A former Microsoft engineer just handed the entire cloud industry a mirror it didn’t want to look into.<\/p>\n\n\n\n
The reflection isn’t flattering, and for accounting and tax firms running deadline-critical software on a shared public cloud, it should be a wake-up call.<\/p>\n\n\n\n
Here is the core of it: cloud reliability is not a platform feature. It is a staffing and architecture decision. When the engineers who built a platform leave and the fixes they wrote become too fragile to touch, what remains is infrastructure that runs not on engineering excellence, but on accumulated fear of breaking something worse. For a CPA firm trying to file returns on April 14, that is not a risk worth taking.<\/p>\n\n\n\n
The good news is that a dedicated private server built for accounting workloads operates from an entirely different playbook.<\/p>\n\n\n\n
Comparing those two playbooks: Azure’s public cloud and Verito’s dedicated private cloud, is exactly what this article does.<\/p>\n\n\n\n
This article explains why the distinction matters more than most firms realize, and why the Azure expos\u00e9 of April 2026 is the clearest articulation yet of a structural problem that has always existed in hyperscale public clouds.<\/p>\n\n\n\n
In early April 2026, Axel Rietschin<\/strong><\/a>, a former Microsoft engineer with nine years at the company including time on the Windows kernel and Azure Core Compute, published a multi-part expos\u00e9 on Microsoft Azure’s internal engineering culture.<\/p>\n\n\n\n The account gained immediate traction in the technical community, accumulating hundreds of upvotes and corroborating comments on Hacker News within 24 hours of publication.<\/p>\n\n\n\n Rietschin’s account is not the grievances of a disgruntled ex-employee. It is a structured, technical diagnosis of a platform that has crossed a specific and dangerous threshold: it can no longer be improved without risking collapse.<\/p>\n\n\n\n The specifics matter here. Vague concerns about “big cloud fragility”<\/em> are easy to dismiss. These are not vague.<\/p>\n\n\n\n Azure’s compute test coverage never cleared 40%, according to Rietschin’s account as reported by The Register<\/strong><\/a>. Engineers were filing approximately 200 just-in-time production access requests per day across a single three-month window, meaning hundreds of manual interventions were needed daily just to keep the platform operational.<\/p>\n\n\n\n A 122-person team was responsible for 173 different VM control agents, with no complete documentation explaining how those agents fit together or depended on one another.<\/p>\n\n\n\n By 2023, roughly half of the Azure compute team were junior engineers. Zero of 64 planned reengineering items were completed. And in perhaps the most alarming detail: bug fixes were being actively rejected because applying them carried too high a risk of cascading failures in neighboring systems.<\/p>\n\n\n\n One site reliability engineer working across AWS, Azure, and GCP added external confirmation to Rietschin’s account in the Hacker News thread. They reported measured production data showing 80 to 90% of the cloud incidents they encountered across all three providers originated from Azure, a figure corroborated by byteiota’s reporting<\/strong><\/a> on the broader engineering community’s response.<\/p>\n\n\n\n There is a specific kind of fragility that emerges when a codebase has accumulated enough technical debt that the team responsible for it can no longer safely refactor it.<\/p>\n\n\n\n Engineers start working around problems rather than through them. Fixes get deferred. Institutional knowledge concentrates on a small number of senior engineers who eventually leave, taking that knowledge with them.<\/p>\n\n\n\n What Rietschin describes is that Azure’s core compute infrastructure reached exactly this point. A 40% test coverage ceiling does not mean 60% of bugs go uncaught. It means that the team does not have a reliable map of what the code actually does.<\/p>\n\n\n\n When something breaks, recovery depends on whoever is available having enough contextual knowledge to trace the failure backward, manually, in a production environment. That is not a process. That is a prayer.<\/p>\n\n\n\n For any organization that needs infrastructure to just work on a predictable schedule, this should be clarifying.<\/p>\n\n\n\n Most enterprise workloads can absorb cloud instability with some grace. A slow API response on a Tuesday afternoon costs someone a few minutes. A delayed batch job gets requeued. The business moves on.<\/p>\n\n\n\n Tax and accounting firms do not have this flexibility. Their exposure to cloud unreliability is structurally asymmetric, and the Azure account makes that asymmetry sharper than ever.<\/p>\n\n\n\n A CPA firm running Lacerte or Drake Tax on shared cloud infrastructure does not have an evenly distributed risk profile.<\/p>\n\n\n\n The risk spikes with the calendar. Tax season compression is real and severe: March and April individual filings, Q4 payroll deadlines, quarterly estimated tax due dates, and extension windows all create narrow bands of maximum exposure.<\/p>\n\n\n\n A two-hour cloud outage on a random Wednesday in August is a minor inconvenience. A two-hour degradation on April 14 at 10 PM is a different category of event entirely.<\/p>\n\n\n\n It means returns that don’t get filed. Clients who reschedule. Staff working through the night to recover lost time. In some cases, it means penalties that the firm absorbs because the failure was the infrastructure’s, not the preparer’s.<\/p>\n\n\n\n Scott Edwards, president of Milner, Howard, Palmer & Edwards, described exactly this pattern before switching to Verito. His firm had been dealing with hosting downtime on a generic cloud provider, and the unpredictability of that downtime during deadline periods was what finally pushed the decision. “The only regret anybody has ever expressed to me,”<\/em> he said after making the switch, “is that they wish they had done it sooner.”<\/em><\/p>\n\n\n\n That sentiment echoes consistently across firms that have moved from generic cloud infrastructure to purpose-built accounting cloud hosting. The regret is never about the move itself. It is always about how long they waited to make it.<\/p>\n\n\n\n Every hyperscale cloud provider publishes an uptime SLA. Azure, AWS, and GCP all advertise nines-based availability guarantees and offer service credits when they fall short.<\/p>\n\n\n\n These credits are calculated in minutes of downtime translated to fractional monthly bill reductions. They are not trivial from an accounting standpoint. But they are entirely disconnected from the actual cost of downtime for a tax or accounting firm.<\/p>\n\n\n\n An SLA credit does not reimburse a missed filing deadline. It does not compensate for the client relationship that erodes when their return is late. It does not cover the overtime billed when staff have to reconstruct work from backups after a production incident. And it does not account for the reputational damage to a firm whose technology failed publicly during the one window in the year when technology must not fail.<\/p>\n\n\n\n The math on this gets uncomfortable quickly, and if you want to examine it in full, the comparison of cloud hosting vs local server<\/strong><\/a> for accountants breaks down the total cost picture in detail.<\/p>\n\n\n\n The short version: the gap between an SLA credit and the real cost of deadline-week downtime is not measured in dollars. It is measured in client relationships.<\/p>\n\n\n\n The Azure expos\u00e9 is a dramatic example, but the underlying problem it illustrates is not unique to Microsoft. It is a symptom of what hyperscale cloud architecture requires at its foundation. Rietschin’s account made a specific company’s specific failures visible. The structural argument runs deeper.<\/p>\n\n\n\nThe Numbers That Should Alarm Every IT Decision-Maker<\/strong><\/span><\/h3>\n\n\n\n
What Happens When a Platform Can’t Be Improved<\/strong><\/span><\/h3>\n\n\n\n
Why Tax and Accounting Firms Feel Cloud Instability the Most<\/strong><\/span><\/h2>\n\n\n\n
When Your Workload Compresses Into Four Weeks a Year<\/strong><\/span><\/h3>\n\n\n\n
Why Hyperscaler SLAs Don’t Cover the Real Cost<\/strong><\/span><\/h3>\n\n\n\n
The Structural Problem with Hyperscale Clouds<\/strong><\/span><\/h2>\n\n\n\n