends beyond software development. NTT DATA Group’s ‘Client Zero’ approach encourages employees across various departments to test AI solutions within their daily work. As a result, Codex adoption is expanding from engineers to nontechnical employees across different functions.

Nontechnical employees are already using Codex for tasks such as building lightweight tools, organizing large volumes of files, analyzing data in Excel, summarizing documents, and scripting repetitive processes. These tasks often require automation but would be too time-consuming to handle manually.

For example, employees use Codex to extract transportation expenses from credit card statements and transfer them into travel expense forms. This process involves working across multiple files, understanding the structure and entry rules of each sheet, and verifying the completed transfer, streamlining work that was previously manual and cumbersome.

Another significant shift is that nontechnical employees can now complete tasks that once required support from engineers. Creating an analytical report, for instance, previously required preparing data in a business intelligence tool and building a dashboard. With Codex, employees can analyze raw data directly and create the reports they need, improving efficiency while reducing dependence on specialized tools and skills.

In this way, Codex is doing more than supporting software development; it’s lowering the barrier to work that once required specialized expertise and enabling employees without development experience to create practical outputs for their roles.

To expand the use of Codex across the organization, NTT DATA Group recognized the need for an environment where everyone can use it safely and with confidence. This foundation is crucial not only for its own employees but also for customers across various industries supported by the company.

The OpenAI Center of Excellence has developed and rolled out security guidelines alongside practical guidance for adoption. These guidelines clarify what data can be used, which systems Codex can connect to, how network traffic is managed, which sandbox mode to apply, what level of automation is appropriate, and where human review is required.

This environment enables employees across technical and nontechnical roles to confidently explore new ways to use Codex. With this foundation in place, Codex adoption is moving beyond a small number of advanced use cases and steadily becoming part of the daily work for approximately 9,000 employees.

NTT DATA Group’s experience with Codex has delivered tangible results across the organization, including completing complex incident analysis in just 30 minutes, increasing weekly active Codex users by 1.4 times after publishing a usage guide and conducting hands-on training, and automating internal system operations using Playwright to reduce time spent on routine daily tasks.

The company’s journey with Codex highlights five key lessons for embedding AI tools across an organization: making ChatGPT Enterprise part of daily work, deploying it broadly to create network effects through peer learning and word of mouth, creating a secure environment where employees can use AI safely, continuously improving adoption programs using usage data, surveys, and employee interviews, and having the CoE identify and generalize high-impact use cases.

NTT DATA Group’s ambition goes beyond making individual tasks more efficient. By embedding AI into everyday work, the company is creating an environment where people and AI can each contribute their strengths and generate greater value faster.

A major breakthrough in human resources technology has been achieved by Omni HR, a leading provider of AI-native HR and payroll platforms. The company has launched native MCP integration, allowing popular AI assistants like Claude, ChatGPT, and Cursor to access live HR data directly within the platform. This innovation marks a significant step forward for businesses seeking to streamline their operations with cutting-edge technology.

The integration ensures that every action taken by an AI assistant is scoped to the user’s existing permissions, eliminating any risk of unauthorized access or data breaches. No new access is granted, and no HR data is stored at the connector level, maintaining the highest standards of security and compliance.

Omni HR has been expanding its AI capabilities in recent months. In June 2026, the company released Mino, an AI agent designed to work seamlessly within the platform. The MCP integration builds on this intelligence by extending it outward, enabling AI tools used daily by teams to access Omni’s unified data layer and act upon it without switching systems or copying data.

The launch of MCP integration addresses a pressing concern in the industry: the lack of trust among HR leaders when it comes to relying solely on AI outputs. According to Omni’s State of AI in HR report, only 21% of HR leaders across Singapore and the Philippines have sufficient confidence in AI-generated results to act without manual review. This ‘confidence gap’ is largely attributed to fragmented and inconsistent HR data.

Omni’s centralized, real-time system consolidates all HR data into one platform, providing a reliable foundation for AI-driven insights. With MCP integration connected, teams can ask their AI assistant to perform tasks such as pulling employee leave information or running headcount snapshots by department, receiving the answer in plain language without needing to switch between systems.

The CEO of Omni HR, Brian Ip, emphasizes that what sets this integration apart is its commitment to maintaining user access controls. ‘Mino brought AI into Omni,’ he explains. ‘The MCP integration brings Omni into the AI tools your teams are already using every day.’ This approach ensures that AI assistants do not gain a broader view of business operations than their human operators, setting a new standard for industry-wide adoption.

MCP is now available in early access to Omni HR customers, who can book a demo at omnihr.co/demo. As the platform continues to expand its capabilities and user base, businesses across Asia are poised to benefit from enhanced efficiency and productivity through seamless integration of AI assistants with live HR data.