Where Are the Smartest Startups Building Future Robots?

London, United Kingdom 

Right now, the costliest mistake for a robotics startup is waiting. Each week spent building the right compute stack, managing simulation environments, and creating enough synthetic training data is another week for competitors to catch up. This bottleneck, rather than a lack of talent, vision, or ambition, has often been the difference between robotics companies that grow and those that get stuck at the prototype stage. London is now the place where this situation changes. 

On June 9, 2026, Nebius and NVIDIA introduced the Physical AI Living Lab, a six-month program that provides early-stage British and European robotics startups with access to infrastructure once available only to large, well-funded companies. The Physical AI Living Lab is different from a typical incubator. It does not provide office space, seed funding, or a demo day. Instead, it offers something even rarer: the computing power, tools, and engineering support needed to quickly take a robot from simulation to deployment. 

The Compute Problem No One Talks About Loudly Enough 

Founders working in physical AI often share the same frustration. Their models work well, and their vision is clear, but building a smooth pipeline from data generation to simulation, training, and production deployment requires infrastructure that consumes both time and budget. Calvin Zhou, co-founder of RoboForce, which builds AI robots for solar farms, construction sites, and farms, explained it clearly: “Manual handoffs between data generation, simulation, and training mean our GPUs can sit idle costing us both time and money.” 

Idle time is more than merely a small inefficiency. For early-stage teams with limited resources, it can determine whether a product reaches the market. 

What the Physical AI Living Lab Actually Provides 

Physical AI relies on large-scale simulation, synthetic data, and fast compute power resources that most early-stage robotics companies cannot build on their own. The Physical AI Living Lab removes this barrier by giving founders access to the same tools and compute power used to build physical AI at scale, helping them move from simulation to physical deployment more quickly. 

The program offers a strong technology stack. Startups gain hands-on experience with NVIDIA OSMO for workload management, NVIDIA Cosmos models for robot simulation and training, NVIDIA Isaac Sim and Isaac Lab, and the NVIDIA Physical AI Data Factory Blueprint. All of this runs on Nebius AI Cloud infrastructure. 

The synthetic data part is especially important. Synthetic data is generated using Voxel51’s FiftyOne integration, which is built on NVIDIA Cosmos models as world base models. For robotics teams training manipulation policies or mobile navigation systems, the ability to create large, varied, and realistic training environments without physical hardware is not just convenient it is essential. Collecting real-world data is slow, costly, and often cannot cover all the unusual situations a robot might face. 

Why Blackwell Changes the Equation 

The hardware is just as important as the software. The first phase of the Physical AI Living Lab uses Nebius’s UK-based infrastructure, built on NVIDIA RTX PRO 6000 Blackwell hardware. Running synthetic data and robot simulation on Blackwell hardware together is a big deal. Both simulation quality and training speed depend on raw computing power. With Blackwell hardware at this scale, startups can test and refine their policies in hours rather than weeks. In the past, this kind of speed required either a large internal GPU cluster or an enterprise cloud deal that most early-stage companies could not get. 

RoboForce used NVIDIA Cosmos models on the Nebius AI Cloud to cut pipeline setup time by over 70% and accelerate the time to production for new policies. This advantage did not come from new algorithms, but from making the infrastructure smoother and easier to use. 

London as a Physical AI Hub 

Choosing London for the program’s first phase was intentional. The UK is known for top robotics and AI research, but there is still a gap between this innovation and real, market-ready physical AI solutions. British universities train excellent robotics researchers, and the country has a strong deep-tech investment scene. However, it has lacked a way to connect academic research with the production infrastructure required to build a real-world system. The Physical AI Living Lab makes Nebius AI Cloud, which is missing an on-ramp. Applications go through the NVIDIA Inception pipeline, and the first group starts in September 2026. Engineers from Nebius and NVIDIA will offer technical support during the program. 

A Model That Could Travel 

Both organizations plan to expand the Living Lab to more locations over time and welcome more participants as demand for robotics infrastructure grows. This London launch is a proof of concept for a broader strategy: treating Physical AI Living Lab environments as repeatable components of a global robotics development network, each supported by Nebius AI Cloud and NVIDIA Cosmos models and tools. 

The Founders Who Should Be Paying Attention 

The program is designed for teams that have a strong model and a real use case, but are held back by the high cost and complexity of building enterprise-level simulation and compute infrastructure on their own. This includes areas such as warehouse automation, agricultural robotics, industrial inspection, and autonomous last-mile delivery any field where synthetic data, robot simulation, and Blackwell hardware can help accelerate the move from prototype to pilot deployment. 

The robotics startups that will shape the next decade are not always the ones with the most funding. They are the ones that move fastest from a working model to a working robot. The Physical AI Living Lab is built on the idea that this path goes through London, and that having the right infrastructure at the right time is what turns a prospective demo into a real product.

Source: Nebius launches Physical AI Living Lab for UK and European robotics startups built with NVIDIA technologies 

Why Do 93% Of Global Bosses Fear Cloud Border Laws?

Dubai, United Arab Emirates 

Ninety-three percent of global executives told researchers they now see data sovereignty rules as a direct threat to their companies’ ability to operate internationally. This number comes from a major new study on technology fragmentation. It depicts a business world where IBM tech sovereignty is now common boardroom language. The real question is no longer if local data laws will affect your AI plans, but how much they will. 

This finding is especially important in Dubai, a city that has spent the last decade building its function as a bridge between East and West in the global digital economy. 

IBM Tech Sovereignty and the Splintering of the Global Cloud 

For most of the past fifteen years, companies saw the cloud as borderless. Data moved freely, AI models used global datasets, and infrastructure choices were based on cost rather than laws. Now, that model is breaking apart. 

Governments from Riyadh to Brussels to New Delhi have created or are planning rules that require certain types of data, such as health records, financial transactions, and biometric identifiers, to stay within national borders. The European Union’s GDPR set the standard. India’s Digital Personal Data Protection Act came next. The UAE’s data protection law, Federal Decree-Law No. 45 of 2021, applies similar rules to one of the world’s busiest trade routes. 

The new study measures how these changes affect investment. Boards that used to approve long-term AI projects confidently now face a patchwork of legal requirements. For example, a financial services company working in seven countries now has seven different, and possibly conflicting, compliance rules. IBM’s research teams and academic partners have closely tracked this. Their data shows that concerns about IBM tech sovereignty have moved from legal teams to decisions about where to invest money. 

AI Governance Frameworks Are Not Keeping Pace 

The gap between what regulators want and what AI governance frameworks can do is growing faster than most compliance teams expected. Most national AI rules were written for local situations. They were not designed for cases in which one AI model, trained on data from 14 countries, makes credit decisions for customers in jurisdictions with different transparency rules. 

Here is a real example: a Gulf-based bank uses a powerful language model to assess trade finance risk. The model uses shipping data, credit histories, and commodity prices, some of which come from EU-regulated sources. Under the GDPR, processing EU personal data outside approved systems poses serious legal risks. The bank’s AI governance frameworks now have to track where data is stored, where it moves, and where the AI makes decisions. These are often different places. 

The study found that 67% of executives had already delayed or changed at least one AI project in the past 18 months due to uncertainty about data sovereignty. These delays are costly. One logistics company in the study estimated it lost $4.2 million in competitive advantage by delaying the launch of a predictive routing system while legal teams sorted out cross-border data issues that regulators had not yet clarified. 

Dubai’s Strategic Bet and the Dubai Future Foundation 

Few organizations have focused on this issue more than the Dubai Future Foundation. As the UAE government’s main group for long-term technology planning, the Foundation has been clear about the challenge. Dubai wants to be an AI hub, but AI hubs need data to move freely, and now, data flows are politically complicated. 

The Dubai Future Foundation is working on both bilateral and multilateral data-sharing agreements that serve as diplomatic tools for technology. At the same time, it is investing in sovereign cloud infrastructure to meet localization rules without cutting Dubai off from global networks. The Emirates is not alone in this strategy. Singapore’s Digital Economy Agreements, Saudi Arabia’s LEAP initiative, and the EU’s GAIA-X project are all similar efforts. They all bet that regional sovereignty and international connectivity can work together if the infrastructure is built with care. 

Whether this strategy works depends on whether international AI governance frameworks develop faster than the patchwork of national regulations. 

Operational Continuity Cloud Infrastructure Sovereignty Regulations: The C-Suite’s Major Headache 

Strip away the policy language, and what executives are actually managing is an operational continuity cloud infrastructure sovereignty regulations problem. The question is simple and brutal: if a regulation changes overnight in a country where you have active workloads, can you move those workloads without disrupting your business? 

For most companies, the honest answer is no. They cannot move workloads quickly, easily, or cheaply. Cloud systems built for efficiency, not flexibility, are often deeply tied to local providers. Moving a production AI system to another cloud region is a big job. It can take six to eighteen months, depending on how complex the setup is. 

This is why the study’s 93% figure is not only about executive worry. It is a logical reaction to real risks. Boards are not scared of technology itself. They are worried about sudden regulatory changes that could render their current systems illegal in markets that account for 20% of their revenue. 

What Boards Are Actually Doing About It 

The best responses fall into three main groups. First is architectural diversification, which means building cloud systems from the start to run workloads across several regional providers, trading some efficiency for greater regulatory flexibility. Second is making policy monitoring a core business task, so companies treat political and regulatory updates as part of daily operations, not just legal work. Third is forming sovereign AI partnerships, in which companies work directly with groups like the Dubai Future Foundation to build AI systems that comply with local regulations while still connecting to global networks. 

IBM tech sovereignty solutions are now a major part of this third approach. These products let companies run AI in sovereign cloud environments without having to rebuild all their technology systems. 

The executives who treat operational continuity, cloud infrastructure, and sovereignty regulations as a compliance checkbox will keep losing ground to those who treat them as a way to compete. In the next 18 months, there will be no global agreement on data borders. Instead, we will see more fragmentation, more country-to-country deals, and more decisions made with regulatory uncertainty. The companies that prepare for this reality now, instead of hoping for a borderless cloud, will be the ones still operating when matters settle down.

Source: Dubai Future Foundation, IBM global study shows UAE ahead of peers in AI governance adoption 

How Does Cisco Stop Hacker Attacks Without a Reboot?

San Jose, California.  

On average, enterprise network security teams take 21 days to patch a critical vulnerability after it becomes public. Meanwhile, attackers can act within hours. The cost of this gap has never been higher, and rebooting a production switch to apply a fix is no longer an option. At Cisco Live 2026, Cisco addressed both issues at once with Cisco Cloud Control.  

At Cisco Live 2026, Cisco launched Cisco Cloud Control as the base for its Agentic Ops platform vision, where people and AI agents work together to manage networking, security, computing, observability, and joint effort from one control panel. The main feature is not just the architecture, but what it enables Cisco to do in real time on a live network without any downtime.  

Cisco Cloud Control And The End Of The Patch Window 

Most enterprise security teams follow a familiar routine: a vulnerability appears, a ticket is created, a maintenance window is scheduled, and the fix is applied at 2 AM on a Saturday. This approach worked when attackers moved slowly, but that is no longer the case.  

Cisco has extended its Live Protect Runtime Security to Cisco Nexus 9000 switches, offering protection against new vulnerabilities without requiring software upgrades or reboots. This means protection is applied while the system is running. There is no need for an upgrade or reboot. The switch continues to forward traffic while the vulnerability is handled at the software level.  

Expanding Live Protect runtime security is a key short‑term advantage, especially as the time between vulnerability discovery and exploitation continues to shrink. For places like hospitals, trading floors, or factories where downtime is costly, this is more than a mere convenience it is a major change in how protection is delivered.  

Live Protect runtime security is first available for Cisco Nexus 9000 series switches and comes with Nexus One entitlement. Later in 2026, it will also be available for campus and branch smart switches and secure routers.  

One Data Plane, Two Operators, Human and Automated 

The real story behind Cisco Cloud Control isn’t about individual features. It’s about the shared data layer that supports everything.  

Cisco Data Fabric, powered by Splunk, brings together telemetry from networks, applications, security, and third‑party sources into one layer. Both human analysts and automated agents use this shared data, which forms the base for Cloud Control and the agentic SOC. This shared foundation is important because it prevents situations where AI agents and human analysts work from different data sets, reach different conclusions, and interfere with each other’s fixes.  

Imagine a hospital network facing a lateral movement attack at 3:47 AM. For older systems, an on‑call analyst would wake up, log into several consoles, manually check alerts, and begin isolating network segments. With the AgenticOps platform, an autonomous agent detects the problem, examines data from network, application, and security layers simultaneously, and starts containing the threat. Meanwhile, the analyst can view the same data and intervene or modify the agent’s actions at any time.  

The platform serves as the foundation for Cisco AgenticOps. It enables people and autonomous agents to solve problems together while keeping humans in control. Autonomous agents manage incident life cycles to accelerate resolution using the Cisco AI Canvas.  

This isn’t about automation replacing human decisions. Instead, automation handles speed so people can focus on decisions that truly require their input.  

Cisco Cloud Control Meets 50+ Ecosystems 

Cisco Cloud Control combines the company’s networking, security, computing, observability, and joint effort tools, letting users manage and secure everything in one place. This removes the need to switch between different consoles. The platform can also connect to over 50 third‑party platforms and tools using built‑in connectors or the open model context protocol.  

The integrated list includes AWS, Microsoft, Google Cloud, ServiceNow, PagerDuty, Slack, and Wiz. For security teams working in hybrid multi-cloud environments, bringing third-party signals into the same data layer as Cisco’s own telemetry helps close a gap that attackers have often used: the spaces between different vendor systems.  

The Quantum Threat That Most Enterprises Are Not Ready For 

Beyond the operational news at Cisco Live 2026, there was a bigger warning that needs attention: the vulnerability of quantum-ready infrastructure vulnerability defense.  

Cisco shared a bold plan for post‑quantum security to address rising concerns about the harvest‑now, decrypt‑later attacks. These attacks, cybercriminals collect encrypted data now to decrypt it later with quantum computers. Cisco has promised to enable quantum‑safe communications for most of its main products by December 2026.  

New quantum-ready assessments, available through Cisco IQ and set for global release in July 2026, help organizations identify which assets are most at risk from harvest‑now, decrypt‑later attacks and show where to focus their defenses. From now on, all new campus, branch, and data center routers, switches, and firewall series will come with Quantum Safe Secure Boot.  

Many organizations have put off defending against vulnerabilities in quantum‑ready infrastructure because the threat seems far away, but it isn’t. Intelligence agencies and nation‑state actors have been collecting encrypted enterprise communications for years, hoping to decrypt them with quantum computers in the future. Companies that wait for compliance rules to update their cryptography may find their data already compromised before they even start patching.  

The Shift Cisco Is Betting On. 

Companies building agentic AI operations. The key message from Cisco Live 2026 is that Cisco is now offering an operating model, not just network hardware. In an agentic enterprise where technologies work at software speed, owning both the infrastructure and the control plane gives Cisco an advantage that single‑solution vendors can’t easily match.   

Security teams that ignore Cisco Cloud Control, Live Protect, runtime security, and the Agentic Ops platforms as just marketing may end up stuck with 2 AM maintenance windows while their competitors move ahead. The Saturday patch window is gone for good. The real question is whether organizations will build the new architecture themselves before the next exploit forces them to do so. 

Source: Cisco Unveils Agentic Platform for Operating and Defending Critical IT Infrastructure 

Who Is Using AI To Make Custom Shirts In Seconds?

Seattle, Washington  

A grandmother in Ohio, who had never used Photoshop, just created matching T-shirts for her family reunion. She wrote one sentence into her phone. Seconds later, she had a finished graphic of a golden retriever dressed as a 1970s astronaut, ready to order on a hoodie, tumbler, or sweatshirt, with Prime delivery on the way. She didn’t hire a designer or open a design platform. She simply used Alexa for Shopping. 

On June 8, 2026, Amazon announced that customers can now design and order custom merchandise using AI through Alexa for Shopping. What used to require a design tool, a print-on-demand platform, and a fulfillment service can now be done with a single text prompt in the Amazon Shopping app. The real story is how this change removes the barriers that have kept everyday people from making the things they actually want. 

How Alexa for Shopping Turned a Prompt Into a Product 

The feature lets customers describe an idea in the Amazon Shopping app or on Amazon.com, and it instantly creates a design that can be applied to T-shirts, hoodies, tumblers, and other products available through Amazon Merch on Demand, Amazon’s print-on-demand service. The tool is free to use. Customers only pay for the physical products they order. 

Consumers give the Alexa for Shopping assistant a prompt that creates custom designs “in seconds.” In announcing the feature, Amazon offered a sample prompt: “make a design of a golden retriever as a 90s corporate lawyer at a disco.” The AI generates the design, and users can then improve it by selecting suggested actions or typing additional changes. 

This is not simply a chatbot that answers questions about shipping. It is a creative tool built on top of a system that already handles hundreds of millions of orders. By May 2026, Alexa for Shopping had become a way to complete transactions, not just provide information. The custom merch feature adds a creative layer onto this existing system. 

The Audience Nobody Expected to Design Custom Merch 

People in the design-on-demand industry used to think that most custom merchandise tools were meant for small business owners, content creators, and independent artists with at least some visual skills. Services like Redbubble, Bonfire, and Spring were made for users who could work with a virtual canvas, understand file exports, and handle a learning curve. 

Amazon’s new capability can be used to create personalized gifts, matching shirts for family reunions, team outings, or friend trips, and custom gear for holidays, game day, and other occasions, according to Amazon’s release with no design skills required. 

In other words, the target is not a creator. It is a regular consumer who has an idea but no way to make it happen. Now, that person has voice-activated AI fashion design tools at their fingertips inside the app they already use to buy paper towels. 

A customer taps the Alexa icon at the bottom right of the Amazon app and describes an idea, such as matching shirts for a family reunion or a pet as a cartoon astronaut. They get an AI-generated design in seconds, can refine it with recommended changes or more text prompts, share it with friends or family through a link so others can order the same item, and check out just like any other Amazon purchase. 

The sharing feature is more important than it seems. One person in a group chat creates the design, and everyone else gets a direct link to order the same item. This is not simply a design tool; it is a viral commerce loop built into a consumer’s product. 

What Amazon Merch on Demand Makes Possible at Scale 

Every order is backed by Amazon Merch on Demand, the print-on-demand system that manages production, fulfillment, and customer service. Designs can be put on T-shirts, hoodies, tumblers, and other products through this service, which organizations like the UFC also use. The infrastructure was already in place. What changed is who can use it and how easy it is now. 

Amazon is not starting from scratch in the custom merch business. Its AI shopping system grew quickly over the past year, with Rufus helping more than 300 million customers in 2025 and generating almost $12 billion in additional annual sales, according to Amazon’s Q4 2025 results. The design custom merch feature is the creative, consumer-facing part of a logistics and AI system that Amazon has built over many years. 

The Risk Hiding Inside the Opportunity 

For platforms built around independent artists and creators, such as Redbubble, Spring, and Fourthwall, this development poses a major challenge. Amazon’s new feature creates even more competition for online merch platforms such as Redbubble, Bonfire, Spring, Fourthwall, and others. 

These platforms still offer features that Amazon does not fully match, such as a creator economy, community discovery, and artist-owned storefronts. But such a casual buyer, such as someone who wants five custom shirts for a bachelorette party, may never visit those platforms again. Now, the buyer has Alexa for Shopping and a simple text box. 

There is also the issue of intellectual property. Generative AI design tools used by many people sometimes create designs that resemble existing artwork, brand logos, or copyrighted characters. Amazon has not shared details about how it moderates or filters these designs. That is a significant gap. 

A Different Kind of Design Democratization 

The phrase “no design experience required” has been used in software marketing for thirty years. What’s different now is that it’s finally true. Voice-activated AI fashion design tools have removed the last barrier: needing to use any interface at all. You describe what you want, it builds it, and you order. 

The custom merch feature stands out from most Amazon product launches. Usually, visibility inside the assistant is a key commercial factor, but here, designs are not listed as pre-existing products competing for placement. Instead, they are created on demand for each user’s request. 

This on-demand, zero-catalog approach is the future of consumer product creation. For every retailer, platform, and independent designer watching from the sidelines, the real question is not whether to use Alexa for Shopping and Amazon Merch on Demand, but how quickly they can adapt before shoppers get used to having just one way to order. The grandmother in Ohio has already made her choice. 

Source: What you need to know about Amazon today: June 10, 2026

What New Hidden Tricks Did Apple Add To Your Phone?

Cupertino, California. 

The dinner bill lands on the table. Eight people look at it. Someone pulls out a calculator app. Someone else debates who ordered the extra guacamole. This routine, familiar and a bit awkward, is now optional. Apple has made it unnecessary, and that is just one of the subtle but important changes announced at this week’s WWDC 2026. 

While most headlines focused on the new Siri AI, the Apple services features announced this week are important in their own right. They are precise, practical, and in many cases, long overdue. Executives traveling to new cities, small business owners splitting client dinners, and anyone who has wanted more control over their location data all have something new to try. 

How Apple Services Features Changed Your Daily Routine Overnight 

The updates Apple announced are not flashy. They show up during everyday moments: when a waiter brings the check, when you buy a surprise gift and want to keep your location private, or when you look down at a city from a plane and notice the map actually matches what you see. 

That last example is Apple Maps Flyover. The feature has been around before, but the new version coming this fall uses aerial photography and AI to create clearer, easier-to-read city views. For municipal planners, architects, or executives vetting an unfamiliar market before a site visit, the difference between a blurry overhead render and a crisp, navigable aerial model is not cosmetic. It is functional. Apple also says the updated Flyover will include a Local Lists feature that highlights trending restaurants and destinations in the United States using privacy-friendly insights. This means recommendations are based on overall trends, not your personal data. 

The End of the Dinner-Bill Standoff 

The Visual Intelligence bill split feature is worth highlighting because it makes splitting the bill much easier, a situation many people deal with several times a week. 

Just point your iPhone camera at a printed receipt or open a photo of one in Messages. Apple Intelligence will recognize each item. You tap what you ordered, and the app calculates your share of the bill, including tax and tip, then sends the exact amount via Apple Cash. The Visual Intelligence bill split feature works in the Camera app’s Siri mode, Apple Wallet, and directly in Messages. You do not have to open another app, enter numbers by hand, or try to split the shrimp appetizer into your head. 

For small business owners who host working lunches or executives who often have client dinners, this feature is a real time-saver. Expense tracking software usually struggles with group receipts. While this feature does not fix everything, it removes the most frustrating part of the process. 

Advanced Location Sharing Privacy Updates iOS: Fine-Grained Control, Finally 

This is where Apple made its most underrated move of the week. 

Find My is getting advanced location sharing privacy updates that iOS users have wanted for years. The old system was simple: you either shared your location, or you did not. The new version offers more choices. You can share your location for a set period, such as a specific number of minutes, hours, or days, or until a specific date and time. You can also pause sharing with a contact until the end of the day without ending the connection completely. 

The uses for this are clear. Maybe you are buying a birthday gift for someone who can see your location in Find My. Or you are meeting a date for the first time and only want to share your location for two hours. Or you are a parent who wants your teenager to have your location during their commute, but not on weekends. Before, each of these situations meant you had to turn sharing off completely, remember to turn it back on, and explain why it was off. 

With the new advanced location sharing privacy updates, iOS treats your location like something you can lend for a while, not give away forever. This is a real change in thinking. People who care about privacy have pointed out that the old all-or-nothing approach forced you to either share your movements all the time or look suspicious by turning it off. Now, custom-duration sharing solves that problem. 

Apple Watch Gets Pulled Into the System 

The Find My update also comes to the Apple Watch. A new unified app replaces the three separate apps Find Devices, Find Items, and Find People with a map-focused interface. Precision Finding now helps you locate a paired iPhone, a second-generation AirTag, or AirPods Pro 3. 

If you have ever searched for your phone in a hotel room at 6 a.m., this new setup is instantly helpful. Now there is just one app, one interface, and one place to check. 

What This Signals About Apple’s Direction 

There is a clear pattern in these new Apple service features. Apple is building systems that understand greater context and need less manual effort. Scanning a receipt is quicker than typing numbers. Sharing your location for a set time is more straightforward than suddenly turning it off. An AI-enhanced aerial view is more helpful than a blurry map. 

None of these features is a novel idea. What stands out is how well they are put together, especially how they work together. The Visual Intelligence bill split works simultaneously in Wallet, Messages, and Camera. The Apple Maps Flyover improvements use the same privacy-focused system Apple has been developing for years. The advanced location-sharing privacy updates coming this fall are based on a system that lets users decide how much data to share, rather than assuming they will share everything. 

Apple does not often say exactly what it is working toward, but you can see the direction in the details. The goal is a phone that takes care of more of the small hassles in daily life, like splitting checks, managing location sharing, and navigation, without asking you to trust it with everything all the time. The real question is whether users will notice these changes before they start relying on them.

Source: Apple introduces innovative features and intelligence experiences across services 

How Will Salesforce Track If Your AI Agent Fails You?

San Francisco, California  

Imagine a customer typing a question to your AI agent in all caps, repeating it, and then asking for a human representative. Most click-tracking software would count this as three interactions and mark the session as ‘active.’ But with the Summer ’26 update, Salesforce sees it differently: as a sign that automation has let someone down. 

This difference between tracking activity and measuring quality lies at the heart of Salesforce Agent Analytics. It denotes a real shift from how companies have usually judged AI performance. 

Why Click-Counting Never Told the Full Story 

For years, the metrics used to judge AI customer service agents were basic. Did the user click? Did the chat stay open? Did they avoid filing a ticket? These signs were seen as proof of success. If a session ended without a support case, it was called a deflection and counted as a win. No one checked if the customer was actually satisfied or just worn out. 

As AI agents became more independent, the problem with this logic became clear. Salesforce’s research team found that in over 2,500 conversations studied for its ICLR 2026 submission, 93% were labeled as successful by standard metrics, even when agents had stopped helping and just repeated what users said without solving anything. This failure is called ‘echoing.’ The metric that missed it is simply ‘inadequate.’ 

This is the problem Salesforce Agent Analytics now directly addresses. 

Summer ’26: The Architecture of Honest Measurement 

The Summer ’26 release, which goes live between June 13 and June 15, 2026, brings in Refined Agent Analytics. This is a unified dashboard that integrates Service Agent and Employee Agent data into one view, with over 40 metrics covering Quality, Health, Effectiveness, and Usage. While this is a solid upgrade, the bigger change is the introduction of Custom Scorers, now in Beta. 

Custom Scorers don’t just count clicks. They actually read the conversations. 

With LLM session evaluation, these scorers look at the entire conversation and grade it based on what matters to a business: Sentiment, Tone of Voice, Product Interest, Escalation Trigger, and Courtesy. For example, a company selling enterprise software might see ‘Product Interest’ as when a user starts comparing features with a competitor. A healthcare portal might define ‘Escalation Trigger’ as the exact words that come before an angry callback. Now, both of these can be set as scoring criteria. 

The practical impact is clear. While legacy analytics might mark a closed chat window as a resolved case, a Custom Scorer can detect when a conversation shifts from neutral to hostile over several messages and ends with the user leaving. That’s not a deflection; it’s a failure. Now, Salesforce Agent Analytics can call it what it is. 

The Deflection Metric Gets a Conscience 

For a long time, deflection metrics have been a vanity stat in AI customer service. High deflection rates made executive dashboards look good, even if customers were just giving up instead of getting answers. OpenTable’s use of Agent force showed a better way. Their team created a live deflection score that updates during each talk, starting at neutral and rising in response to real signals. For example, typing in all caps raises the score, and asking for a human raises it more. The agent uses this live score to decide in real time whether to keep trying, open a case, or escalate. 

This approach is fundamentally different from just counting closed windows. It treats frustration as real data, not just something missing. With Summer ’26, this idea is now built into the platform itself, so any company can use it, not just those with custom solutions. 

Qualitative Scoring at Machine Speed 

The deeper shift here is a methodology. Qualitative automated customer service agent scoring  the practice of using a language model to judge the performance of another language model  was considered a scholarly exercise as recently as 2024. The concern was obvious: what keeps the evaluating model from having the same blind spots as the model being evaluated? The answer Salesforce has landed on is human-defined rubrics. An enterprise writes the scoring criteria. The LLM applies that criterion at scale to every session, not just a sampled subset. 

This is what makes qualitative automated customer service agent scoring practically viable for businesses with thousands of daily agent interactions. A human QA team might only review about 2% of the sessions. A Custom Scorer checks all of them, catching the same escalation triggers and tone signals that a trained reviewer would notice, and does so before the customer can complain. 

Developers set up these scorers using the Metadata API and store their definitions in source control under the aiAgentScorerDefinitions folder. They turn them on from the Scorer Hub. The whole process is designed to make LLM session evaluation a repeatable, auditable engineering practice, not just an occasional review. 

What Executives Should Actually Be Watching 

With Custom Scorers now part of Salesforce Agent Analytics, CX leaders need to shift the conversation with their teams. Instead of asking, ‘What is our deflection rate?’ the better question is, ‘Of the sessions we deflected, how many ended with a sentiment score that shows real resolution?’ 

Over time, this difference will separate companies that build trust in their AI agents from those that simply reduce ticket volumes while harming customer relationships. Deflection of metrics without context have always shown problems only after the fact. LLM session evaluation lets you see problems before they get worse. 

Salesforce has now built a quality-control function directly into its platform’s measurement tools. Companies that use it well will not only know when their agents fail, but also how they failed, and they’ll have the tools to fix problems before the next customer interaction.

Source: New Implementation of LLM-based Deflection and Abandonment Metrics within Agent Analytics (Update on July 1, 2026) 

Why Are 89% of Tech Bosses Unready for Massive AI?

Armonk, New York 

Eighty-nine percent of the world’s top tech executives say they are not ready for what is ahead. This isn’t a distant problem. AI agent deployment is expected to hit its organizations within a year. For an industry that prides itself on anticipating change, this is a real wake-up call. 

This finding comes from a new IBM study 2026, conducted by the IBM Institute for Business Value and Oxford Economics. Researchers surveyed 2,000 C-level tech executives from 33 countries and 19 industries earlier this year. The results show that the leaders responsible for enterprise AI lack control. 

The Gap Between Mandate and Capability 

Eighty percent of those surveyed said their CEOs have told them to speed up AI transformation, but only 11% feel fully ready for the scale of AI agent deployment expected next year. The numbers are clear: executives are being pushed to move quickly on a path they can barely see. 

70% of executives said their teams are deploying AI faster than IT can keep up with. Two-thirds of CIOs and CTOs said they are responsible for AI systems they do not fully control. These are not junior staff; they are the top tech officers in their companies, yet they are approving results from systems they cannot fully audit, monitor, or govern. 

Matt Lyteson, CIO at IBM, described the problem in a way that should concern any board. He said tech leaders underprepared for this shift need to rethink how their organizations control and manage AI financially. The goal, he said, is “embedding control and visibility from the start, so they can scale with confidence.” The warning isn’t about technology failing. It’s about human-speed governance being overwhelmed by systems that move at machine speed. 

When Governance Can’t Keep Pace 

77% of organizations said AI adoption is already outpacing their current governance. This shows a structural problem that has been growing for years. Companies built their compliance, audit, and risk review processes within a world where new software took months to deploy. AI agent deployment moves much faster. 

Here’s a real-world example. A financial services firm lets an AI agent handle customer loan evaluation. Six months later, the agent has made 400,000 decisions. In a board meeting, the CTO is asked to explain those decisions. She cannot, at least not fully, because the oversight system her team uses was built for quarterly reviews, not for instant autonomous systems at scale. 

This is the gap tech leaders are now underprepared for in the agentic era. It is not simply operational friction. The artificial intelligence corporate governance risks embedded inside this dynamic are existential for some organizations: regulatory exposure, brand damage, and financial losses from systems that optimize for the wrong outcomes before anyone notices. 

The Structural Performance Divide 

The IBM study for 2026 does more than point out a problem. It clearly shows the cost of doing nothing. Organizations that build control into their AI systems deploy 16 times more agents than those using manual governance, achieve 18% higher operating margins, and spend four times less on their AI budgets. 

This performance gap changes the governance of conversation. Corporate governance risks with artificial intelligence are not just legal or regulatory issues for the risk officer. They affect profit margins. Companies that treat governance as an afterthought pay four times more for slower, smaller deployments. Those that build control into their systems from the start scale faster and earn more. 

The companies making progress are not trying to overhaul all of IT. Instead, they are making targeted investments in adaptable infrastructure, governance by design, and portfolio discipline. These three pillars work together to build structural readiness. 

What Lyteson’s Warning Really Means 

Lyteson’s concern that machine-speed systems overwhelm human-speed architectures is not merely a theory. It’s a real issue that CIOs are dealing with right now, whether they succeed or not. Many tech leaders who are not ready for this shift still use IT systems built for stability, governance models that rely on manual review, and investment plans designed for multi-year asset lifecycles. These approaches cannot keep up with the speed of AI. 

By 2027, executives expect a 38% increase in the number of AI agents in their organizations. This trend means the governance gap will not close on its own. It will only get worse. Every quarter a company delays building control systems is another quarter in which AI agent deployment outpaces the oversight architecture. 

The Accountability Reckoning 

The IBM study for 2026 highlights corporate governance risks associated with artificial intelligence, already prompting greater regulatory accountability. European Union’s AI Act, new U.S. state-level AI liability proposals, and the SEC’s growing disclosure rules all focus on one question that 89% of tech executives cannot answer confidently: Who is responsible when your AI system causes harm? 

Regulators will eventually make the answer clear: the executive is responsible. This means the 89% of tech leaders who are not ready for large-scale AI agent deployment are not merely facing an operational problem. They are taking on personal and institutional liability every day they wait. 

The executives who keep their reputation and their companies strong will be those who see governance not as an obstacle, but as the foundation for large-scale deployment. The gap in structural readiness is real, and the data closing it is clear. The only question left is whether leaders will act before the AI agents do. 

Source: ESGDIVE 

Where Amazon Spent Billions Creating Jobs To Upgrade Online Fiber 

Charlotte, North Carolina  

The next big competition in artificial intelligence will not just be about computer chips. It will also be about glass. 

On June 8, 2026, Amazon spent a staggering, undisclosed investment, described by both companies as a multibillion-dollar commitment, to secure a key part of America’s optical fiber supply. The focus was Corning Incorporated, a 175-year-old specialty glass company whose cables already support some of the world’s top computing centers. The result, announced at Corning’s New York headquarters, is a multi-year agreement that will shape how the next generation of cloud systems are built, connected, and protected from foreign disruptions. 

This is more than a typical vendor contract. It shows the direction American industry is taking. 

How Amazon Spent Its Way Into a Fiber Lock 

Under the deal, Corning will provide optical fiber, cable, and connectivity solutions for Amazon’s growing data center infrastructure across the United States. While the exact amount was not shared, both companies confirmed it is worth several billion dollars. Corning’s stock rose by as much as 9.5 percent after the news, showing how much this agreement changes Corning’s business outlook. 

At the center of the Amazon multibillion-dollar Corning optical fiber procurement contract details is a straightforward strategic logic: Amazon’s data centers need more fiber than the global market can reliably provide. Instead of competing for a limited supply, Amazon chose to secure its own domestic source. This deal shows that the next bottleneck in AI expansion is not computer chips, but the glass that connects them and which company controls that supply. 

That glass is widely used in modern large-scale data centers. Every server rack uses fiber strands that are thinner than a human hair to communicate. As AI tasks require faster connections between graphics processing unit clusters, connections measured in petabits per second—the amount of fiber needed per square foot of data center space has increased significantly. Amazon, which operates one of the world’s largest cloud networks through AWS, can no longer overlook the importance of data center cabling. 

Manufacturing Job Creation in North Carolina: The Ground-Level Impact 

The impact of this Corning fiber agreement on people is just as important. The investment will create 1,000 new manufacturing jobs at Corning’s North Carolina facilities, plus hundreds of construction jobs for expanding those sites. Manufacturing salaries are expected to be over $65,000, which is much higher than the state’s average manufacturing wage. These jobs are considered long-term careers, not temporary positions. 

AWS CEO Matt Garman framed the deal in explicitly economic terms. “Amazon’s investments in North Carolina have created more than 26,000 jobs across the state. This multibillion-dollar agreement with Corning continues that commitment, channeling investment into American manufacturing and creating 1,000 new jobs at their facilities near our data centers. Location is just as important as numbers. The fiber Corning makes in North Carolina will be used in Amazon’s data centers in the same state. This creates a closed loop of production and use, protecting both companies from the overseas shipping delays that affected supply chains during the pandemic. The Amazon agreement is in addition to the company’s earlier plan to invest $10 billion in North Carolina to grow its cloud computing infrastructure. Altogether, Amazon’s total investment in North Carolina has now passed $20 billion since 2010.since 2010. 

Manufacturing job creation of this scale does not materialize overnight. Corning will have to expand its current plants and build new ones to meet the contract’s demands. To help train enough workers, Amazon and Corning will expand the Fiber Optic Technician Training Program at Catawba Valley Community College, preparing students for jobs in fiber-optic manufacturing and fusion splicing. This partnership shows that the real challenge is not just having enough fiber, but also having enough skilled workers to produce it. 

Supply Chain Strength: Why Domestic Production Changes the Risk Calculus 

The supply chain strength argument behind this deal is not theoretical. When COVID-era shipping disruptions cascaded through global electronics supply chains in 2021 and 2022, American hyperscalers discovered how exposed they were to overseas component production. Optical fiber, much of it historically sourced from Asian manufacturers, proved vulnerable to exactly those pressures. 

The Corning fiber agreement is one of several major hyperscaler moves to address that exposure. In January 2026, Corning signed a supply agreement with Meta worth up to $6 billion. In May, Nvidia and Corning announced a partnership to expand U.S.-based manufacturing of advanced optical connectivity, with Corning committing to increase its domestic optical-connectivity capacity tenfold and its U.S. fiber-production capacity by more than 50%, including three new plants in North Carolina and Texas. 

Amazon’s deal lands as the third major hyperscaler pledge to Corning in a single year. Together, these agreements are changing where American fiber gets made and who controls access to it. Corning’s Optical Communications sales grew 36% year over year in the first quarter of 2026, a figure that illustrates the severity of the demand wave now hitting domestic producers. 

What stands out about this deal for supply chain strength rests in its focus on location. By making fiber in North Carolina and using it in Amazon’s North Carolina data centers, lead times are shorter, logistics are simpler, and sensitive network equipment stays within a secure, domestic area. For a company like Amazon, which serves federal agencies, hospitals, and banks through AWS, this closeness is not just efficient; it is also a security measure. 

Upgrade Online Fiber: What the Infrastructure Buildout Means for American Industry 

Amazon’s decision to upgrade online fiber network in the United States rather than sourcing it from abroad signals a major shift in how the tech industry views physical infrastructure. For many years, the industry assumed that materials could always move easily and cheaply across global markets. That is no longer the case. 

Corning CEO Wendell Weeks described the agreement as a major turning point for both Corning and American manufacturing, saying it helps build a stronger U.S. supply chain. This way of talking using expressions like resilience and national capacity shows how much the global political situation has changed how big companies make purchasing decisions. 

For American workers, especially in North Carolina’s manufacturing regions, this means a steady flow of skilled technical jobs that were not available a year ago. For Amazon’s competitors, the message is also clear: companies that secure domestic fiber supplies early will have a significant advantage in quickly building infrastructure for the rest of the decade. 

The billions Amazon spent on this Corning fiber agreement represent more than a procurement decision. They represent a calculated bet that the companies that control the physical parts of AI infrastructure, the glass, the cables, and the connectors will have an advantage that software alone cannot match. When the next global supply chain problem occurs, Amazon plans to source its fiber from a plant just 40 miles from its own servers. 

This is not simply a strategy for data center infrastructure. It is a form of industrial policy, shown by the money spent and its impact on North Carolina. 

Source: https://www.aboutamazon.com/news/company-news/amazon-corning-fiber-optics-1000-jobs-north-carolina

How NetApp Teamed Up To Secure AI Factories Worth Multi-Millions

San Jose, California 

Last year, a major U.S. financial services company found that its proprietary trading algorithms, developed over four years, were accessible to an AI model training pipeline on the same internal network. There was no outside breach or complex hacking. Instead, the problem was a misconfigured storage layer that the security team missed because there was no policy for AI workloads. The incident cost the company about $47 million in fixes, regulatory work, and delayed product launches. 

This situation is not simply a rare warning. It shows a real risk present in thousands of enterprise data centers today. NetApp teamed with Cisco to solve this problem, not by releasing a software patch, but by redesigning hardware and administrative frameworks to secure AI factories before the next training run starts. 

How NetApp Teamed With Cisco to Build Secure AI Factories 

The partnership led to a new generation of FlexPod solutions. These are converged infrastructure stacks that combine Cisco’s Unified Computing System servers and networking with NetApp’s ONTAP storage operating system. The earlier FlexPod version worked well for traditional workloads in enterprise data centers. The new design tackles a different challenge: AI compute clusters packed with GPUs that process huge datasets and face ongoing regulatory scrutiny. 

NetApp and Cisco FlexPod AI data engine infrastructure operates as a single, integrated system rather than separate parts managed independently. This matters in practice. When storage, computing, and networking are managed together, security policies apply to all three at once. For example, a rule that limits access to personal data updates storage permissions, network rules, and compute controls in one step, rather than requiring three separate updates across different systems. 

For companies in regulated markets like healthcare (under HIPAA), finance (under SOX), or defense (under CMMC), this consistency is essential. It can mean the difference between meeting compliance requirements and leaving a gap that an auditor could find. 

The Intelligent Data Infrastructure Layer Inside the AI Data Engine 

At the core of the joint architecture is what NetApp calls its intelligent data infrastructure. This is a storage and data management system built specifically to meet the demands of large-model training in enterprise settings. 

Training a big language model on company data creates data movement patterns that traditional storage systems were not built to handle. In one training run, a model might read the same dataset hundreds of times in random order, pulling from different storage levels at once, while the data keeps being updated by production systems.ion systems. Standard access controls, made for people making single-file requests, do not work well with this kind of fast, machine-driven, parallel access. 

NetApp’s intelligent data infrastructure solves this by treating AI workloads as a separate access class with its own policies. Data marked for AI training moves through special pathways, with cryptographic checks at each step. This ensures that a model using a selected dataset cannot accidentally access nearby data stores containing sensitive or regulated information. The system keeps unchangeable access logs, so compliance teams have a clear record for audits without needing to document every data movement by hand. 

Enterprise Security Architecture and the Governance Gaps AI Exposes 

Five years ago, enterprise security documents did not mention the governance gaps that AI workloads now reveal, because these workloads were rare. Security teams built their frameworks to handle known threats such as external attacks, stolen credentials, or insider data exfiltration through clear human actions.  

AI training pipelines create a new kind of risk that does not fit into the usual categories. A model trained on internal data does not steal information in ways that traditional data loss tools can detect, but it can still encode sensitive patterns in its weights that might later appear in its outputs. For example, a customer service model trained on unedited support tickets might start giving answers that mention details from past customers. A financial prediction model trained on unmasked deal data could produce projections that reveal confidential information. FlexPod solutions handle this at the infrastructure level by defining data boundaries before training starts, rather than trying to catch leaks after deployment. deployment. The system’s policy engine sorts data by sensitivity and uses that to decide what can go into each training dataset, blocking any uncleared data, no matter how the training job asks for it. 

This is what secure AI factories look like in real life: instead of just building a perimeter defense around an AI system, there is a governance layer built into the data infrastructure itself. 

What the NetApp and Cisco FlexPod AI Data Engine Infrastructure Needs From Enterprise Teams 

NetApp and Cisco FlexPod AI data engine infrastructure guide asks organizations to do something most IT departments have put off: create a formal data inventory with sensitivity labels for each asset before any AI workloads use it. This requirement brings to light governance gaps in corporate data management that have existed for years. Most big companies have decades of data stored without appropriate classification. Switching to a secure AI factory setup forces this classification work, which may be expensive and uncomfortable for organizations, especially when it shows that some data has been stored, shared, or accessed beyond its original consent or regulatory limits. 

Enterprise Security architecture teams that see this as just an infrastructure project may not realize what it really takes. The hardware and software are ready to use, but the real challenge is the ongoing discipline needed to classify, manage, and audit the data going into AI systems. That is where most companies will find the real work starts. 

The companies that complete that work first will operate AI infrastructure that regulators and auditors can properly review. As AI governance laws advance in the U.S., the EU, and Asia-Pacific, the ability to audit systems may become just as important as the AI features themselves. NetApp and Cisco built the vault, but it is up to each company to fill it the right way. 

Source: https://www.businesswire.com/news/home/20260603146499/en/NetApp-and-Cisco-Accelerate-and-Secure-AI-Innovation 

Where Workday Opened Up Hidden Rules To Fast-Track Smart App Building

Las Vegas, Nevada 

A mid-sized logistics company with 4,000 employees recently had to wait 11 weeks for its tech team to set up a new overtime-calculation module in its HR platform. This delay forced the company to manually correct two payroll cycles and triggered a compliance flag from its finance auditor. Unfortunately, this kind of timeline is common in enterprise software development, where even small changes mean working through complicated backend systems that most developers don’t fully understand. 

Workday opened a direct line through that complexity at its annual Workday DevCon conference in Las Vegas, unveiling a Developer Agent embedded in the Workday Build Platform that accepts plain-English instructions and converts them into production-ready code. The announcement marks a meaningful change in how corporate engineering teams approach smart app building inside one of the world’s most widely deployed enterprise software ecosystems. 

How Workday Opened the Workday Build Platform to Plain-Language Development 

The Developer Agent works as a built-in AI assistant within the Workday Build Platform. Now, a developer using Claude Code or Cursor, two popular AI coding tools in businesses, can describe a workflow in plain English and get a complete Workday Object Definition Language output in response. 

Here’s how it works in practice. Imagine a payroll specialist at a healthcare network who needs to set up a PTO rule that treats salaried nurses on rotating twelve-hour shifts differently from regular administrative staff. Before, this engineer would spend days reading API guides, checking Workday’s data structure, and writing and testing scripts by hand. Now, with the Developer Agent, the engineer can simply describe the logic in plain language and get a ready-to-use script in minutes, already checked against Workday’s data model. 

This time savings is real. According to Workday’s own benchmarks shared at WorkDay DevCon, tasks that used to take three to four weeks of backend engineering can now be done in less than an hour with the Developer Agent. For companies using Workday for HR, payroll, and finance, this speed can change what small development teams can deliver. 

AgentSkills Open Standard and the Architecture of HR Tech Automation 

The bigger announcement at Workday DevCon wasn’t merely the Developer Agent, but the AgentSkills Open Standard that supports it. 

Workday launched AgentSkills as an open standard, allowing external developers to create modular AI features that plug directly into the Workday system without requiring Workday’s engineers to build or maintain them. It serves as a shared language for HR tech automation: external developers define skills using the standard, which are then available to the Developer Agent as tools during setup. 

This move has big strategic effects. By making the standard open rather than keeping it private, Workday invites its entire developer community, including tens of thousands of certified implementation partners worldwide, to extend the platform’s AI capabilities. A benefits administration firm specializing in COBRA compliance could write an AgentSkills-compatible module that any Workday developer could use. A workforce analytics startup could publish a scheduling optimization skill that slots directly into the same interface that developers already use for payroll. 

This approach enables smart app building across the whole ecosystem, not just for individual tools. 

Why Workday DevCon Developer Agent Platform Configuration Changes Enterprise Timelines 

The Workday DevCon Developer Agent platform configuration guide addresses a bottleneck that enterprise CIOs have complained about for years without a credible solution: the gap between business requirements and deployed functionality. 

Business leaders, like a VP of People Operations who needs a new headcount dashboard or a CFO who wants real-time labor cost alerts, usually describe their needs in business terms. Developers then turn these needs into technical specs, write the code, and translate it back into business language for testing. Each step can cause delays, confusion, and extra work. 

The Developer Agent removes the first translation step. Now, a developer can type something like “build a report that flags any department where headcount exceeds approved budget by more than 5 percent, refreshing every Monday morning” and get a working Workday configuration. This makes the gap between business goals and technical results almost disappear. The rest of the process testing, governance review, and deployment approval still happens, but it starts with a validated setup rather than a blank file. 

For HR tech automation specifically, this matters because HR workflows carry legal and compliance weight. A misconfigured FMLA tracking rule does not just create inconvenience it creates liability. Speed is valuable; speed built on a platform that validates configuration logic against a known-correct data schema is what actually moves the needle in regulated industries. 

The Risk Embedded in Democratized Smart App Building 

Workday opened access to sophisticated backend tooling to a larger developer population, which carries a genuinely mixed risk profile. Faster configuration means more configurations  and more configurations mean more surface area for errors that governance teams need to catch before they reach production. 

The AgentSkills Open Standard adds a further layer of complexity. Third-party skills bring outside logic into a platform that customers have frequently chosen for its closed and auditable design. Workday’s documentation includes certification rules and sandbox testing, but the responsibility for checking third-party skills still falls on the IT teams that use them. 

These risks do not invalidate the approach. Instead, they highlight the real-life challenges that enterprise architects will need to manage as the Workday Build Platform grows. 

Companies that learn to combine faster configuration with strong governance will have a lasting advantage over those still waiting weeks for new modules. The Workday DevCon Developer Agent platform has made the technical side much easier. The challenge of management and oversight still remains.

Source: https://newsroom.workday.com/2026-06-02-Workday-Launches-New-Tools-for-Developers-to-Build,-Connect,-and-Verify-AI-Agents-For-HR,-Finance,-and-IT