This week on Support Engineering Weekly, Alex and Maya explore a growing tension in Microsoft's cloud ecosystem: administrators are gaining new visibility into AI usage, performance, and spending while legacy APIs and authentication methods are being retired. We examine new Agent 365 activity insights, tenant-level adoption metrics, and beta Microsoft Graph APIs for Copilot cost management, including how programmatic spending policies could help organizations monitor consumption and establish financial controls as AI deployments expand. We then turn to the operational changes that demand preparation. Exchange Web Services (EWS) is entering retirement in Exchange Online, requiring affected archive and compliance integrations to transition to Microsoft Graph APIs. Microsoft Entra's move toward customer-managed telephony providers for SMS and voice authentication also raises important questions about authentication strategy, passkey adoption, user communications, and help-desk readiness. Alongside these changes, new remote repair capabilities for Microsoft 365 Apps and read-only diagnostic checks in Microsoft Purview eDiscovery offer administrators more ways to troubleshoot problems without relying entirely on manual intervention. Finally, we investigate several Microsoft 365 service-health incidents that reveal the risks of interconnected cloud systems and fragile telemetry pipelines. From Copilot model visibility changes and missing contacts to Global Secure Access disruptions and prolonged delays in Microsoft 365 usage reports, these cases show how backend dependencies, data-processing bottlenecks, and user-interface behavior can affect administration and support. We close with a question for every organization planning its 2027 IT and compliance strategy: is your internal monitoring infrastructure as resilient as the applications you are trying to support? This episode is based on the Microsoft Message Center and Service Health updates covered for October 19-23, 2026. Chapters 02:02 - AI Agent Visibility and FinOps 06:03 - Exchange Web Services Retirement 08:12 - Entra SMS and Passkey Transition 11:10 - Remote Repair and Compliance Diagnostics 14:41 - Copilot Visibility and Contact Failures 18:34 - Global Secure Access Outage 19:20 - Usage Reports and Telemetry Backlogs Learn more: Visit the Support Engineering Blog at https://blog.sadhan.ch/ for the full articles, technical references, and additional support-engineering resources behind the discussion.
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
You are listening to support engineering weekly. You know, upgrading your organization's infrastructure right now, it feels a bit like buying a state of the arts for its car. Oh, absolutely. Like, suddenly the dashboard has, you know, 50 new, highly precise metrics showing you exactly how the engine is performing, it's optimizing every single drop of fuel, giving you all these predictive insights. Right. It's a beautiful new display. Exactly. And simultaneously, while you're totally distracted by that shiny display, the manufacturer
is like in the backseat, actively removing the physical emergency break you've relied on for a decade. The operational whiplash is just, it's very real. You're gaining this unprecedented kind of futuristic visibility into your environment, especially around AI. But your legacy safety nets, you know, the protocols and tools you've trusted for years, they're just quietly being dismantled in the background. It creates this really precarious balancing act for IT teams. Yeah, it really does. Well, I'm Alex and I'm Maya and welcome to this deep dive into the Microsoft weekly updates
for October 19th through the 23rd, 2026. We really want to thank you for joining us today. Thanks for being here. So if you manage a Microsoft environment or, you know, if you're architecting the future of one, you aren't probably feeling that push and pull we just described. Our mission today is to unpack the week's message center and service health updates. Right. We're going to translate all these massive architectural shifts from Microsoft into, well, practical operational impacts for your environment. Because I mean, reading a bulletin about an API deprecation is one thing.
It's totally different from understanding the underlying mechanism of that change. Like realizing how it's going to fundamentally break your third party compliance solution next April. That's another story entirely. Exactly. We have to look at the architecture behind the announcements. We really do. So let's start with the positive side of that sports car analogy, you know, the shiny new dashboard. The fun stuff. Right. Microsoft is finally bringing AI out of the shadows and giving administrators the tools to see what's actually happening. We've got a cluster of updates here starting with message center 1506727.
Right. The agent 365 update. Yeah. Microsoft is expanding agent 365 usage and activity insights right in the Microsoft 365 admin center. And the critical detail in this rollout, and this is huge, is that Microsoft is providing this telemetry without requiring a dedicated agent 365 license. Oh, wow. Really? Yeah. So if your organization is using Microsoft built agents or even, you know, agents built on Microsoft's platform by your internal devs, you're now getting foundational visibility natively. That's a big deal.
It is. We're talking about metrics that track active users, session durations, total runtime, and crucially exception rates. Which is just a massive shift. I mean, before this, managing AI costs felt like getting a single enormous credit card charge at the end of the month with zero context. Just a giant line item that says AI compute. Exactly. Just a terrifying dollar amount. Now this update is like getting the fully itemized receipt for a massive corporate dinner party.
You finally see exactly which specific agents are like eating all the expensive appetizers. That's a great way to put it. And you know, taking that dinner party idea a step further, they're pairing this with update 1520901, the insights agent V2 frontier. Okay. What does that add? So this provides tenant level visibility into adoption metrics. But what really stands out structurally is how they're handling user feedback. It's strictly opt in for sharing user ratings and comments with the agent owners. Oh, interesting.
So the devs don't automatically see what users are saying. Exactly. So administrators maintain complete control over the feedback loop. It prevents developers from seeing raw, potentially sensitive user comments about their specific internal agents unless the organization explicitly allows it. I mean, that makes total sense from a data privacy standpoint, like, especially if an internal HR agent is getting feedback that includes, I don't know, personally identifiable information. Right. You don't want that bleeding over to the dev team.
Definitely not. The real heavy hitter for visibility here, like the tool that actually lets you manage the check of that dinner party is 1521616. Microsoft is introducing beta graph API's specifically for co-pilot cost management. Yes. I look at this and I see like the true birth of AI FinOps. It is the absolute foundation. It is. I mean, safely scaling AI in an enterprise is basically impossible without programmatic controls. Because things can spiral so fast.
Exactly. Until now, organizations have really struggled to limit compute spend dynamically. But with these new REST APIs, you can retrieve tenant, user, and per service credit balances. Nice. And more importantly, you can author, update, and delete spending policies and then assign those policies directly to entry security groups. Wait, wait. If I can assign a spending policy to a security group, that means I could hypothetically build an automation where if an experimental internal agent gets caught in a logic loop and starts
burning through thousands of compute credits, it hits a threshold. Right. It hits a threshold. And my FinOps tool immediately pulls that security group's access to the agent. Exactly. And I don't even have to wake up an admin in the middle of the night to do it. That is the exact architectural goal they are aiming for. You tie compute exceptions and run time directly back to hard financial limits. You set up programmatic threshold alerts. That's so powerful. It really is. Because without these APIs and those admin center metrics feeding your monitoring tools, deploying AI at scale is essentially flying blind.
You're just like hoping developers write efficient prompts, which is not a viable financial strategy. No, definitely not. All right. So we've got our shiny new telemetry. We can see where the money is going and we can build automated breaks. Let's transition to the other side of the equation. The missing emergency break. Right. First off, this handing you the keys to AI cost control. They are aggressively taking away the legacy controls and protocols you might still be relying on. Let's look at this wave of forced modernization. Yeah.
This area is where things get operationally dangerous. If your architecture team isn't paying close attention, the most urgent shift is 1469565. The EWS retirement. Exactly. The final retirement of exchange web services or EWS in exchange online. The global disablement is beginning right now in October, 2026 and it will be completely finalized by April, 2027. So for anyone who might not be deep in the exchange trenches, EWS is the old soap based API before Microsoft graph came along and standardized everything on rest.
EWS was the backbone. It was everything. Yeah. It was how basically every third party tool connected to exchange to read or move mail. And the structural shift away from it is monumental. Microsoft is forcing all archive mailbox scenarios. So we're talking compliance, e-discovery, data lifecycle management, all of it over to Microsoft graph APIs. Specifically, the mailbox import, export and mailbox folder APIs, right? Yes, exactly. But I mean, if a vendor has to migrate from a soap protocol to a completely different rest
architecture and graph, that's not just a patch. No, not at all. That is a ground up rebuild of how their application authenticates and handles data payloads. So if I'm an admin, relying on a third party tool for my daily e-discovery holds or my legal backups, and that vendor hasn't finished this rewrite, come April, does it just degrade or does it completely break? The tool literally stops functioning, full stop. Wow. Yeah, the authentication requests will just be rejected by exchange online and it will not pull a single byte of data.
If your organization relies on any third party vendor for compliance exports, journaling or archive backups, you need to demand a hard confirmation of their graph migration status today. Because there's no grace period. None. Microsoft is not extending this deadline. Yeah, that is a terrifying cliff for legacy applications. And speaking of hard deadlines and cliffs, let's look at 1519855. The Entra update? Yeah. Entra is moving to a choose your own telephony provider model. They call it CYOT for SMS and voice authentication.
Microsoft is completely retiring its own provided SMS and voice off by February 1st, 2027. This is such a fascinating strategic move from an engineering perspective. Microsoft is entirely washing its hands of the telecom carrier relationship. Well, they used to just handle all that. Historically, yes, they absorb the complexity of global SMS delivery. Now if your organization absolutely insists on retaining SMS as a multi factor authentication method, you have to go to the Azure security store, select a supported third party telecom
provider, bring your own API keys and manage the commercial relationship and billing yourself. So instead of Microsoft handling the like Twilio or bandwidth contracts on the back end, I have to go set up an enterprise telecom contract, get an API key, paste it into my enter portal and pay a separate monthly bill just so my users can get a six digit text message. That is the new architecture. That is. And honestly, it's designed to create friction. It definitely sounds like it. But I have to push back on this timeline a bit.
Like if a company just ignores this enter update because they have a million other priorities, their end users whose only MFA method is SMS are going to hit a brick wall in February. They absolutely will. The documentation says they'll be forced to register a Passkey during sign in. And it explicitly says the prompt is blocking. Are organizations genuinely ready for a blocking Passkey registration prompt to hit thousands of users on a random Monday morning? Operationally, most organizations are absolutely not ready for a mandatory Passkey rollout.
No way. And that is precisely why Microsoft designed the architecture this way. They want to force a global transition to phishing resistant Passkeys by default. Managing global SMS delivery is just an engineering nightmare filled with carrier filtering, exorbitant international routing costs. It's a mess. It is. So by making you pay a third party directly for the SMS routing, it suddenly becomes a highly visible line item on your IT budget. So it forces the financial conversation.
Exactly. It forces chief information security officers to ask why they are paying a telecom vendor thousands of dollars a month for a highly vulnerable outdated authentication method. And device bound Passkeys are essentially free and infinitely more secure. It's forced modernization through financial friction. I respect the strategy, but man, if I'm an admin listening right now, my head is spinning. You know, I have to manage new telecom contracts, audit my compliance vendors for graph readiness, figure out how to monitor beta FinOps APIs.
I really pity the help desk on February 1st if the communications haven't gone out to the users. Well, the balance out that immense pressure. Microsoft is actually throwing IT teams a few administrative life preservers this week to help manage these shifts. And goodness. Yeah, they're rolling out some new centralized tools to handle remote diagnostics. I'm really glad you brought that up because this next one, 15 18, oh 32, is going to make desktop support engineers literally weep with joy. The remote repair tool. Yes. The update repair in inventory feature for Microsoft 365 apps.
This fundamentally changes how help desks operate. It really is a massive shift in remote device management through the M365 apps admin center. Administrators can now remotely trigger an online repair for Windows devices that are reporting out of date or corrupted office installations. Let's walk through like a realistic scenario of how this changes the daily grind. Right now user calls in because their Excel client keeps crashing on launch. The classic ticket. Exactly. The traditional workflow requires a technician to initiate a remote screen share session,
interrupt the user's work, navigate through the Windows settings to add remove programs, and run the office repair tool manually. Or worse, if the remote tools fail, it requires a physical desk site visit, which eats up so much time. So much time. But with this update, you can identify all the corrupted clients in a single dashboard and trigger an automated backend script that silently reinstalls the latest healthy build based on their assigned update channel. The architectural caveat here though is the word silently. Oh, right.
Yeah. The remote command forcefully closes any open office applications to execute the repair and swap out the binaries. So you could lose work if you're not careful. Exactly. Yeah. So operationally, while you don't need a screen share session anymore, you still need an established communication workflow to warn the user to save their work before you push that button. Make sure they hit save. Yeah. But moving from manual one off interventions to remote automated remediation at scale, that reclaims thousands of hours for an enterprise help desk. Oh, without a doubt.
And we're seeing the same philosophy of safe remote diagnostics elsewhere in the message center this week too. Like the fabric database hub is becoming enabled by default in November, which gives data teams a central dashboard to discover database endpoints without risking unauthorized alterations to the underlying resources. Right. And update 151-5937 brings three brand new read only diagnostic checks to purview e-discovery. Those purview diagnostics are structurally critical for compliance teams.
They cover case access, security filters, and hold policy distribution, which is huge for troubleshooting. It is because when you're troubleshooting a complex compliance hold that isn't capturing the right custodian data, the absolute last thing you want to do is accidentally alter the configuration while you're testing it. Because altering a compliance configuration mid investigation could break the chain of custody and create a massive legal liability for the company, right? You'd essentially tamper with the search parameters of an active legal hold. Precisely. Having read only diagnostic tools built natively into the product allows your tier three engineers
to safely figure out why a search scope is failing or why a security filter is blocking access. And they can do all that without ever executing a right command that would alter the legal parameters of the hold. That's brilliant. So, we've got incredibly powerful new dashboards. We have aggressive forced modernization, and we have these great remote diagnostic tools. But even with all of this centralized control, you know, complex systems still break in production. Always. Let's look at the service health logs for this week and do a bit of an autopsy on what happens
when these interconnected systems fail. A prime example of complexity creating confusion is incident 1514746. Others suddenly lost the ability to see the GPT models within their co-pilot chat model selection drop down. Right. I saw this one. The service was flagged as degraded, helped us take it spite globally. But then Microsoft investigated and updated the status to false positive. Yeah. Which is a really rare and fascinating resolution in the service health dashboard. What actually broke here? Well, broke was actually a fix.
Wait, what? Yeah. It was a UI bug that had been hiding the truth from administrators. So the affected tenants had admins who had previously gone into their centralized settings and explicitly created policies to disable those specific GPT models for their users. Okay. However, a bug in the front end code meant the user interface was completely ignoring that backend setting. It was displaying the models in the drop down menu regardless of the policy. So the UI was essentially lying to the users. Exactly. It's like fixing a completely broken stuck fuel gauge on a car.
The driver gets it back from the mechanic and immediately panics because the tank reads empty, but it was always empty. The gauge is just finally telling the truth. That is the exact mechanism of what happened. Microsoft pushed a backend update that finally fixed the UI bug. The front end interface suddenly respected the admins original deny policy. And the models vanished. The models disappeared for the end users exactly as the admins had originally configured them to do months ago. But because the users were so accustomed to the broken UI, the sudden disappearance
caused a massive panic. They all believed the underlying service was offline. That is just the ultimate task failed successfully scenario for an IT department. The policy worked. It just took a UI patch to enforce it. Exactly. But unfortunately, not all the outages this week were false positives. Let's unpack incident 1489432. The new outlier was the new outlier. And finding all their contacts, speed dials and organizational charts, just completely
blank. Gray boxes everywhere. This was a genuine architectural failure within the unified contact fabric. To understand this, you really have to look at how Microsoft handles contacts now. They don't use static vcards anymore. The root cause was a failure within a contact processing component that handles what Microsoft calls profile based contacts or PBCs. Okay. What exactly is a profile based contact? A PBC is essentially a dynamic pointer. Instead of storing a user's phone number and job title locally on your device, your
outlook or team's client holds a PBC pointer. And that pointer constantly queries the unified backend to pull the most recent active profile data, including their real time present status. Okay. So if it's a dynamic pointer, why did the whole contact card disappear? Like if a presence field is missing or a phone number isn't populated, shouldn't it just default to a gray offline circle or leave the field blank? Why would it nuke the entire visual component? Because the processing logic didn't handle null values gracefully. Oh, no.
Yeah. The component encountered contact pointers were specific upstream telemetry, like a subtle change in how a custom presence notification was formatted. Just wasn't explicitly stored or returned in the expected JSON format. So instead of defaulting to an offline state, the processing logic just panicked. It severed the underlying link between the contact object and the user profile rendering it. And because Microsoft's backend is a unified fabric, a failure in a minor team's presence formatting setting just cascaded instantly. It wiped out the visual rendering of contact lists in Outlook desktop, Outlook mobile and
team simultaneously. The interconnected nature of the platform means a single point of failure has a massive blast radius. We saw another severe cascading infrastructure issue with incident 1505-952. The GSA outage. Yes. Underperforming infrastructure within global secure access, which is Microsoft's edge security service, completely broke access to Outlook, OneDrive and Teams for users specifically routed through the UK and Germany data centers. And it required targeted traffic redirection to secondary notes to fix.
It really highlights that when your secure edge network underperforms, it doesn't just result in slow load times or latency. It functionally drops the packets and takes your core productivity suite completely offline for entire geographic regions. Which leads us into a critical underlying theme for the week. We've talked about dashboards telling the truth and edge infrastructure breaking under load. But what happens when the telemetry pipelines, the actual data streams that feed your admin center and your FinOps tools, what happens when they completely clog up?
Yeah, this was the multi month saga of the Microsoft 365 usage reports and the co-pilot health data pipelines. Looking at incidents 1481173 and 1501829, this wasn't just a brief data hiccup. Administrators were looking at their dashboards in late October and seeing stale usage data frozen from July. If you want to understand the sheer scale and fragility of Microsoft's backend data ingestion, you have to study this outage. It wasn't just a single failure.
It started with an upstream telemetry process responsible for aggregating raw user data being temporarily disabled. That initial pause caused a backlog of missing data files. When engineers tried to restore the pipeline and process the backlog, they collided with a planned infrastructure migration. Perfect timing. Right. The interaction between the old data files and the new infrastructure caused widespread quote, corrupt file formatting. And reading through the historical logs on this one, it was an endless game of engineering
whack-a-mole. It really was. They would apply a patch to fix the corrupt formatting and get the email activity reports flowing again, only to discover that the compute resources were so overwhelmed that the Microsoft 365 apps reports and the forms reports were now experiencing severe ingestion delays. The operational consequences for our listeners here are massive. And think about it, if your IT department relies on these usage reports to justify upcoming enterprise agreement license renewals or to track the adoption metrics of incredibly
expensive co-pilot seats, you were completely blind for months. You had no idea if your investment was actually being utilized. Exactly. But I have an architectural question about this. Once they found the code bug and fixed the corrupt formatting, why does it still take weeks to fix a data ingestion issue? Like why can't they just run a script and catch up overnight? It comes down to hard physical limitations on data request volumes. You cannot just flip a switch and instantly ingest three months of missing global telemetry
across tens of thousands of enterprise tenants. The pipeline has fixed compute, memory, and throughput limits. If you dump a three-month backlog of raw telemetry into the ingestion engine all at once, you will crash the active production systems. It's like trying to drain a flooded football stadium through a standard garden hose. You have a path, but the physics of the hose dictate how fast you can move the volume. That's a perfect analogy. The telemetry engineers had to slowly methodically process the backlog during off-peak hours to
ensure they didn't degrade active service performance. It requires throttling the data flow, which naturally stretches the recovery out over weeks. It really puts the immense complexity of these cloud systems into perspective. We started this deep dive talking about gaining new unprecedented visibility into AI costs and agent performance through those beta graph APIs. We talked about how you need this data to safely deploy the technology. This multi-month telemetry outage is a stark sobering reminder that your beautiful new
dashboards are only as reliable as the invisible, highly fragile data pipelines feeding them. As we saw with the aggressive retirements of Exchange Web Services and the SMS telecom shifting, you are being forced to rely entirely on these modern APIs and telemetry systems, whether you feel operationally ready for them or not. The safety nets are being removed. Which brings us to our final thought for you to consider this week. We are in an era where organizations are rapidly deploying incredibly advanced, autonomous
AI agents, while simultaneously being forced to retire old, reliable administrative safety nets. Yet, as the months long usage report outage proves, the foundational telemetry and data pipelines that track these cloud environments are currently struggling to hold up under the sheer weight of enterprise scale. As you build your IT architecture and compliance strategies for 2027, ask yourself, is your internal monitoring infrastructure actually as resilient as the cutting edge applications you are trying to support?
It is the defining engineering question for the next year. You've been listening to Support Engineering Weekly from the Support Engineering blog. Until next episode, keep troubleshooting, keep learning and keep engineering. Thank you for listening to Support Engineering Weekly. This podcast is available on Apple Podcast, Spotify, iHeartRadio and other major podcast platforms. Follow the show on your preferred platform and never miss an episode.
What happens when a routine network maintenance operation triggers a failure deep inside a cloud provider's authentication infrastructure? In this week's episode, Alex and Maya investigate how a seemingly isolated...
How can a routine network capacity upgrade trigger a global cloud outage - and why can the very systems designed to protect availability make the failure worse? In this Support Engineering Weekly bonus episode, Alex and...
Microsoft 365 administrators deal with information from several places: the Microsoft 365 Roadmap, Message Center, and Service Health. But knowing where to find an update is only part of the challenge. The real question...
A server under pressure fails its health checks. Automated traffic management removes it from service, pushing even more load onto the remaining infrastructure. What was designed to protect the system becomes part of...
Microsoft Project Online is approaching retirement - or, in this case, its retirement date has already passed. But replacing a project management platform is not as simple as moving tasks from one product to another. In...