Skip to content
Tech Newsletter

Your Microsoft 365 uptime promises just got harder to trust

ContentBuffer
ContentBuffer
November 17, 2026·~6 min read
ContentBuffer

ContentBuffer

Welcome, Tech Leaders.

A hidden bug in Microsoft's cloud gateway caused one of the biggest outages in the company's history, and it ran for almost 24 hours. Businesses and public offices that depend on Microsoft 365 and Exchange Online felt it immediately. The company had two independent gateways for resilience, and both still went sideways. If you run anything on Microsoft 365, this one is a wake-up call. The root cause was a change to the gateway that, under certain conditions, exhausted CPU and memory until servers stopped responding. Pre-release testing missed it because the test hardware was weaker than systems already in production. Your staging environment may be lying to you in the same way. The timeline: the outage ran from November 12 at 16:20 UTC to November 13 at 16:10 UTC. About 15 hours earlier, Teams traffic was moved from the alternate gateway onto the primary, which carries Exchange Online access, and capacity was running 14 percent below optimal. Restarts didn't help, and a targeted fix from Microsoft engineers was needed. Microsoft extended email delivery timeouts so delayed messages wouldn't bounce. The full final report is due November 20.

Coral just dropped as an open-source, self-hosted gateway that sits between your agents and the systems they work on. Every tool call gets checked against your workspace policy first. Only then does Coral add the credentials and forward the call to the provider. Agents never see the tokens. If you're wiring agents into GitHub, Linear, your CRM, or a warehouse, this is the layer you've been missing. Allow, Ask, or Block gets set per provider tool and per custom tool. Custom tools are code, so they can hold rules that flat allow lists can't express, like only rolling back the services named in a ticket. Teammates can share tools without sharing API tokens, which is the part your security team will actually care about. The numbers are the reason to look. Coral is used by thousands of developers at more than 450 companies, and it's ahead of other MCP gateways in blind benchmarks of 100 enterprise tasks. Custom tools cut tokens by 37%, cost by 31%, and time by 29%. Coral runs on your infrastructure with zero data egress, is written in Rust, and is licensed Apache 2.0. Let's dive deeper…

In today's ContentBuffer update:

  • Microsoft's 15-Hour Email Outage Came From a Bug Hiding Since July

  • AWS Emulator Lets You Test 200+ Services Locally, For Free

  • AWS Nitro Sandboxes give every pod its own kernel, free of charge

  • Uno Platform 7.0 adds Apple TV, WebGPU, and drops Mac Catalyst

  • 5 new AI tools & 5 new AI jobs

  • More tech news

Latest Development

Outages And Reliability

Microsoft Outage Traced to Hidden Gateway Defect, Lasted Nearly 24 Hours

Image source: cdn.arstechnica.net

Summary: Microsoft's cloud gateway bug knocked out Microsoft 365 for almost 24 hours, from November 12 to November 13. A change plus a traffic shift exhausted server resources, and the test hardware was too weak to catch it. If you rely on Exchange Online, check your fallback plans before the final report lands on November 20.

Key Points:

  • The outage ran from November 12 at 16:20 UTC to November 13 at 16:10 UTC, just under 24 hours, tracked under incident IDs EX1550543 and MO1550821.

  • Over 80 percent of the affected gateway servers froze, and restarts were ineffective because large request queues were still waiting when they came back.

  • About 15 hours before the failure, Teams traffic moved from the alternate gateway onto the primary, which handles Exchange Online access.

  • Capacity was running 14 percent below optimal when the traffic shift hit, exposing a latent bug that had been sitting in the gateway change.

  • Microsoft extended delivery timeouts on affected emails and added capacity and traffic-distribution guardrails, with its full final incident report due November 20.

Why it matters: If your team runs Exchange Online or Teams behind Microsoft 365, this outage hit your mail flow and chat directly, even with Microsoft's two-gateway design in place. Check whether your retry logic and queue handling can survive a full day of delayed deliveries, since Microsoft's extended timeouts only soften the damage. Ops teams should compare Microsoft's November 20 post-incident report against their own SLA credits and failover runbooks before the next maintenance window.

Ai Agent Gateways

Coral Gateway Puts Policy Checks Between AI Agents and Your Tools

Image source: opengraph.githubassets.com

Summary: Coral is an open-source, self-hosted MCP gateway that checks every agent tool call against policy before attaching credentials. Custom tools cut tokens 37%, cost 31%, and time 29% in Coral's own tests. Worth a look if your agents touch production systems.

Key Points:

  • Agents write short TypeScript programs that Coral runs in a fresh V8 isolate with no network access, no credentials, and no imports, keeping untrusted code boxed in.

  • Default policy lets read-only tools run immediately, while every other provider or custom tool waits for approval, so writes never fire silently.

  • Custom tools can aggregate warehouse, CRM, and billing data, returning only the final metrics to the agent instead of raw rows.

  • Coral's shared server runs under Docker Compose or Kubernetes, and the desktop app ships for macOS on Apple silicon, Windows x64, and Linux x86_64.

  • Every run is attributed to the person or agent that started it, and training data is user-owned, capturing the goal, tools used, and outcome as a structured trace.

Why it matters: If your support or SRE team is running agents against Linear, GitHub, or a billing system, Coral gives you per-tool Allow, Ask, or Block controls without rewriting each integration. The 37% token cut matters most for teams paying per-call on big-context workflows, since custom tools can let a smaller model perform like a larger one. The zero-egress, self-hosted design also means your security team can approve it without a data processing review.

Incident Postmortem

Microsoft Email Outage Lasts 15 Hours After Lock Contention Bug Spreads Fleetwide

Image source: image.theregister.com

Summary: Microsoft's email went dark for 15 hours after a July bug in network gateway servers spiraled across the fleet. Over 80 percent of servers went unresponsive, and rollback wasn't possible. If you ship to large fleets, test on hardware that matches production.

Key Points:

  • The defect was introduced in July in network gateway servers and passed load testing because the tests ran on weaker hardware with fewer cores.

  • The bug caused lock contention inside an internal Windows process, and that contention got worse on machines with more cores and newer hardware.

  • Within two hours of the surge in load and drop in capacity, the defect was encountered across Microsoft's fleet, and over 80 percent of servers became unresponsive.

  • Rollback was impossible because the defect was months old, so engineers built a patch for the patch and restored groups of servers gradually over 15 hours.

  • Connection negotiation was the main access failure, with competing processes blocking connections. Mitigation was applied at 0720 UTC, and the final update went out at 0415 UTC on November 14.

Why it matters: If you run Microsoft 365 integrations or any service that depends on Exchange Online, this outage hit your uptime whether you touched Microsoft's code or not. The real lesson is for SRE and platform teams: if your load tests run on older, lower-core hardware, they may never surface lock contention that only shows up on your newer fleet. Check whether your canary and staging environments match production core counts before the next big rollout, because a bad change that is months old can be impossible to roll back.

Local Development And Testing

AWS Ships Free Local Emulator Covering 200+ Services for Coding Agents

Image source: res.infoq.com

Summary: AWS released a free local emulator covering over 200 services, built for coding agents and tests. It runs without provisioning cloud resources or incurring charges. Great for iteration, but skip it for IAM and infrastructure-as-code validation.

Key Points:

  • The emulator is generated from more than 400 Smithy API models, so its coverage tracks AWS's own API definitions closely across services.

  • Fidelity varies by service. AWS publishes versioned fidelity tests and per-service scores with every release, so you can check specific services before relying on them.

  • The source is on GitHub under CC0 1.0 Universal, except third-party code in the vendor directory, which is licensed under Apache License 2.0.

  • An embedded Rust test mode runs the emulator inside the test process and wires its HTTP client directly into the AWS SDK, no separate server needed.

  • The emulator is generally available at no additional service fee, and it has no prebuilt image for macOS or Windows, only a short Dockerfile for container builds.

Why it matters: If your coding agent provisions real S3 buckets and Lambda functions just to test a change, this emulator removes that cost and latency entirely. Teams running integration tests in CI can use the embedded Rust mode to avoid hitting a shared dev account. But if your pipeline depends on IAM policy checks or CloudFormation-style infrastructure validation, keep testing against real AWS because the emulator explicitly does not cover those paths.

Container Isolation

AWS Launches Nitro Sandboxes: Per-Pod VMs at No Extra Charge

Image source: res.infoq.com

Summary: AWS launched Nitro Sandboxes, giving each opted-in Kubernetes pod, ECS task, or EC2 workload its own Firecracker microVM and guest kernel at no added cost. On EKS it's one pod field, so multi-tenant and regulated teams can get VM isolation without touching their containers. No GPU passthrough at launch, and SMT on x86 eats into your sandbox density.

Key Points:

  • Each sandbox runs as a Firecracker microVM with hardware-assisted virtualization, sharing neither kernel nor memory mappings with the parent instance or any other sandbox on the node.

  • Total sandbox vCPUs may not exceed five times the node's vCPU count, and sandbox count is capped at the node's vCPU count, halved on CPU-overcommit instance types.

  • Confidential mode uses a Cedar policy generated from the pod spec to pin image digests, arguments, environment, mounts, and capabilities, while denying exec and stdin entirely.

  • The Nitro Secure Module measures image and policy into PCRs, so a KMS key policy can release secrets only to sandboxes whose measurements match the expected values.

  • Warm pools via the open-source agent-sandbox project hand out a pre-booted VM in about two seconds, which is the number to check against your current pod startup latency.

Why it matters: If you run multi-tenant SaaS on EKS, your platform team can now set runtimeClassName to confidential-sandbox on the risky tenant pods and stop arguing about shared kernels. Anyone on standard EKS needs to build a sandbox-enabled AMI for their node groups or Karpenter, which is real work and the step most teams will underestimate. Pod Identity is required for credentials, so IRSA-based services need migrating before they can run sandboxed at all.

Cross-platform Ui Frameworks

Uno Platform 7.0 Ships With Apple TV Support and 5x Smaller Builds

Image source: res.infoq.com

Summary: Uno Platform 7.0 is out with Apple TV support, an experimental WebGPU backend, and builds roughly 5x smaller than 6.7. Mac Catalyst and native rendering are gone, so check your project format before upgrading.

Key Points:

  • Uno Platform 7.0 reports 5,219 commits and 704 closed issues since 6.7, making it the framework's largest release to date for .NET developers building cross-platform apps.

  • Median performance is about 2x faster than 6.7, with some operations hitting 38x speedups and output binaries roughly 5x smaller across all supported targets.

  • The experimental WebGPU backend delivers up to a 50% frame-rate improvement, using wgpu-native outside the browser and native browser WebGPU on WebAssembly targets.

  • Apple TV support arrives via the net10.0-tvos target framework, supporting tvOS 15 and later, with .NET 11 supported from day zero of this release.

  • SkiaSharp 4 becomes the default renderer, co-maintained by Uno Platform and Microsoft's .NET team, with Vulkan rendering by default on Android, Windows, and Linux.

Why it matters: If you're maintaining a WinUI-style app on Mac Catalyst, you'll need to move to macOS or another target before upgrading. Teams already on Uno.Sdk single-project layouts can adopt 7.0 with minimal friction, while anyone running a legacy multi-project setup should plan the migration now. The new WebGpuCanvasElement also gives graphics teams a path to issue raw WebGPU commands inside the UI tree.

New Tools & Job

  • APOB AI - APOB AI is a browser-based workspace for creators who use recurring AI characters in images and short videos. Generate video from text or an image, make a still avatar speak, synchronize lip movements to new audio, or edit elements of an existing clip. It suits character-led social videos, visual prototypes, and product demos. An account and generation credits are required; limited free access and optional paid plans are available.

  • Babydoll Dance AI - Babydoll Dance AI helps creators turn their own portrait, or photos of up to three people, into a coordinated Babydoll dance video. Upload a clear photo for the center dancer and optionally add friends on the sides. The output is a downloadable 15-second, 9:16 MP4. The site shows the example, photo requirements and credit price before generation. It uses one-time credit packs rather than a subscription.

  • TunedGift - TunedGift is a browser-based AI music tool for turning personal stories into songs for birthdays, weddings and anniversaries. Start with a recipient, an occasion and a few details from a shared memory. Review and edit the title and lyric draft, then listen to a one-minute audio excerpt before choosing whether to purchase the song. The workflow focuses on making a gift from your own details and gives you a way to review the lyrics and hear the result before checkout. Submitted by the TunedGift team.

  • ReelAI - Submitted by the ReelAI team with AI-assisted preparation. ReelAI helps creators animate photos into short videos using AI-powered motion templates and effects. It also offers AI image generation. Paid credit-based plans and current pricing are available at https://tryreelai.com/pricing.

  • Repostit Clips - Repostit Clips helps creators turn long YouTube videos, podcasts, interviews, webinars and livestreams into vertical short clips. It can find highlight moments automatically or let users choose them in a manual editor. Each clip receives a score, and users can schedule clips across their connected apps. Repostit is affiliated with this submission.

  • Senior AI Enablement Engineer, GTM - Why Harvey At Harvey, we’re transforming how legal and professional services operate. By combining frontier agentic AI, an enterprise-grade platform, and deep domain expertise, we’re reshaping how critical knowledge work gets done for decades to come. This is a rare chance to help build a generation

  • Lab Technician, AI Systems - Tenstorrent is leading the industry on cutting-edge AI technology, revolutionizing performance expectations, ease of use, and cost efficiency. With AI redefining the computing paradigm, solutions must evolve to unify innovations in software models, compilers, platforms, networking, and semiconductor

  • Research Scientist, Open Benchmarks Red Team - About Snorkel Snorkel AI is the frontier AI data lab, helping teams build the data and environments behind high-performing frontier and agentic AI. We combine technology with research-driven AI data development to create datasets, benchmarks, evals, and custom solutions for real-world AI systems. Fo

  • Application Security Engineer - Agents platform - About ElevenLabs ElevenLabs is an AI research and product company transforming how we interact with technology. We launched in January 2023 with the first human-like AI voice model. Today, we serve millions of users and thousands of businesses - from fast-growing startups to large enterprises like D

  • Director, Machine Learning Engineering - About the role Chime's Data Science & Machine Learning team powers decisions across the company, from how we protect members to how we build products for them. We're looking for a Director to lead that team through its biggest shift yet: moving from traditional models to a modern, AI-driven approach

QUICK HITS

Read more news here →

THAT'S A WRAP

Partner with Us

Showcase your product to our engaged community of Tech enthusiasts

Our newsletter reaches a diverse audience of tech professionals, investors, engineers, managers, and business leaders worldwide. Get in touch today.

More from the Tech Newsletter

Tech Newsletter

Get the next issue

One email a week. Free. Unsubscribe anytime.

How we handle your email: Privacy policy

Comments

Subscribe to join the conversation...

Be the first to comment