What many people forget when talking about the benefits of AI coding initiatives is that (a) coding is only one piece of the puzzle and (b) building good apps is a team sport. Those two truths mean that a 50% improvement in dev productivity will not translate into 50% better time-to-market, reduced costs, etc. etc. etc. I believe that the proper injection of AI to incorporate scalable customer feedback will create better products at higher velocity and quality. I also firmly believe that this is a transformation that could dwarf the agile movement. But first we need the technologies and processes to help product managers, product marketing managers, and product owners get the right requirements fed into the toolchain.

For that reason, I have been keeping an eye out for technologies that enable better work up front in the design and planning process — before the AI starts making code. Miro is a leader on that list. Even though it’s pretty easy to connect to JIRA via MCP, that can end up being just a lot of forms lacking nuance. If I am out there as a product manager, marketer, or owner, I’d rather just integrate the artifacts straight into the flow versus typing them into JIRA. Placing live insights and artifacts alongside my prioritized requirements will be key to driving context for agents to build the new apps. Miro helps you create great artifacts, but so far those have been mostly document-driven. So, I was intrigued when Miro announced its intent to acquire Reforge. Reforge’s playbooks, surveys, and customer feedback collection capabilities are a really nice complement to what Miro provides, and it’s worth a double-click.

A question I get a lot from clients trying to scale their AI efforts is “How can we attract and retain enterprise customers?” My conclusion is that while the technology is quite different, the buying process is not. And while security and pricing models for agentic solutions are at the top of the list of concerns, there are even more issues that lurk underneath. For instance, I am starting to also wonder about longevity and supportability as reasons to buy or not buy. When OpenAI decided to discontinue the Sora video platform and app last week, it stuck out as an interesting decision. From a business perspective, OpenAI is walking away from a consumer-oriented business that was not getting traction, which makes business sense. I am going to assume that the company is reallocating its resources on more enterprise-centric capabilities like Codex. But in doing that, it is also signaling a behavior that enterprises hate. All too often, we are seeing AI companies unwilling to share roadmaps more than a quarter out, with no commitments to lifecycles, migration paths (even basic guidance), or even a replacement product strategy. Enterprises do not like change and really don’t like rapid change. So, let me make it easy for everyone: If you can’t commit to a lifecycle and roadmap, you can’t sell your product to enterprises in a meaningful way.

Last week I was introduced to another advanced agentic use case. Overall, I am excited that vendors are starting to engage in specific storytelling on how AI is actually being used. Ideally, there is a customer name attached to the story, but that is a very tricky proposition that can take some vendors months, and often the detail has to be light to protect customer interests. Sometimes a “headless” reference case is just as good, especially when it allows the vendor to share more technical depth and texture.

Which brings me to the new use case from MongoDB that lays out an agentic loan process. I just bought a house, and the financing steps were the most fraught aspects of the entire process. Part of the challenge was corralling a ton of unstructured data and then manually typing it into a range of forms and other electronic documents. This use case really shows how agents can be used to collect disorganized and unstructured data, then extract and transform it so users don’t need to have three windows open to make sure the data gets in there.

Also, the seamless integration between AI outputs and structured data processes stands out. We need to get to an understanding that 100% generative AI agentic apps will be very expensive to run at scale. But agents paired up with lower-cost, higher-accuracy RPA or other deterministic apps will hit the new sweet spot of customer experience, ease of use, and lower costs that enterprises want. And frankly, this is also a great case for why data platforms will ultimately transform into context platforms over time. In this connection, MongoDB may not be among the first couple of names you hear when it comes to data platforms, but its Atlas platform is coming on strong.

What many people forget when talking about the benefits of AI coding initiatives is that (a) coding is only one piece of the puzzle and (b) building good apps is a team sport. Those two truths mean that a 50% improvement in dev productivity will not translate into 50% better time-to-market, reduced costs, etc. etc. etc. I believe that the proper injection of AI to incorporate scalable customer feedback will create better products at higher velocity and quality. I also firmly believe that this is a transformation that could dwarf the agile movement. But first we need the technologies and processes to help product managers, product marketing managers, and product owners get the right requirements fed into the toolchain.

For that reason, I have been keeping an eye out for technologies that enable better work up front in the design and planning process — before the AI starts making code. Miro is a leader on that list. Even though it’s pretty easy to connect to JIRA via MCP, that can end up being just a lot of forms lacking nuance. If I am out there as a product manager, marketer, or owner, I’d rather just integrate the artifacts straight into the flow versus typing them into JIRA. Placing live insights and artifacts alongside my prioritized requirements will be key to driving context for agents to build the new apps. Miro helps you create great artifacts, but so far those have been mostly document-driven. So, I was intrigued when Miro announced its intent to acquire Reforge. Reforge’s playbooks, surveys, and customer feedback collection capabilities are a really nice complement to what Miro provides, and it’s worth a double-click.

A question I get a lot from clients trying to scale their AI efforts is “How can we attract and retain enterprise customers?” My conclusion is that while the technology is quite different, the buying process is not. And while security and pricing models for agentic solutions are at the top of the list of concerns, there are even more issues that lurk underneath. For instance, I am starting to also wonder about longevity and supportability as reasons to buy or not buy. When OpenAI decided to discontinue the Sora video platform and app last week, it stuck out as an interesting decision. From a business perspective, OpenAI is walking away from a consumer-oriented business that was not getting traction, which makes business sense. I am going to assume that the company is reallocating its resources on more enterprise-centric capabilities like Codex. But in doing that, it is also signaling a behavior that enterprises hate. All too often, we are seeing AI companies unwilling to share roadmaps more than a quarter out, with no commitments to lifecycles, migration paths (even basic guidance), or even a replacement product strategy. Enterprises do not like change and really don’t like rapid change. So, let me make it easy for everyone: If you can’t commit to a lifecycle and roadmap, you can’t sell your product to enterprises in a meaningful way.

Last week I was introduced to another advanced agentic use case. Overall, I am excited that vendors are starting to engage in specific storytelling on how AI is actually being used. Ideally, there is a customer name attached to the story, but that is a very tricky proposition that can take some vendors months, and often the detail has to be light to protect customer interests. Sometimes a “headless” reference case is just as good, especially when it allows the vendor to share more technical depth and texture.

Which brings me to the new use case from MongoDB that lays out an agentic loan process. I just bought a house, and the financing steps were the most fraught aspects of the entire process. Part of the challenge was corralling a ton of unstructured data and then manually typing it into a range of forms and other electronic documents. This use case really shows how agents can be used to collect disorganized and unstructured data, then extract and transform it so users don’t need to have three windows open to make sure the data gets in there.

Also, the seamless integration between AI outputs and structured data processes stands out. We need to get to an understanding that 100% generative AI agentic apps will be very expensive to run at scale. But agents paired up with lower-cost, higher-accuracy RPA or other deterministic apps will hit the new sweet spot of customer experience, ease of use, and lower costs that enterprises want. And frankly, this is also a great case for why data platforms will ultimately transform into context platforms over time. In this connection, MongoDB may not be among the first couple of names you hear when it comes to data platforms, but its Atlas platform is coming on strong.

OpenAI has killed its Sora video platform, which I imagine was probably one of the company’s deepest money pits. That said, OpenAI was working with Disney and didn’t tell Disney that it was going to discontinue the Sora service before it told the world. I think this will be a big problem for OpenAI because it will need some kind of video-generation model in the future if it wants to be a mainstream AI company. Also, I think OpenAI’s sudden rug-pull on Disney will likely affect how others partner with OpenAI in the future. Disney was expected to invest as much as $1 billion in OpenAI through this partnership, and now that all seems to have dissolved. I get that OpenAI wants to focus on foundational models, but it cannot give up on partnerships, especially not ones that have companies as big as Disney involved.

I wasn’t surprised to hear that Google Quantum AI is set to begin the development of a neutral-atom quantum computer — a departure from its traditional focus on superconducting qubits. It is likely that other quantum companies are quietly experimenting with modalities other than the technology they were funded for. In this case, co-development of neutral atoms and superconducting qubits can be complementary. Even though a superconducting quantum computer can rapidly create deep circuits using millions of gate cycles at microsecond speeds, scaling up to thousands of qubits can be a problem.

On the other hand, neutral atoms can hold arrays of thousands of qubits with flexible connectivity, but they struggle with achieving the circuit depth of superconducting machines. Google is betting that it will be able to advance both technologies simultaneously by allowing each modality to compensate for the other’s weakness. At the same time, it may also be able to share research breakthroughs between the two teams.

Dr. Adam Kaufman, a prominent physicist at JILA (formerly the Joint Institute for Laboratory Astrophysics) at the University of Colorado Boulder, will be heading up the Google effort. Kaufman will retain his faculty affiliation while building a dedicated hardware team at Google, which will also continue collaboration with portfolio company QuEra.

Supporting two fundamentally different hardware programs simultaneously can strain engineering resources, talent, finances, and coordination. There’s also a possibility the two teams could cannibalize each other’s talent pipelines or produce competing internal roadmaps. On the other hand, a breakthrough in one modality could be useful for the other, particularly in error correction architectures.

Afeela, the joint venture between Sony and Honda, will no longer produce any cars, including its first model, for which the joint venture was already taking orders — and which was expected to ship this year. Additionally, all future car development has ended, and the brand and joint venture seem to have reached an end altogether. The venture was established to develop electric vehicles that brought together Honda’s automotive know-how with Sony’s experience in electronics. The main problem was that the starting price for the first vehicle was $90,000, and for a luxury product like that, people tend to prefer established brands. Not to mention, the vehicle’s range wasn’t very competitive at that price, plus I think there were concerns about long term viability, especially after EV pullbacks from larger makers including Honda, VW, GM, Ford, and Stellantis. The Afeela brand and vehicles were unique and fun in their own way with features like a built-in PlayStation, but ultimately they weren’t that competitive, and I think Sony and Honda realized that. Also, Sony seems to be on a path to cut businesses that aren’t as profitable, which it recently did by spinning off its TV unit into a joint venture with competitor TCL.

Oracle has introduced its Private Agent Factory as part of its broader AI Database 26ai strategy, positioning it as a low-code/no-code platform for building, deploying, and managing enterprise AI agents directly within the data layer. The platform integrates prebuilt agents, vector search, and workflow orchestration into the database, allowing enterprises to create data-driven agents without exposing sensitive data to external systems. The emphasis here is running AI “close to the data,” reducing the need for complex pipelines while maintaining governance, security, and performance.

From a competitive standpoint, Oracle is taking a different approach than hyperscalers and AI startups that rely on external orchestration layers and fragmented data stacks. By embedding agent orchestration, memory, and security directly into the database, Oracle is positioning the data tier as the control plane for agentic AI. This contrasts with platforms like Databricks and Snowflake, which emphasize open data layers, as well as emerging agent frameworks that sit above the data rather than within it. The strategy also leverages Oracle’s installed base, where so much enterprise data already resides.

There is a lot of goodness in what Oracle launched. And for those competitors that argue that enterprise data doesn’t exist solely in an Oracle database, the company would counter with the Autonomous Data Lakehouse. This enables the real time import of data from virtually any data source or repository.

My thoughts on this? Oracle is making a deliberate architectural bet — and that bet is that AI in the enterprise will be won at the data layer, not the model layer. If that holds, Private Agent Factory could reduce friction for companies moving from AI pilots to production by collapsing the stack. The open question is whether enterprises will prioritize that level of integration over the flexibility of more modular approaches.

I expect this will resonate with organizations that have adopted Oracle, and those in industries where data governance and control are critical.

Arm has formally entered the merchant silicon market with its AGI CPU, marking a shift from IP licensing to delivering production silicon. The processor targets agentic AI environments with a 136-core design, high memory bandwidth, and aggressive efficiency claims, including up to 2x performance per rack versus x86 and significant capex savings at scale. Meta is a lead co-design partner, underscoring the chip’s alignment with hyperscale AI infrastructure.

The AGI CPU should also find a market for traditional cloud workloads as smaller CSPs, hyperscalers, neoclouds, and even large enterprise IT organizations look to replicate the “Graviton model.”

From a competitive standpoint, this move places Arm squarely into the datacenter CPU battle against Intel and AMD, both of which are already evolving x86 architectures for AI-era workloads. Arm’s pitch centers on efficiency, density, and deterministic performance, while x86 incumbents retain advantages in ecosystem maturity, software optimization, and entrenched enterprise deployments. At the same time, hyperscaler adoption of Arm-based CPUs has already validated the architecture at scale, creating a more credible competitive baseline than prior Arm server efforts.

Arm may also find itself butting against some of its partners as companies like Marvell and Broadcom currently build custom silicon for larger-scale customers. However, I expect this will be a tension that will be limited in scope and very manageable.

This is a bold move. And it is a heavy lift to move into the merchant silicon market — especially for the datacenter. As we have witnessed with Qualcomm’s attempt to move into the client market, producing a high-quality chip is not enough.

Still, I expect Arm will find success beyond Meta and OpenAI. There is a real hunger in the market, and the disruptive nature of the agentic AI explosion will enable an easier path to success for the company.

I was at RSAC 2026 last week talking to more vendors than I can count. And it got me thinking about what’s missing in information security.

It’s not because the technology isn’t impressive. It is. At the show, there were more than 600 exhibitors interacting with 40,000-plus people. And the announcements kept coming: Google closed its $32 billion Wiz acquisition and rolled out an Agentic SOC. AccuKnox launched AI-Security 2.0. Dell Technologies doubled down on PowerProtect and cyber resilience across its storage portfolio. Infoblox is expanding into digital risk protection with its planned acquisition of Axur. AT&T is enhancing its security measures through its dynamic defense strategy. Zenity unveiled Guardian Agents. Spektion introduced runtime exposure management. TuxCare is tackling open source lifecycle risk. Intruder, InfoSEC, Commvault, Cohesity, Rubrik, Veeam Software, HYCU — these are just a few names from a long list of innovative companies in the information security industry.

But here’s what kept nagging at me. I sat through demos, briefings, and booth conversations where vendors showed me architecture slides, feature matrices, and competitive comparison charts. They walked through their platform, stack, integrations, AI capabilities, and dashboards. Yet very few of them started where the customer starts. What actually changes for me if I deploy this? What outcome are you delivering? How does my risk posture actually change? What does this do to my operational overhead? Can you show me how a CISO walks into a board meeting and explains the business value of what I just bought?

That gap between what vendors talk about and what buyers actually need is getting wider, not narrower. And that matters because the complexity is real. The M-Trends 2026 report — issued by the Mandiant unit within Google Cloud — tells us that attackers are compressing the time between initial access and execution to 22 seconds. Industry research shows that 43% of enterprises plan to add even more security vendors this year. We’re not simplifying anything. We’re stacking more tools onto environments that are already hard to run.

It’s not enough to just list features. I would prefer vendors to show how their AI and zero trust approaches actually reduce attack surface, save time, or lower costs in measurable terms. Plus how this all ties directly to business continuity, compliance, and trust. The vendors who figure this out will separate themselves. The ones who keep leading with architecture slides will wonder why their pipeline stalled. I’m putting together a deeper piece on RSA and what outcome-based security actually looks like. More soon.

Miro is acquiring Reforge, integrating its AI-driven product development tools and learning platform into Miro’s collaboration workspace for product teams. Financial terms and deal timing have not yet been announced, but the deal includes Reforge’s technology and team, with Reforge CEO Brian Balfour joining Miro as chief growth officer and COO Tom Willerer appointed as Miro’s chief strategy officer.

The move points to Miro evolving from digital whiteboarding to an AI-native innovation workspace. This means helping teams decide what to build, not just document how to do it. By absorbing Reforge’s product management methodology and AI tools, Miro is positioning itself across the full discovery-to-delivery lifecycle.

The acquisition reflects a broader shift in AI canvases, as horizontal collaboration platforms layer in specialized product intelligence to compete with suites like Microsoft 365 and Google Workspace at the point where discussions and planning occur and decisions are made. (For more on this, see my recently published article about AI canvases here.) For Miro, the opportunity is to translate Reforge’s community-driven rigor into scalable systems that sharpen practitioners’ insight. This would look like turning Reforge’s frameworks and community practices into in-product workflows and playbooks that effectively guide decisions. I will be watching with the hope that it can do that while preserving the depth and credibility that make Reforge valuable to users.

Data platforms are everywhere. What’s still missing is the ecosystem around them that actually connects the data, keeps it consistent, and makes it usable across the business.

ERP still captures the transaction. That’s not changing. But the context behind that transaction can live somewhere else. Demand patterns, cost drivers, delays, customer behavior — all of that is spread across CRM, supply chain, logistics, and external data sources that no single vendor owns.

ERP is evolving to be more event-driven, more agentic, and more connected. For example, Microsoft is bringing ERP data into Microsoft Fabric, while SAP is extending its Business Data Cloud beyond its own environment. Meanwhile, Snowflake, Databricks, Teradata, and Cloudera are all positioning themselves as the connective layer across systems. A lot of movement, a lot of overlap.

But here’s what I keep coming back to: The technology is moving faster than the governance. Data can lose its meaning when it moves between platforms. Definitions don’t align. Calculations break. Relationships get lost. And that’s exactly what vendors and enterprises alike need to be addressing.

The winners in this field won’t necessarily be the ones with the best platform. They’ll be the ones whose data stays consistent, governed, and usable as the ecosystem around it grows. Read more on this in my latest Forbes article, “How ERP Data Fits Into The Enterprise Data Ecosystem.”

SAP’s acquisition of Reltio is really about strengthening its data foundation. What SAP is working toward is more consistent, governed data across both SAP and non-SAP systems so that AI and automation can operate in real-time environments. This is not just for reporting, but for day-to-day decisions across finance, supply chain, and customer processes. The goal is to help enterprises reduce the amount of time spent reconciling data. It also extends SAP a bit further into the decision layer, not just the system of record.

Most customers are running mixed environments, so having a more unified way to manage data across systems matters. It also reflects what I’m seeing across the market. As AI gets embedded into workflows, data quality and consistency are becoming more visible issues. Reltio should be able to bring a more mature, cloud-native approach that SAP can build on rather than starting over.

With any acquisition, success comes down to how well it’s integrated into what customers are already running. The bigger issues I keep seeing are still around data governance and ownership, which ultimately drives how much value this delivers. The technology can help unify data, but it doesn’t replace the need for clear accountability and consistent processes.

This move makes sense for SAP. The impact will depend on how simply it fits into existing environments and how directly it connects to real operational outcomes.

Supply chains don’t need a full government shutdown to feel disruption. The partial DHS closure is a good example of how even limited constraints start to show up in day-to-day operations. What I’m seeing is less about any single issue and more about how quickly conditions shift while visibility lags. That’s where teams start to struggle. Planning cycles break down when inputs are incomplete or delayed, and decisions get made with less confidence than they should.

A clear example is what happened when the Census Bureau paused construction spending data. That data feeds directly into demand planning for materials like steel. Without it, producers were making sourcing and capacity decisions without a reliable signal. That’s not a dramatic disruption, but it’s enough to throw off execution.

At the same time, approvals and inspections are taking longer as agencies like FDA, EPA, and USDA operate with fewer resources. For regulated manufacturers, that creates delays that ripple into production schedules, supplier commitments, and inventory levels. Layer on tighter freight capacity and longer transit times, and the system has less room to adjust.

This is where technology should be doing more of the work. Not just visibility dashboards, but scenario modeling, exception management, and real-time signals across suppliers, logistics, and inventory. And vendors should be doing what they can to help customers get the most out of these functionalities. The goal is to help teams see issues earlier and make adjustments before they turn into larger problems. 

The challenge is that many organizations still rely on fragmented data and disconnected workflows. When conditions are stable, that can work. When things start shifting, it breaks down quickly. What tends to work better in these environments is staying close to what you can actually see and control. Keep alignment with key suppliers. Understand where dependencies exist, especially with government processes. Make decisions based on current signals, not assumptions about where things will settle.

This isn’t about large transformation efforts. It’s about being steady, informed, and responsive. The teams that handle this well are the ones that stay connected to their data and can adjust without overreacting.

With the beginning of the 2026 season, Major League Baseball has upgraded the fan experience for VisionOS. This clearly indicates not only that the league is continuing to invest in XR as a whole, but that it wants to improve the experience for that slice of its fan base as it iterates towards what I believe will inevitably be an AR implementation. The future development path for AR glasses on Apple will go through the Apple Vision Pro, which is why I’m glad to see the company continuing to work with developers to add enhancements that make the user experience more enjoyable and immersive.

On March 16, 2026, both NXP and STMicroelectronics announced solutions incorporating NVIDIA’s Holoscan Sensor Bridge (HSB). This big bet on distributed physical AI reaches all the way down to real-world devices that sense and act in real time.

HSB accelerates connections between physical-world devices (sensors and actuators) and NVIDIA’s AI compute engines by creating a direct, low-latency data path to GPU memory, reducing system complexity, latency, and cost. In other words, it connects the real world directly to GPU-powered intelligence, minimizing reliance on intermediate software stacks and gateways. Holoscan is a nervous system that connects sensors and actuators directly to AI brains over standard Ethernet.

NXP announced integration of HSB across its edge processing and networking portfolio, including the i.MX 95 applications processor, i.MX RT1180 crossover MCU, S32J TSN switch, and Aviva Links high-speed asymmetric connectivity chips. The result is a unified real-time substrate for sensing, data movement, and actuation that aligns with NVIDIA’s compute stack. In effect, this creates a real-time sensor layer for physical AI.

STMicroelectronics announced HSB solutions across its STM32 microcontrollers, sensor portfolio (IMUs, image sensors, and time-of-flight depth sensing), and motor-control solutions. The company also offers sensor simulation models that enable NVIDIA Isaac Sim to model sensor behavior before prototyping. In parallel, Leopard Imaging and ST introduced a multimodal vision system combining imaging, 3-D scene mapping, and motion perception with HSB Jetson connectivity.

Connecting the physical world directly to GPU DRAM is a significant step in the evolution of edge AI systems. The direct path from sensor to memory — at wire speed — changes the latency, complexity, and cost profile for real-time perception and control applications. Most deployment examples for this technology are humanoid robotics, and that’s certainly a familiar and compelling entry point. Still, the technology applies to any situation that requires high-bandwidth, low-latency device-to-AI connectivity: automotive, manufacturing, healthcare, infrastructure, and more. Robots can take on any physical embodiment, with perception and motor control directly connected to AI brainpower via standard Ethernet.

I’ll leave you with this example of a compelling use case from my recent conversation with Gowri Chindalore, an NXP VP who leads business, technology, and AI strategy for the company’s Secure Connected Edge solutions. He said robots (humanoid or not) need to communicate effectively with humans, and that requires more than words. A listener’s facial expressions can cue a speaker’s speech pattern in less than 200 milliseconds. That’s how you know, for example, that someone is just about to interrupt you, and you act accordingly. Human interface developers are turning to visual cues to make robotic engagements more human-like, which requires an upgrade in perception latency. That’s what Holoscan does, and NXP and STMicroelectronics are making it practical.

NXP Announcement: NXP Delivers New Innovations for Advanced Physical AI with NVIDIA

STMicroelectronics Announcement: STMicroelectronics and Leopard Imaging accelerate robotics vision with NVIDIA Jetson-ready multi-sensor module – ST News

The FCC’s decision to ban any new foreign-made routers blindsided the industry and really didn’t take industry input into account. The ban cites security risks as the rationale for banning any new routers made abroad, but it doesn’t do anything to help companies onshore manufacturing or address the primary issue, which is that router firmware is inherently insecure — and nobody is paying router companies to keep their firmware up to date and secure. I believe this process could have been accomplished more effectively with a multi-year transition and more accountability for firmware vulnerabilities, but I don’t think current FCC leadership is interested in that approach. This has also created alarm among some consumers who think they won’t be able to buy any routers in the future because pretty much all consumer routers are currently made abroad.

HP announced HP IQ at its Imagine conference in New York City last week. IQ is the company’s new AI suite that leverages on-device compute paired with open models like OpenAI’s GPT-OSS 20B to perform tasks such as summarizing documents, taking meeting notes, and enhancing device-to-device connectivity by automating some wireless processes. While HP seemed to communicate that this was for enterprise users, I think it will be better suited for SMB users who simply don’t have the time to set up AI tools and models — yet can benefit from the enhanced privacy and security.

Atom Computing and Cisco signed an MOU to test distributed quantum computing architectures using network-neutral-atom quantum processors. The partners also plan to test technical issues such as linking Atom’s neutral-atom quantum processing units together and extending Cisco’s network-aware distributed quantum computing compiler to orchestrate workloads across multiple nodes.

By combining Atom’s scalable quantum hardware with Cisco’s quantum networking stack and protocols, it should be possible to test a full-stack distributed platform. If successful, this could prove that utility-scale quantum computing can be achieved through multiple networked, mid-sized processors rather than a single monolithic machine.

Moving from a standalone environment to a distributed quantum computing architecture is needed for distributed quantum datacenter architectures, and we can expect more of these tests in the future involving other quantum companies. Look for demonstrations of multi-node entanglement, workload partitioning across QPUs, and early full-stack prototypes.

At RSAC 2026, Cisco laid out a security model for the agentic workforce, introducing new Duo IAM and Secure Access capabilities that extend zero trust to AI agents and other non-human identities. Customers can register agents, map them to accountable human owners, and require all tool calls to flow through an MCP-based gateway with tight, short-lived permissions.

It seems like each week brings another “claw.” Cisco also announced its new DefenseClaw, an open-source secure agent framework that ties together skills scanning, MCP validation, AI asset inventory, and a sandboxed runtime with NVIDIA OpenShell. Cisco is framing this as a move to govern a mixed workforce of humans and agents (rather than defending a human perimeter) with controls embedded into identity, runtime, and development workflows. This also shows me that Cisco is wisely nudging security toward being a shared operational responsibility across the enterprise.

Elon Musk’s plan to build a “Terafab” to manufacture chips for SpaceX, XAI, and Tesla is one of the most ambitious (or, depending on whom you talk to, pie-in-the-sky) ideas he’s had so far. While I absolutely believe there is some value for Musk’s companies to build their own chips and own those capabilities, I don’t think that the $25 billion earmarked for the effort is enough based on the intended scale of producing 1TW of compute per year. There’s also the issue of tools and depreciation and how both of those will impact the financials of the Terafab and the companies buying from it. For this effort to be successful, the Terafab would need to be completely full for years purely from Musk’s own companies, which I think is unlikely. For a parallel example, just look at the challenges Intel is having as it tries to grow capacity and gain customers for its Foundry business — and Intel has a long legacy of cutting-edge semiconductor production.

This Opening Day, MLB officially launched the use of ABS (automated balls and strikes) into big-league play after testing it last year during spring training and in the minor leagues. The technology, which uses Sony’s Hawk-Eye system with 12 cameras paired with T-Mobile 5G networking, is delivering real-time results for ball and strike calls challenged by the pitcher, catcher, or hitter. I got to witness the system in use during a game last week, and it happened so quickly that I almost missed it — which is a good thing. I saw that the call was overturned and that the pitch was off by more than 1.5 inches, and then the game moved on, which is exactly how it’s supposed to work. I believe that this will only enhance the game of baseball and improve its accuracy, which is ultimately what fans and sponsors want to see. I also think that T-Mobile did a great job snagging the sponsorship, because ABS sponsored by T-Mobile is going to come up at least a few times every game. Each team gets two challenges, which they get to keep if the call is overturned, but lose if the call is confirmed.

My two-year-old Oura Ring 3 died on me, and based on my research, this a very common issue with the Oura Ring — and, I imagine, most wearables with such small batteries. If the battery ever gets discharged too low or for too long, it can struggle to regain charge, which is what I believe happened to my unit. Oura did offer me a discount of $200 off of a new $350 Oura, but I just acquired a Whoop health-tracking wristband, and I’m going to give that a try to replace the Oura for activity and sleep tracking. I suspect I might still prefer a ring for tracking, although I like it that Whoop requires just a monthly subscription — and, if your device breaks, they will send you a new one. So, do you want to pay $350 plus $70 a year for the Oura service? Or $360 a year for Whoop and never pay for the device itself?