Microsoft Ignite 2026: Every Announcement, Availability and Caveat Microsoft opened Ignite 2026 in San Francisco on November 17 with a set of Copilot and model announcements, including general availability for Home and Office in Copilot, Copilot Managed Runtime, MAI Code 2 and MAI Code 2 Flash in GitHub Copilot, Microsoft-Decision-1.1, Microsoft IQ, Perception and Cobalt 200 VMs, which Microsoft says is up to 50% faster than Cobalt 100. Copilot Code, Autopilot and Workspaces remain in the Frontier program, with Copilot Code reaching general availability in December per the Book of News, while Copilot Today's Brief and Feed sit in private preview. Satya Nadella framed Copilot as "the new OS for work" and described the model goal as "the most intelligence out of every model, for every task, at the best possible price," with MAI Cyber 2 Flash (Codename MDASH) in preview at 12 billion active parameters and Agent 365 Cost Management also in preview. Microsoft opened Ignite 2026 on November 17 in San Francisco with one message: AI needs context, choice, control and a bill you can read. Satya Nadella called Copilot "the new OS for work," then spent the keynote on four layers under it: models, enterprise context, agent governance and security. This page is the single index. It sorts the Ignite 2026 announcements by status generally available, Frontier, preview because Microsoft's own wording blurs them, and it links to our deeper posts on each piece. Facts come from Microsoft's live blog, official blog post https://blogs.microsoft.com/blog/2026/11/17/ignite-2026-building-the-frontier-ecosystem/?ref=explainx and Book of News https://news.microsoft.com/events/ignite/2026/book-of-news/?ref=explainx . TL;DR: what shipped and what is still preview | Announcement | Status per Book of News | What it does | |---|---|---| | Home and Office in Copilot | GA | Chat and Cowork in one home; Word, Excel, PowerPoint inside Copilot | | Copilot Managed Runtime | GA | Secure hosting for agents and workflows you build | | Copilot Code, Autopilot, Workspaces | Frontier | Plain-language app building, a persistent agent, a shared team space | | Copilot Today Brief and Feed | Private preview | Morning briefing that refreshes through the day | | MAI Code 2 and Code 2 Flash | GA in GitHub Copilot | Agentic coding models; Flash is the high-volume tier | | Microsoft-Decision-1.1 | GA | Cheap model for routing, classification and verification | | MAI Cyber 2 Flash | Preview Codename MDASH | Security model with 12B active parameters | | Agent 365 Business Analytics | Frontier | AI spend, adoption and outcomes in one view | | Agent 365 Cost Management | Preview | Spend policies, limits and alerts for admins | | Microsoft IQ | GA, no new SKU | Context layer: Work IQ, Fabric IQ, Foundry IQ, Web IQ | | Perception | GA | Red, blue and green security agents driven by playbooks | | Salesforce in Copilot | Preview | 38+ sales skills via Salesforce MCP access | | Cobalt 200 VMs | GA | Up to 50% faster than Cobalt 100 | What is the new Copilot, and who gets it today? Microsoft split the release in two. The surfaces a normal Microsoft 365 tenant can switch on now are Home, where Chat and Cowork meet, and Office in Copilot, which puts Word, Excel and PowerPoint beside the chat. Both are generally available. Copilot Managed Runtime, which hosts what you build, is also generally available. The agent-heavy parts sit behind the Frontier program. Code turns plain language into working apps and dashboards. Autopilot is a persistent agent with its own agent ID and memory that you can @mention in Teams. Workspaces is Nadella's "multiplayer form factor, where people, their agents and other people" share context. Copilot Code reaches general availability in December, according to the Book of News. Copilot Today is the most limited. The Today Brief ships with a Microsoft 365 Copilot license, but the Today Feed needs Copilot credits, and the whole experience is in private preview in Outlook and the Copilot app. Our deeper reads: Copilot Today, Pages and Workspaces https://explainx.ai/blog/microsoft-copilot-today-pages-workspaces-office-ignite-2026 and Autopilot and the Frontier rollout https://explainx.ai/blog/microsoft-copilot-autopilot-m365-frontier-rollout-november-2026 . Which models did Microsoft ship, and what do they cost? The model story is cost per task, not a new leaderboard. Nadella described the goal as "the most intelligence out of every model, for every task, at the best possible price," with three purpose-built areas. - Coding: MAI Code 2 targets complex, long-horizon, multi-file work and repo-wide reasoning. It is generally available in the GitHub Copilot model picker and in Copilot Fusion. MAI Code 2 Flash is the fast, high-volume tier. - Cybersecurity: MAI Cyber 2 Flash has 12 billion active parameters. Microsoft says it lands within a few points of larger cyber models at a fraction of the price. The Book of News lists it as preview only in Codename MDASH. - Decisions: Microsoft-Decision-1.1 is generally available in Foundry, OpenRouter and Vercel. The Book of News says it is built on Gemma 4 26B A4B, adds image input, and scores 68 on the Artificial Analysis Decision Index v2.1, up six points from the first model. More than 4,000 customers already use the first version. No source gives token prices for the MAI models. Benchmark and price-ratio claims in the press come from Microsoft and have not been independently reproduced. We cover the numbers and the caveats in MAI Code 2 and the six specialist models https://explainx.ai/blog/microsoft-ignite-mai-code-2-seven-models-november-2026 . Mustafa Suleyman framed the family as "frontier-level capability at a price organizations can afford to run at scale." Jay Parikh added the routing side: Microsoft Foundry now lists more than 11,000 models, and its model router expanded to the top 50. Auto routing in Copilot picks the model per task, trading quality against cost on every request. GPT 6.2 and Meta's latest models are listed as coming to Foundry with no date; see GPT-6.2 Astra Sol launch pricing and benchmarks https://explainx.ai/blog/openai-gpt-6-2-astra-sol-launch-pricing-benchmarks-november-2026 for the OpenAI side. What is Agent 365 Business Analytics? This was the most practical announcement for finance and IT leads. Microsoft cited a study claiming AI spending can nearly quadruple as organizations move to enterprise-wide adoption, and that 93% report exceeding their AI budgets. That is a vendor-cited figure, so treat it as directional. Agent 365 splits the problem in two. | Layer | Audience | Status | What it shows | |---|---|---|---| | Business Analytics | Business leaders | Frontier | Spend, adoption, success rate, user satisfaction, assisted hours | | Cost Management | IT admins | Preview | Spend by user, agent and model; policies, limits, alerts; Fabric chargeback | Nadella's line was that Business Analytics is "your ERP system for AI usage." Leaders can ask questions through agentic Q&A, build custom reports and push data into existing reporting tools and data lakes. Charles Lamanna's pitch was measuring "the work AI helps complete, not just the tokens it consumes." Licensing is where the fine print lives. Insights on Microsoft services need at least one Microsoft 365 Copilot license. Insights on non-Microsoft services need an Agent 365 license. Forecasting and optimization recommendations reach the Frontier program "this month," and value insights across Microsoft and non-Microsoft agents arrive in December. What is Microsoft IQ, and does it need a new license? Microsoft IQ is an enterprise context layer. It combines Work IQ, Fabric IQ, Foundry IQ and Web IQ with more than 350 third-party connectors, so an agent starts with a governed picture of how your business works instead of a blank prompt. Per the Book of News, it is generally available across Microsoft Copilot, GitHub Copilot, Foundry and MCP-compatible harnesses, with no new SKU. The unit you build is a business plugin that captures know-how, skills and tools. Business plugins themselves are in the Frontier program. | IQ layer | Status highlights | |---|---| | Work IQ | Memory and longer workflows GA; Listen event-triggered agents and granular approvals in preview; OpenAI integration GA, Anthropic "coming soon" | | Fabric IQ | Copilot and Fabric app integration GA; ChatGPT and Claude connectors in preview | | Foundry IQ | SharePoint knowledge source GA; knowledge graphs and Microsoft-Retriever-1 in preview | | Web IQ | GA for all Azure customers at a lower price no figure given ; Places covers 207.5M entities | If you build agents, the practical point is that context ships as portable plugins over MCP https://explainx.ai/blog/what-is-mcp-model-context-protocol-guide and agent skills https://explainx.ai/blog/what-are-agent-skills-complete-guide rather than as per-agent prompt files. The related Power BI path is covered in Power BI authoring plugin and Fabric agents https://explainx.ai/blog/microsoft-powerbi-authoring-plugin-skills-fabric-agents-2026 . What is Copilotforce, the Salesforce tie-up? Nadella announced integrations of Salesforce across Copilot, Teams and Fabric. Microsoft's own page references it as "CopilotForce"; the live blog spells it "Copilotforce." - Salesforce in Copilot is in preview. It ships 38+ prebuilt sales skills and uses authenticated Salesforce MCP access, so sellers act inside their existing Salesforce permissions. - Sales Cloud in Teams is in pilot. - Data Federation between Salesforce Data 360 and Fabric OneLake is generally available, as is the Salesforce CRM migration experience. Salesforce's Rohan Kumar and Accenture CIO Tony Leraris joined Judson Althoff on stage. Microsoft also made its own Dynamics 365 Sales and ERP apps plugins generally available, so the message is "bring your CRM, we will meet it." How does Perception change security work? Perception is now generally available. Microsoft describes a three-layer cyber stack: signals, sensors and actuators; context; and agents, harnesses and models. Perception sits at the top. It runs blue, red and green team agents through playbooks, in a closed loop, with guardrails and human sign-off for high-impact actions. Key details from the Book of News: - The Triage Agent covers about 80% of alerts in a typical Defender queue and now handles AWS and Salesforce signals. - The Web App Pen Test Agent and the Reduce exposure risk playbook are in preview. Secure AI agents playbooks are in private preview. - Codename MDASH, in preview, runs 100+ specialized agents in a four-stage pipeline. It supports MAI Cyber 1.1 Flash now, with MAI-Cyber-2 after evaluation. - Defender runtime protection for local agents is generally available on Windows and covers Claude Code, GitHub Copilot CLI and VS Code extensions. It needs Defender for Endpoint P2. - Entra Agent ID for local agents such as GitHub Copilot CLI and Claude Code is in preview. Microsoft also launched the AI Transparency Initiative, an industry effort to set a common approach to AI transparency, security and governance across models and providers. Details are in our AI Transparency Initiative post https://explainx.ai/blog/microsoft-ai-transparency-initiative-ignite-november-2026 . Nadella's framing, "These nested black boxes are simply unacceptable," matches his earlier emergency-brake argument on compromised models https://explainx.ai/blog/nadella-assume-ai-models-compromised-emergency-brake-containment-2026 . For teams already running coding agents on laptops, the actionable item is discovery: Intune can now mark devices with unapproved local agents as noncompliant. Pair platform controls with runtime checks on what an agent is allowed to do; AgentBeam https://agentbeam.com/?ref=explainx is the agent security platform from the explainx.ai team for that second layer. The supply-chain risk is real: see the Microsoft Agent Framework PyPI credential stealer https://explainx.ai/blog/microsoft-agent-framework-pypi-1-26-credential-stealer-november-2026 from the same week. What changed in infrastructure and developer tools? - Cobalt 200 VMs are generally available, with up to 50% better performance than Cobalt 100. Microsoft says customers like Databricks, OpenAI and Snowflake use them. Regional counts differ: the live blog says 23 regions by year end, while the Book of News says 24 on top of 11 existing. We use "23 to 24" until Microsoft reconciles them. - Maia 300 was cited at 40% more tokens per watt than leading GPU-based systems; see our Maia 300 analysis https://explainx.ai/blog/microsoft-maia-300-inference-chip-40-percent-tokens-per-megawatt-2027-2026 . - Copilot SDK 2.0 is generally available with a native Rust runtime and six languages. Model choices in GitHub Copilot are covered in Copilot Fusion https://explainx.ai/blog/github-copilot-fusion-generally-available-effort-levels-single-cascade-critique-30-percent-discount-2026 . - Windows Server 2028 is in preview through the Insider Program, and Windows 365 for Agents is eligible for Copilot Credits. - Foundry made multi-turn evaluation, production insights and Agent Optimizer generally available. What should builders and buyers do this week? 1. Check your tenant. Home and Office in Copilot are GA; everything agentic is Frontier. Ask your admin whether the tenant is enrolled before planning pilots. 2. Price the licenses, not the demo. Business Analytics, Agent 365 and the Today Feed each carry a different license or credit requirement. 3. Test the cheaper models on your own tasks. MAI Code 2 Flash and Decision-1.1 are claimed as cost levers. Run them on a real repo and a real routing task before you move traffic. 4. Inventory local agents. Defender and Intune can now see Claude Code, Copilot CLI and others; decide your allow-list first. 5. Wait on preview items. Perception playbooks, Dynamic Agent Risk and MDASH are not production commitments yet. Changelog Each entry below is dated, so you can see when a status or announcement changed. - Nov 18, 2026: First version, built from the live blog, official blog and Book of News. Related reading Status labels follow Microsoft's Book of News as of November 18, 2026 and may change during Ignite week.