BriefTechNews
Microsoft 365 falls over, Gemini gets a decision engine, and bots get banned

Microsoft 365 falls over, Gemini gets a decision engine, and bots get banned

7 min read · 11 sources

TL;DR
  • Microsoft 365 outage on November 12 hit email, Teams, and Intune due to failed server health checks and an overwhelmed process.
  • Google's Gemini Decisions API scores up to 1,000 questions per request in one pass with 150-250 ms latency at $0.10 per million input tokens.
  • Gartner forecasts 10.6 million AI-created jobs versus 8.9 million lost by 2030, a net gain of 1.7 million.
  • United Airlines blocked customer bots, citing data leakage and FAA/DOT violations, following Walmart, Target, and Amazon.
  • AWS Cloud WAN now supports VPC Flow Logs with no additional charge, capturing IP traffic across global networks.

Your email was down again. Microsoft’s November 12 outage took out Exchange Online, Teams features, and the Intune admin console, with the company blaming failed server health checks and an overwhelmed server process. Recovery meant rerouting traffic and restarting servers, which is a fancy way of saying the fleet got tired and needed a nap. It’s the kind of day that makes you wonder why you migrated to the cloud in the first place, and then you remember: because you didn’t want to rack servers.

The bigger story is that Google just shipped a model that makes decisions instead of generating text. The Gemini Decisions API is a closed-set scorer, not a chatbot. It reads the question, outputs a probability, and charges you almost nothing for it. If you run agents, route traffic, or enforce policy, this changes your cost math. And if you run a Mac fleet, Claris wants to turn it into an AI cluster.

At $0.10 per million input tokens and zero output tokens billed, the Gemini Decisions API is roughly five times cheaper than Gemini 3.5 Flash-Lite for decision workloads.

Gemini Decisions API: the 150 ms answer to agent routing

Source: blog.google ↗

Google’s new Gemini Decisions API is powered by Gemini Decision 1 (gemini-decision-1), a model built for fast, closed-set decisions: tool selection, policy checks, model routing, yes/no, multiple choice, rubric scoring. It takes multimodal input (text, images, audio, video, PDFs) up to 256K tokens and answers up to 1,000 independent questions in a single forward pass. End-to-end latency is 150-250 ms, and the model returns a confidence score from 0 to 1 for each answer.

The pricing is the headline: $0.10 per 1M input tokens, with zero output tokens billed. Google claims that’s nearly 5x cheaper than Gemini 3.5 Flash-Lite for decision workloads. For engineers building agent orchestrators, this is the difference between a routing layer that costs real money per call and one that costs fractions of a cent. The calibrated confidence scores also mean you can set thresholds: route to a human or a bigger model only when the score drops below 0.8. That’s the kind of knob SREs like to turn.

Gartner's job math: net positive by 2030, painful in 2026

Source: computerworld.com ↗

Gartner’s forecast says AI will create 10.6 million jobs globally (excluding China and India) by 2030, while eliminating 8.9 million, for a net gain of 1.7 million. The technology industry sees the largest net gains: 1.9 million new tech jobs by 2030. The catch is the near term. Gartner expects 5 million jobs to be deferred as employers hold off hiring while they figure out what AI can do. In 2026 alone, 2.2 million jobs are lost and only 1.6 million gained.

Entry-level candidates are most exposed to the deferral, which means the junior engineers you’d normally hire to run scripts and write boilerplate are the ones getting squeezed. The report’s logic is that AI deployments mature, productivity gains compound, and new roles appear. But if you’re a hiring manager, the next 18 months look like a hiring freeze with extra steps. Plan your succession accordingly.

The B2A economy: agents are your new customers

Source: a16z.news ↗

a16z’s argument is blunt: AI agents now make up most web traffic, write over half of new articles, and generate most production code. That means the customer you’re serving is increasingly a machine, not a human. The numbers back it up: Shopify reported AI-driven traffic growing eightfold year-over-year in Q1 ‘26, and agentic browser traffic grew 8x between August and October, with Muse driving 60% of that activity.

The consequence is architectural. If agents are your customers, your site needs to serve them without JavaScript, without CAPTCHAs, and with structured data they can parse. Blocking bots is now blocking revenue. The flip side is that agent-referred traffic reportedly converts better than human traffic, because the agent already did the filtering. If you’re not building an agent-friendly API layer, you’re ceding that distribution to a competitor who is.

Claris Harbor: your Mac fleet is now an AI cluster

Source: computerworld.com ↗

Claris Harbor, from Apple’s business subsidiary, turns Apple Silicon Macs into private AI servers. The app sets up an endpoint on a single Mac, or pools unified memory across multiple machines via Thunderbolt 5 for larger models. Data and prompts stay on physical hardware. It’s in private preview with a mailing list for early access.

For enterprises with hundreds of Macs sitting idle overnight, this is a way to run inference without buying GPU servers or paying cloud egress. The Thunderbolt 5 pooling is the interesting part: it means a rack of Mac Studios becomes a memory-coherent cluster for models that wouldn’t fit on one machine. The obvious caveats are management overhead and the fact that Apple silicon’s GPU performance is not a data-center GPU. But for confidential workloads where data cannot leave the building, this is a genuinely new option.

Databricks Genie: from analysis to autonomous action

Source: databricks.com ↗

Databricks’ new Genie is an autonomous AI coworker that handles long-running tasks continuously, tracking progress and acting as new information arrives. It’s grounded in business context via Genie Ontology, governed by Unity Gateway and Unity Catalog, and accessible through Genie One, email, Slack, and Microsoft Teams. Use cases include monitoring claims for fraud patterns and managing marketing campaigns, with users setting guardrails and approving actions.

The shift from analysis to execution is the thing to watch. Genie doesn’t just tell you there’s fraud; it files the report and alerts the right team. That requires permissions, approval flows, and cost visibility, which is what Unity Gateway provides. If you’re evaluating it, ask how it handles the approval loop: the difference between a useful tool and a liability is whether the agent can act without a human in the loop, and whether you can see what it spent.

Gemini on CarPlay, and the open standards that keep agents honest

Source: blog.google ↗

Google’s Gemini app is rolling out to Apple CarPlay, enabling hands-free conversations with Gemini Live from the car display. It integrates with Google Workspace for daily briefs and calendar prep, and syncs with the iPhone app. It’s a voice-first UX problem: the interface is the road, and the model has to be good enough to handle interruptions and background noise. For engineers, it’s another platform to support, but the interesting part is the distraction-minimizing design.

Meanwhile, Anthropic donated MCP and Agent Skills to the Linux Foundation’s Agentic AI Foundation, with Google, OpenAI, and others co-maintaining the standard under vendor-neutral governance. MCP standardizes live connections to tools and data via a stateless transport model; Agent Skills use progressive disclosure to load procedural knowledge only when needed. The MCP Toolbox for Databases solves the N-by-M integration problem between clients and data hosts. Without these standards, every agent framework reinvents the same database connector, and every enterprise ends up supporting a zoo of incompatible protocols.

United bans bots, and the ecosystem pays the price

Source: computerworld.com ↗

United Airlines blocked customer bots, citing data leakage, unintended purchases, and violations of FAA and DOT regulations. It follows Walmart, Target, and Amazon. The airline is restricting access via its digital platforms and travel agencies, updating booking and distribution policies. Analysts point to a lack of industry standardization: supporting multiple bots drains IT resources, and distinguishing a helpful agent from a scraping bot is genuinely hard.

This is what happens when agents outpace standards. The fix isn’t more bot-blocking; it’s a protocol that lets an agent prove intent and scope before it touches a booking engine. Until MCP or something like it becomes the default for enterprise transactions, expect more of these bans, and expect your own agent integrations to get collateral damage.

AWS Cloud WAN Flow Logs, and your brain on AI

Source: aws.amazon.com ↗

AWS Cloud WAN now supports VPC Flow Logs, giving central visibility into IP traffic across global networks. Logs capture IP addresses, ports, protocol, traffic volume, packet-loss counters, and Cloud WAN metadata like segment and Region, published to CloudWatch Logs, S3, or Data Firehose. No additional charge. If you’ve been stitching logs from multiple VPCs and Regions to troubleshoot a cross-network issue, this kills that whole class of problem.

Finally, a former neuropsychologist at Criteo argues that as AI speeds up code production, human skills like understanding and validation become more critical. The practical advice: leverage the Default Mode Network for background problem-solving, use spacing and retrieval practice, respect working memory limits, and apply scientific debugging. It’s a 17-minute read that’s worth it if you’ve noticed your comprehension slipping as your AI assistant writes more of your code. The tools write faster; your brain still has to verify.

Get the brief

Liked this one? Pick the beats you care about and get just those, five minutes a morning. Free, no hype, one click to leave.

About Me Author

My name is

BriefTechNews

A daily digest of what actually moved in AI, tech, crypto and fintech, assembled and written with AI, with every item linked to its source. Read More
Tags

You May Also Like