Last week was Microsoft’s Ignite event and, like last year, there was a major focus on agentic development. Last year we saw the release of Azure AI Foundry, which was one of the first platforms for rich agent development, alongside a host of more canned agents to supplement existing Microsoft products. This year presented a major evolution in Microsoft’s agent strategy. For starters, we now have a complete suite of capabilities for agents delivered either through enhancements to existing products such as FoundryIQ (the new name for AI Foundry) or via net-new products and WorkIQ. While last year’s announcements were competitively solid, this year’s really stood out for how the entire agent story and product strategy hung together. It fits in very well with Microsoft’s target audience, which are enterprises seeking a total vendor solution versus a customized best-of-breed play.

What stood out most to me from Ignite was Agent 365 thanks to how Microsoft extended its existing agent operational capabilities. The concepts from the built-in observability in AI Foundry have been extended significantly. The agent observability features have been improved and now include a personalized developer dashboard to simplify a user’s ability to manage and monitor their own agent fleet. There’s also significant improvement in the catalog features, including an updated model comparison tool. But the biggest standout capabilities were for security; these are largely new features centered on agent-specific threats, which are becoming increasingly difficult for standard cybersecurity tools to assess and remediate. Looking back on the notion of agent management and operations, I am pleased to see that Microsoft was very deliberate in sticking to this path because it was an area of differentiation over the last year.

Google announced a preview of its own agentic development tool this week, called Antigravity. And as with many of the other agent IDEs out there, Google gave its own spin on the new age of application development. In line with other tools like GitHub Agent Mode, developers can spawn and oversee multiple agents developed asynchronously. There is also a feature called Artifacts, which seems like spec-based development where agents will document the pre- and post-activities to help ensure a better-quality deliverable — very similar to what we have seen with AWS’s Kiro IDE. And while I tend to think that the continued evolution of AI in the developer space is possibly one of the most interesting AI use cases, I also would prefer to see more cooperation in the tooling space, as opposed to everybody having their own IDEs — which are starting to all resemble one another.

Last week was Microsoft’s Ignite event and, like last year, there was a major focus on agentic development. Last year we saw the release of Azure AI Foundry, which was one of the first platforms for rich agent development, alongside a host of more canned agents to supplement existing Microsoft products. This year presented a major evolution in Microsoft’s agent strategy. For starters, we now have a complete suite of capabilities for agents delivered either through enhancements to existing products such as FoundryIQ (the new name for AI Foundry) or via net-new products and WorkIQ. While last year’s announcements were competitively solid, this year’s really stood out for how the entire agent story and product strategy hung together. It fits in very well with Microsoft’s target audience, which are enterprises seeking a total vendor solution versus a customized best-of-breed play.

What stood out most to me from Ignite was Agent 365 thanks to how Microsoft extended its existing agent operational capabilities. The concepts from the built-in observability in AI Foundry have been extended significantly. The agent observability features have been improved and now include a personalized developer dashboard to simplify a user’s ability to manage and monitor their own agent fleet. There’s also significant improvement in the catalog features, including an updated model comparison tool. But the biggest standout capabilities were for security; these are largely new features centered on agent-specific threats, which are becoming increasingly difficult for standard cybersecurity tools to assess and remediate. Looking back on the notion of agent management and operations, I am pleased to see that Microsoft was very deliberate in sticking to this path because it was an area of differentiation over the last year.

Google announced a preview of its own agentic development tool this week, called Antigravity. And as with many of the other agent IDEs out there, Google gave its own spin on the new age of application development. In line with other tools like GitHub Agent Mode, developers can spawn and oversee multiple agents developed asynchronously. There is also a feature called Artifacts, which seems like spec-based development where agents will document the pre- and post-activities to help ensure a better-quality deliverable — very similar to what we have seen with AWS’s Kiro IDE. And while I tend to think that the continued evolution of AI in the developer space is possibly one of the most interesting AI use cases, I also would prefer to see more cooperation in the tooling space, as opposed to everybody having their own IDEs — which are starting to all resemble one another.

The Supercomputing 25 conference came and went, and as usual it didn’t disappoint. A few observations:

• St. Louis is not equipped to handle a large convention, nor is it nerd-friendly.
• SC has become more like the industry giant NetWorld Interop of old — an unbiased forum for every tech vendor. It’s a breath of fresh air.
• Not surprisingly, liquid cooling remained ever-present, and collectively this industry perhaps had more representation than any other. Only this time, it wasn’t just companies like Jet Cool, Vertiv, CoolIT, etc. I saw a huge presence from the back-end players — chemical companies making propylene glycol (PG 25), hose and coupling manufacturers, pump manufacturers, you name it. If you thought these were the small booths lining the outer perimeter of the show, you’d be right. However, the traffic was nonstop. At one of the PG 25 manufacturers (not a big name), I saw every major CSP walk through in a span of 20 minutes.
• Dell made a huge splash across the board — compute, storage, and networking. As did HPE. As did Lenovo. As did Cisco. There is too much to include in a short format like the MI&S weekly report, but suffice it to say that these IT solutions vendors have moved on from selling NVIDIA chips to selling OEM value-added solutions. Furthermore, there appears to be a genuine understanding that this value-add must be accompanied by a services-led approach focusing on outcomes.
• VAST Data is still the darling of the Supercomputing crowd. No matter how many arrows its competitors shot, the company owned the energy at SC25. (By the way: it’s time to retire the cowboy hats.) If I hear one more competitor say “But they’re really just a storage company,” I may go crazy. This company, valued at well over $10 billion at this point (and backed by NVIDIA) has clearly found a happy and paying customer base. VAST and its AI OS are not smoke and mirrors, nor is the company dealing in irrational hype. While its DASE architecture may be the star of its show, and while its enterprise transition may be difficult, it is well established where absolute performance matters.
• Keep an eye on Pure Storage. While this company has been so smartly pragmatic in its approach to the market, it is quietly establishing its own hyperscale and AI footprint — one that could be wholly disruptive.
• Did I mention that St. Louis is not a good conference destination?

Microsoft held its Ignite event last week. The company made several announcements regarding Microsoft Foundry and agentic AI, which I found to be quite compelling. However, its announcement of the Cobalt 200 CPU for the Azure infrastructure caught my attention, and the amount of press inquiries I received validated my thinking.

At this point, most people familiar with this market know about Arm’s move into the hyperscale datacenter. While AWS is probably the most recognizable success story with Graviton, each CSP (and large-scale datacenter) is seeing incredible results. In fact, some estimates suggest that Graviton will populate more than half of AWS’s new deployments.

But Azure is different to me — and perhaps more impactful. Because while AWS and Graviton are impacting Intel and AMD from a revenue and share perspective, Azure is telling the traditional enterprise that this Arm architecture is not only permissible, but preferred in some cases. I’ve been a bit more bullish on Arm than most, and am no doubt continuing this behavior. But I think Azure’s leaning heavily into Cobalt is going to open doors for Arm in the enterprise. Not Arm in the enterprise via cloud services — that is already happening. Rather, Arm physically in the enterprise datacenter, starting with large enterprises and, over time, reaching commercial enterprises.

This horizon is not short-term, but rather medium-term. However, this is the future of enterprise IT, and the x86 players would be well-served to track Cobalt adoption closely.

Based on what I saw at Microsoft Ignite 2025, I think Fabric is getting closer to being a unified analytics platform. This should jump out for companies dealing with a mix of data types and systems. Arun Ulag, the head of Azure Data, highlighted some updates that matter. With Fabric’s new release, teams can use the SQL Database and Cosmos DB integration to bring data directly into Fabric. This makes manual data work simpler and gives easier access to relational, JSON, and document data all in one place with OneLake.

Microsoft and SAP are also working together more closely. Fabric is now able to connect to SAP live data across Business Data Cloud (BDC), Datasphere, and S/4HANA through zero-copy, bidirectional data sharing. This eliminates the need to copy or duplicate data, keeping everything secure, consistent, and automatically up to date for analytics, reporting, and daily operations across both platforms. Teams can access SAP data directly in Fabric for better reporting and day-to-day use.

Microsoft also added more compliance controls to Fabric, improving audit trails, backup options, and ways to manage your own encryption keys. These changes help with security, but you still need smart setup and monitoring so you can avoid any access problems.

Even with these changes, the basics remain important. Real-time data handling, tuning for performance, and keeping things secure still take planning and discipline. The Ignite updates move Fabric closer to an all-in-one analytics platform and can make things more manageable for teams. Understanding the complexities will also make it easier to realize the benefits.

Salesforce has officially completed its acquisition of Informatica, pulling core data-management capabilities — integration, governance, data quality, metadata, and MDM — directly into its platform. This gives Salesforce a stronger foundation for its apps + data + AI agents vision and could make it easier for customers to work with cleaner, more consistent data without juggling as many tools. It could also round out the MuleSoft story and position Salesforce deeper in the enterprise data stack. The company expects the deal to improve margins and EPS within a year, which shows confidence in how quickly Salesforce believes it can put the pieces together.

Of course there are real challenges for any acquisition of this magnitude. Merging two large platforms and ecosystems isn’t simple, and Salesforce has acknowledged that the benefits may take time or may not land exactly as planned. Informatica has long been its own ecosystem with its own roadmap and partner model, and customers will want clarity on which things will change, what will be consolidated, and what will stay intact. There’s also the typical integration friction with product overlap, cultural alignment, and the risk of confusing customers during the transition. So even if the combined platform does simplify things long-term, many organizations will face short-term effort around migrations, contract changes, or reworking data pipelines.

I do think this move pushes Salesforce beyond CRM and workflow automation into more of the trusted data layer that AI systems depend on. If Salesforce integrates Informatica effectively, customers could see more reliable data flows and a more straightforward path to using AI agents in real-world operations. But it’s all about execution, and companies will need strong data quality, governance, and security practices.

It will be interesting to see how Salesforce evolves the Informatica portfolio without adding complexity, as Informatica will continue to support its existing partner ecosystem while Salesforce works to rapidly integrate the technology stack into Data 360 and the broader Salesforce platform. The test will be whether this combination actually helps teams work better with their data or just moves the challenges to a new place.

AMD and Cisco recently announced a new partnership with Humain to support its ambitious AI datacenter deployments in the Middle East. This builds upon previous collaborations for smart switching and compute infrastructure, creating a meaningful alternative to NVIDIA platforms.

I suspect that Cisco will lean into its custom silicon capabilities with Silicon One to provide further feature and functionality differentiation relative to HPE, Dell, and other competitors.

Manufacturers using legacy ERP systems feel the strain because everyday workarounds end up costing more than people realize. Manual scheduling, scattered data, late quality catches, slow reporting, and reactive maintenance all seem manageable until they start showing up as higher freight bills, missed production windows, extra scrap, compliance issues, slow onboarding, and growing technical debt. These pressures add up quietly and become part of daily operations. 

By contrast, modern ERP systems take a more practical approach by automating routine work, improving data quality, offering earlier signals of issues, and giving teams clearer insight so problems can be addressed before they turn into delays or unplanned costs. The chart in my latest Forbes article breaks down the areas of hidden cost where modern platforms can make a difference.

ERP vendors are moving toward smart, event-driven platforms with AI-enabled agents at their core. For example, SAP is focusing on connected data and event-driven automation. Oracle prioritizes autonomous decisions based on live operational context. Microsoft is using MCP-based agents to connect work across operations, finance, and supply chain. Infor offers industry-specific ERP CloudSuites designed to address the unique manufacturing and compliance needs of each vertical. Epicor focuses on modular upgrades that let mid-sized manufacturers adapt as business needs change. IFS brings ERP, asset management, and service into one event-driven platform. QAD|Redzone gives frontline teams real-time tools to address plant-floor issues as they happen.

The longer companies wait to modernize their ERP systems, the more these costs accumulate. Modernization does not need to be disruptive. Small, targeted steps and focused pilots can show value quickly. The manufacturers that treat modernization as an ongoing part of running the business, rather than a one-time project, will be the ones who succeed and stay competitive. Read more about “The Hidden Costs Of Legacy ERP For Manufacturers — And What To Do About Them.”

Last week I was in San Francisco for Microsoft’s Ignite conference. Microsoft gave many updates; the focus was on pushing its business apps deeper into the agentic era, with new capabilities across Dynamics 365 ERP, supply chain, sales, and Power Platform. Some of the bigger announcements were about new Model Context Protocol (MCP) servers for ERP. The Dynamic ERP Integration Server (public preview) replaces the old static model and gives AI agents secure, real-time access to hundreds of thousands of ERP functions. A second piece, the Analytics Access Server (preview in December), brings the same context model to metrics and semantic insights so agents can query business intelligence directly inside day-to-day applications. On the supply chain side, these MCP servers power new automation scenarios — monitoring trade updates, flagging risks, and even handling tariff-related compliance checks without manual effort.

Microsoft also introduced a Sales Development Agent through its Frontier Program. It’s built to run prospect research, lead qualification, and outreach around the clock, ideally freeing up human sellers for actual conversations and closing. The Power Platform is getting pulled into this agent ecosystem as well. A new agent-powered maker workspace in Power Apps lets users build multi-page apps using natural language prompts to Copilot. The new Power Apps MCP Server (public preview) makes those apps discoverable to agents so they can take actions like triggering approvals or submitting forms. And with the Dataverse MCP Server now generally available, agents can provide real-time answers grounded in Dataverse data using simple natural-language requests.

I like how Microsoft (like some other ERP vendors) is framing a shift from using ERP as a system of record to a system of action. Making this shift effective will require an effort to clean up data, having a strong governance program in place, and investing in organizational changes that have real value. This is critical because AI agents aren’t being added to the edges of the stack anymore; they’re being built into the core of ERP, CRM, workflows, and app development.

Microsoft has rolled out its enhanced Xbox full-screen experience to all Windows 11-based PC gaming handhelds, expanding beyond the previous rollout on the ROG Xbox Ally X handheld. I believe this demonstrates Microsoft’s commitment to the platform and helps ensure that the ecosystem gives customers a choice of how they want to experience handheld PC gaming. It remains to be seen whether the performance of this new interface will be the same on all systems since Microsoft launched on the most performant handheld available to date, but I could easily see Lenovo and others offering this as an option fairly soon. This also ought to improve Microsoft’s position against SteamOS.

Matter 1.5 marks an inflection point for the smart home. It adds the most-requested device type — cameras — with robust features like WebRTC streaming, multi-stream support, PTZ control, privacy zones, and flexible recording, making cameras first-class citizens across all major Matter ecosystems (Apple, Google, Alexa, SmartThings).

This update also introduces advanced energy management, adding support for EV charging and local generation/storage equipment, while standardizing access to real-time pricing, tariffs, and carbon usage. “Advanced” management means enabling cost-aware, carbon-aware automation that shifts energy usage intelligently and saves consumers money. Yes, that’s right, a smart home system with a positive ROI.

Most importantly, Matter remains true to its multi-ecosystem brand promise. After three years of steady improvement, devices from any brand can now work with any ecosystem. There are still some rough spots, but Matter 1.5 has come a long way. Interoperability is the foundation for the next era of residential automation — agentic, event-driven, multi-vendor systems that coordinate and optimize the entire home.

In short, Matter 1.5 delivers critical device coverage, unlocks intelligent energy optimization, and reinforces the multi-vendor smart home brand promise. For product makers and consumers, it’s the moment where Matter moves from “promising” to “strategic.”

T-Mobile has launched its first “uncarrier” event in Las Vegas under new CEO Srini Gopalan, with its F1 partnership as the backdrop. The company announced its next major effort to solve consumers’ pain points with a new app-based way to switch to T-Mobile with its T-Life app. This ability to switch, paired with an expanded DoorDash partnership to offer same-day delivery of new phones, really changes the entire process. T-Mobile is also allowing customers more time to switch their phone and take advantage of new deals by giving them up to 90 days to make a decision. T-Mobile also announced a deepening of its partnership with F1, not only by expanding its 5G services in Las Vegas, but also by expanding the partnership to COTA in Austin for the (2026) U.S. Grand Prix.

This will be a welcome improvement, because COTA’s biggest weakness as a venue is its infrastructure, and it has always suffered from a lack of connectivity, even with some of the private 5G upgrades it has made recently. T-Mobile’s partnership with F1 goes beyond just enabling fan connectivity as it also powers commerce and broadcast. This will likely be necessary for F1 given its new broadcast partnership with Apple, which I assume will once again raise the bar on F1’s television production.

At Microsoft Ignite in San Francisco last week, the company’s Frontier Firm AI Initiative was one of several major initiatives announced. Launched in partnership with Harvard Business School’s Digital Data Design Institute (D3), the initiative convened an inaugural cohort of fourteen global organizations, including Barclays, BNY, Eli Lilly, Nestlé, and Mastercard, to collaboratively research and define what it means to become a human-led, agent-operated enterprise. I attended a luncheon with these first cohort members and saw firsthand how major enterprises are navigating beyond AI pilots toward operational transformation that embeds intelligence across every layer of work. What became clear through the discussions is that these organizations are unified not by having already arrived at some AI utopia, but by their willingness to experiment openly, share learnings, and confront the hard questions around governance, data quality, business-IT alignment, and scaling beyond proofs of concept.​

Leaders from companies including GHD, which already reports more than 90% active Copilot usage among 5,000 seats, and BNY, which aspires to run an AI-first bank, described journeys marked by pragmatic iteration rather than wholesale revolution. BNY has put digital labor on the org chart with 150 digital employees working across the bank with unique employee IDs, while GHD’s teams pilot agents for coaching, information retrieval, and technical drawing reviews. Barclays reduced a legacy database migration from three months to 20 days by using squads of agents working alongside human experts in tech operations, and deployed AI assistants in contact centers that summarize customer calls and guide human agents in real time. Mastercard has developed an Agent Pay feature that facilitates payments within conversations, enabling agents to conduct transactions on behalf of users while maintaining human control over decisions.

The cohort’s participation in field-based experiments with Harvard researchers signals a recognition that the path to becoming a “Frontier Firm” requires evidence-based blueprints, executive upskilling through custom workshops, and a culture shift that treats AI not as a tool for incremental productivity gains but as a fundamental reimagining of how work gets done. For these 14 pioneers, the initiative represents both accountability and community, as well as a commitment to lead publicly, learn together, and translate ambition into measurable outcomes that reshape customer engagement, employee experience, and business processes from the ground up. The progress these organizations are making suggests the trajectory is moving beyond pilot programs toward structural changes in how enterprises deploy intelligence across their operations.

The adoption of the term “frontier” for AI initiatives across the industry mirrors how Microsoft shaped the AI conversation through its Copilot movement, establishing a vocabulary that vendors and enterprises now widely use when discussing agentic AI and enterprise transformation. I look forward to tracking the learnings of these cohort companies as they progress through the initiative and observing how their experiences influence broader AI adoption patterns across industries. Their work will likely set benchmarks for what it actually takes to move beyond experimentation into enterprise-scale deployment.

Dell launched two new products last week. First, the Area 51 desktop finally got AMD’s new Ryzen X3D chips. Second, Dell launched a unique AI-focused laptop that has a Qualcomm AI100 accelerator paired with a ton of VRAM. Both products have been talked about for a while, and both of them clearly demonstrate that Dell is moving quickly to address industry trends and offer differentiated solutions.

Quantum Advantage
IBM anticipates that it will officially achieve quantum advantage — where a quantum computer outperforms any classical computer on a given task — sometime next year, and that possibility has generated much discussion. IBM’s next quantum processor slated for release is the Nighthawk, which has been specifically designed to attain and manage quantum advantage.

Quantum advantage is most likely to emerge in handling certain problems where quantum devices are able to compute correctly at scales that classical computers can’t match. There are three types of technical problems that stand out as the most promising near‑term candidates for attaining quantum advantage.

• Carefully structured sampling tasks that admit nontrivial verification, such as peaked random circuits or other constructions with detectable signatures.
• Variational and sample‑based ground‑state algorithms in quantum chemistry and materials science, where classical methods already encounter difficulty with moderate system sizes — but quantum‑generated subspaces can be classically analyzed at high precision.
• Expectation‑value and time‑evolution computations for many‑body systems, particularly in areas where classical algorithms struggle.

The important theme is not just speeding up the handling of these three types of problems. It is also important to be able to verify quantum advantage for tasks where quantum devices can be trusted to deliver answers that classical computation can’t match.

Quantum Healthcare Applications
IBM plans to turn its quantum research into practical tools for life sciences and sustainability, and to help industry partners with problems such as molecular simulation for drug discovery, genomic analysis, and carbon‑capture optimization. As discussed in the previous item, the company’s roadmap targets quantum advantage by 2026. And as I covered in my detailed overview of the IBM roadmap earlier this year, IBM expects to deliver fault‑tolerant quantum systems by 2029. These efforts are set to be enabled by two new processors, Nighthawk and Loon. These processors have approximately 10x faster error correction and vastly higher operation counts than previous quantum processors.​

In life sciences, IBM has made it a practice to collaborate with organizations such as Cleveland Clinic, which uses hybrid quantum‑classical workflows for simulating complex molecular interactions. IBM’s goals are to help accelerate drug and vaccine development and enable more accurate, personalized medicine.

IBM also helps with sustainability by combining quantum, AI, and cloud technologies to optimize energy grids, design better carbon‑capture systems, and support broader net‑zero strategies. Its quantum computers are the tools for climate and environmental modeling.​

For interdisciplinary, ecosystem‑driven development, IBM quantum scientists work with biologists, clinicians, and environmental scientists to co-develop designs. Because IBM works closely with channel partners and other stakeholders, the use cases can be targeted at real industry needs.

IBM’s work in error correction and FPGA‑based systems are intended to make quantum more accessible and reliable for production‑scale workloads. The company has differentiated itself by focusing on high‑impact applications in health, environment, and materials that will benefit from quantum advantage.​

Modern AI, including generative and agentic applications and workloads, is poised to transform enterprise operations through massive automation and force-multiplier improvements in knowledge-worker productivity. However, it also presents new cybersecurity challenges in areas including agent identity access management, runtime protection, and data loss prevention.

Palo Alto Networks recently released six predictions for 2026 that map to its platform depth in evaluating cyber risk and securing the use of AI at scale. It is a compelling read, providing valuable insights into what organizations should consider in fortifying security posture and moving security operations from a defensive to proactive stance.

Qualcomm used its first Snapdragon Tech Days to demonstrate the specifics of its performance claims for the Snapdragon X2 family of processors that it announced earlier this year in Hawaii. It detailed the CPU, GPU, NPU, and even platform decisions it made to be able to offer competitive performance and battery life, which I believe offer the best competition to Apple in the PC ecosystem. We likely won’t see products with this family of chips until some time in Q1 2026, but it will make for an interesting comparison to Intel’s Panther Lake, the rest of Apple’s M5 lineup, and whatever AMD has cooking for 2026.

In a surprising move, Nokia recently announced its intention to sell its enterprise campus edge business unit in 2026. ECE is responsible for the majority of the company’s sales in private 5G networking — a segment that the cellular infrastructure provider has dominated for the past several years.

The reality is that private cellular networking is a crowded and highly competitive market, with other offerings from Celona, Ericsson, NTT, and a host of operators including T-Mobile for Business’s Advanced Network Solutions. The shift in strategy may be based on Nokia’s route-to-market challenges into enterprises, something evidenced by its exit from providing compute infrastructure in early 2024 and partnering with Dell.