<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6" -->

---
title: Microsoft Ignite: new MAI models, Azure agent tooling...
description: Microsoft Ignite opened with new MAI models and Azure agent tooling for .NET developers. Microsoft announced MAI Decision 1, MAI Code 2, Code 2 Flash and Cyber...
canonical: https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Microsoft Ignite: new MAI models, Azure agent tooling for .NET, and Satya Nadella on verifying AI code | daily.dev
og:description: Microsoft Ignite opened with new MAI models and Azure agent tooling for .NET developers. Microsoft announced MAI Decision 1, MAI Code 2, Code 2 Flash and Cyber...
og:url: https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6
og:image: https://api.daily.dev/og/posts/sIMnRyld6.png
og:image:alt: Microsoft Ignite: new MAI models, Azure agent tooling for .NET, and Satya Nadella on verifying AI code
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Microsoft Ignite: new MAI models, Azure agent tooling for .NET, and Satya Nadella on verifying AI code

**[Collections](https://daily.dev/sources/collections)** · 6 min read · 1 upvotes · 0 comments

## Summary

Microsoft Ignite opened with new MAI models and Azure agent tooling for .NET developers. Microsoft announced MAI Decision 1, MAI Code 2, Code 2 Flash and Cyber 2; MAI Decision 1 is claimed to be 10x faster and 200x cheaper than a general model and is available on Foundry and OpenRouter. New Azure features include Container Apps sandboxes that run tool calls in microVMs starting in under a second with egress rejected by default, Azure Functions hosted skills defined in markdown with MCP tools, an AI gateway for model routing and budgets, and an SRE agent used by the team on hundreds of thousands of incidents. Separately, a podcast with Satya Nadella covers verifying AI-generated code, avoiding model lock-in, and Gergely asking why Microsoft made voluntary exit offers to most of its software engineers.

## Content

Microsoft used Ignite to show that it is building its own model stack alongside its OpenAI partnership. New in-house MAI models cover code, decisions, image and cyber. Satya Nadella also made a fairly consistent argument in interviews: don't tie yourself to one model.

## The new MAI models

**MAI Code 2 and Code 2 Flash.** Microsoft AI says these are aimed at everyday enterprise coding at lower cost than frontier models.

- MAI Code 2 targets complex, long-horizon work across multiple files or a whole repository, including frontend projects. Averaged over four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, FrontierCode v1.1), Microsoft says it reaches 98% of GPT 6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5's performance at 5% of the cost. In blind frontend evaluations, reviewers preferred its output to GPT-6 Sol 67% of the time.
- MAI Code 2 Flash is for high-volume work like code explanations, quick edits, routine fixes and sub-agent workloads. Microsoft says it matches GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their respective costs.
- Neither is tuned for deep science research or long-horizon agentic work.
- Both arrive through Microsoft Foundry in private preview. Kilo is a launch partner and will add them to its library of more than 500 models.

**MAI Decision 1.** Microsoft says it is 10x faster and 200x cheaper than a general model. It is available on Foundry, OpenRouter and other providers. Rather than free-form text, it returns structured answers for jobs like routing support requests or grading AI responses.

Vercel has already put Decision-1.1 on its AI Gateway. It supports boolean, choice and score answers, and several questions can share one input in a single request. The request format matches Decision-1, so upgrading means changing the model ID to `microsoft/decision-1.1`. It works through the AI SDK (7.0.128+), the OpenAI-compatible Decisions API, the OpenAI Python SDK (3.26.0+) or the AI CLI.

**Image models.** MAI-Image-2.7 and MAI-Image-2.7-Flash are in public preview on Vercel AI Gateway. Both output up to 2048 x 2048 pixels, support transparent backgrounds, and accept up to five reference images for edits or for combining subjects. The first is for final-quality assets; Flash is faster and cheaper for exploring prompts. Developers need only an AI Gateway key, not a Microsoft account.

MAI Cyber 2 was also announced, with few details so far. @finkd said Muse Wonder, Muse Spark 1.3 and Muse Ember are coming soon to Microsoft Foundry on Azure.

## Microsoft is shipping more than it looked like

According to Epoch AI's public model database, as charted by @Heinric98694776, Microsoft had zero models logged for July to September when the chart was first drawn in early October. For 1 October to 16 November it leads the quarter with 6 releases, and 7 for the quarter overall: two MAI-Voice 2.1 models, a streaming transcriber, a new base model, an image model and a cyber model. Google has 5, OpenAI, Anthropic and Meta have 3 each, and only 4 of the 32 models logged are open weights. The caveat is that databases fill in late, so recent counts keep rising.

## What's new for .NET developers on Azure

A session for .NET developers building agents covered several features:

- Azure Container Apps sandboxes run tool calls in microVMs with sub-second startup. In the demo, network egress was rejected by default and managed identity was withheld when a document contained malicious instructions.
- Azure Functions hosted skills let developers define LLM-driven routes in markdown files with MCP tool definitions. A durable functions example batched and classified 20 delivery photos.
- The AI gateway adds model routing, per-client token and monetary budgets, and failover priorities without redeploying agents.
- The SRE agent, used in production for hundreds of thousands of incidents, narrows hypotheses and can mitigate issues before a human responds.

## Nadella on staying flexible

In a conversation with @petergyang and a podcast with @GergelyOrosz, Nadella returned to a few ideas.

**More than one model, more than one cloud.** Asked whether Microsoft can catch the frontier labs, he compared it to people asking whether anyone needs a cloud besides Amazon. His answer: the world will demand more than one cloud and model, and customers want flexibility and optionality at sensible economics.

**Every employee manages agents.** He described a company shaped like a startup: a founder with a token budget, who hires people who also get token budgets. All ICs become managers of agents. As evidence he said Copilot Cowork was built by 11 developers, and one developer maintains the entire GitHub Copilot runtime. Business apps, he said, are becoming headless plugins to Copilot, used more than before because agents work around the clock. His own Autopilot reads his OneNote and meetings, pulls the customer's Dynamics CRM data and checks the supply chain system.

**Treat models as insider risk.** He used to think alignment would emerge with scale; based on empirical results, he now says maybe not. Since the liability sits with Microsoft, he wants a trust architecture independent of the model, and compares agents to insider risk. Orosz relayed the same point: models are intelligent but unreliable, so build a separate system to verify agent behavior.

**Don't build around one model.** Assume the next model drop forces a replan so the model doesn't become a boat anchor. Nadella wants to turn the model business into a database business: log what agents do, fine-tune on those traces, and repeat with each upgrade. He also suggested keeping harness, memory and model decoupled, using open models to protect core IP, and running private evals to choose and upgrade models.

**Software factories.** On the podcast he pointed to the Azure networking team, which manages 500+ fiber network operators and built an agentic system to handle emails and processes. He called this the reality of AI transformation: brownfield work on factories that improve how teams operate while the daily work continues.

Orosz also asked directly about Microsoft's voluntary exit offers to most of its software engineers, a first for the company, and whether layoffs could follow. By his account, Nadella's reply was diplomatic.

The podcast also covers verification of AI-generated code, junior hiring, open models, local and hybrid intelligence, and Microsoft's model strategy.

## Questions this post answers

### What are Azure Container Apps sandboxes for running AI agent tool calls?

They run tool calls in microVMs that start in under a second, with network egress rejected by default. In a demo, managed identity was withheld when a document contained malicious instructions, showing the sandbox denying access even when an agent is tricked into acting on injected content.

_Developers hardening agent tool execution can track Azure's sandboxing approach on daily.dev._

### How much faster and cheaper is Microsoft's MAI Decision 1 model compared to a general model?

MAI Decision 1 is described as 10 times faster and 200 times cheaper than a general model. It is available on Microsoft Foundry, OpenRouter, and other providers, with more details on the related MAI Code 2 model still to come.

_Teams comparing model costs and latency can follow MAI Decision 1 updates on daily.dev._

### Why did Microsoft offer voluntary exit packages to its software engineers?

Microsoft offered voluntary exit packages to most of its software engineers for the first time in company history. Satya Nadella gave a diplomatic answer when asked directly about this and whether layoffs could follow, without a fully direct explanation of the reasoning.

_Engineers tracking shifts in Microsoft's workforce strategy can follow the story on daily.dev._

## Community take

How the wider developer community reacted, aggregated from 1 discussion and 108 comments across hackernews (as of 2026-11-18).

**TL;DR:** The release is seen as a credible but unspectacular entry: people are intrigued that Microsoft trained frontier-scale models from scratch and hit good price/performance, but many are skeptical of the benchmark choices, the promotional pricing, and the usual Microsoft branding/availability confusion.

**Sentiment:** 20% positive · 45% mixed · 35% skeptical

**The case for**

- Price-to-performance numbers look genuinely competitive if they hold up in real use.
- Some see it as validation that smaller, task-focused models can rival larger ones on coding specifically.
- Owning the hardware/compute stack is viewed as a real long-term advantage for Microsoft.
- Training from scratch rather than relying on distillation earns some credit.

**The pushback**

- Benchmarks cited (DeepSWE 1.1, Terminal Bench 2.1, FrontierCode 1.1) are seen as outdated versions, undermining confidence in the comparisons.
- Comparisons use a 50% promotional price and max-effort settings for competitors, seen as cherry-picked.
- Hard to find where to actually access or try the models — no clear pricing page, API keys, or playground.
- Doubts that any real users are adopting MAI models, with claims it mainly appeals to decision-makers rather than practitioners.
- GitHub Copilot's broader product is seen as not competitive enough regardless of the new models.
- Frustration with Microsoft's branding churn and unclear relationship between 'MAI Code' as model vs. 'Copilot' as harness.

**By community**

- hackernews (mixed): Discussion splits between genuine interest in Microsoft's compute advantage and skepticism about cherry-picked benchmarks, pricing gimmicks, and poor product discoverability, with a long tangent debating Chinese model distillation.

**Hottest debate:** Whether Chinese (and now Microsoft) models are closing the gap with frontier labs through genuine innovation or largely via distillation of GPT/Claude outputs.

**Open questions**

- How would these models actually compare against current Deepseek, GLM, and Qwen models on price and performance?
- Will the benchmarks and blind-preference results hold up once developers test the models directly rather than relying on launch marketing?
- Where and how can developers actually access and pay for these models outside of GitHub Copilot?

**Highlights**

> TBH, DeepSWE 1.1 has been busted for a while (is anyone out there arguing Gemini 3.8 Flash is as good as Astra lol?), and TB2.1 as a top line when they are currently at 4 (which is substantially harder) and FrontierCode at 1.1 when it's currently at 2.0 does not bring a lot of confidence. Of course I always want to see more competition in this space, though!
> — [johnfn on hackernews](https://news.ycombinator.com/item?id=50538426)

> What annoys me a lot about MSFT releases are, that is not possible to find how to use these models and where to try them. I didn't hear anyone in my circle using MAI models, none. Because even on this blog post, they are talking about pricing, but there is no Pricing page on that website, neither way to create API keys, nor playground. With this success, I can also say, I am releasing a model which costs 0.0001$ / input/output MTok with 100x cheaper cache read and write which is on part with Opus 5.5. (its your problem to find that model)
> — [throwaw12 on hackernews · 4 comments](https://news.ycombinator.com/item?id=50537289)

> The claimed 20% the cost suggests that the models are 20% the size. Which means these models are flying in the face of an orthodoxy that says the only way to make a model better at coding is to make it better at everything. I think this coming year is when we'll see that orthodoxy fall, as people realize you can make a model better at coding by making it better at coding.
> — [kennywinker on hackernews · 3 comments](https://news.ycombinator.com/item?id=50536994)

> While showing MAI Code 2 at a 50% promotional discount, and comparing to Max reasoning effort for Opus 5.5 and 6.1 Sol...
> — [user43928 on hackernews](https://news.ycombinator.com/item?id=50536992)

> It seems they trained from scratch: https://microsoft.ai/news/frontier-pre-training-at-microsoft...
> — [doomroot13 on hackernews · 1 comments](https://news.ycombinator.com/item?id=50537563)

> We just cancelled our Github Copilot subscription. It was hard to even justify the 19,- dollar tier. All it gave us was some credits (at API prices) which most people burned through in days, some in hours, code reviews (that we couldn't customize) and another service to manage. Having AI integrated in Github is a decent idea but just like Microsoft Copilot, it's just not good enough at the moment. These two models might help, I will definitely give it a try but while they're cheap they still do not compare to what feels like nearly unlimited usage offered by Claude Code and Codex (if you stick to the lower tier models).
> — [LaurensBER on hackernews · 2 comments](https://news.ycombinator.com/item?id=50538591)

**Source threads**

- [hackernews](https://news.ycombinator.com/item?id=50535885) · 189 points · 108 comments

## Similar posts on daily.dev

- [Exploring Microsoft’s AI foundation models](https://daily.dev/posts/exploring-microsoft-s-ai-foundation-models-gieb4896e) · InfoWorld · 3 upvotes · 1 comments
- [Microsoft is racing to make OpenAI optional](https://daily.dev/posts/microsoft-is-racing-to-make-openai-optional-ip9gjmzpu) · The New Stack · 10 upvotes · 11 comments
- [Microsoft swaps in its own AI over OpenAI in some apps](https://daily.dev/posts/microsoft-swaps-in-its-own-ai-over-openai-in-some-apps-ru7dnzqi7) · The Next Web · 0 upvotes · 0 comments
- [Microsoft joins AI cost-cutting trend by relying more on its own models](https://daily.dev/posts/microsoft-joins-ai-cost-cutting-trend-by-relying-more-on-its-own-models-paf5bf3lb) · TechCrunch · 0 upvotes · 0 comments

---

Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#llm](https://daily.dev/tags/llm), [#microsoft](https://daily.dev/tags/microsoft), [#azure](https://daily.dev/tags/azure)

[View this post on daily.dev](https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Microsoft Ignite: new MAI models, Azure agent tooling for .NET, and Satya Nadella on verifying AI code","url":"https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6"},"datePublished":"2026-11-17T18:33:15.663Z","dateModified":"2026-11-18T00:42:44.570Z","description":"Microsoft Ignite opened with new MAI models and Azure agent tooling for .NET developers. Microsoft announced MAI Decision 1, MAI Code 2, Code 2 Flash and Cyber...","image":"https://pbs.twimg.com/media/HXWOfhwWgAAr7MO.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HXWOfhwWgAAr7MO.jpg","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,llm,microsoft,azure","timeRequired":"PT6M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Microsoft Ignite: new MAI models, Azure agent tooling for .NET, and Satya Nadella on verifying AI code"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/microsoft-ignite-new-mai-models-azure-agent-tooling-for-net-and-satya-nadella-on-verifying-ai-co-simnryld6#faq","mainEntity":[{"@type":"Question","name":"What are Azure Container Apps sandboxes for running AI agent tool calls?","acceptedAnswer":{"@type":"Answer","text":"They run tool calls in microVMs that start in under a second, with network egress rejected by default. In a demo, managed identity was withheld when a document contained malicious instructions, showing the sandbox denying access even when an agent is tricked into acting on injected content. Developers hardening agent tool execution can track Azure's sandboxing approach on daily.dev."}},{"@type":"Question","name":"How much faster and cheaper is Microsoft's MAI Decision 1 model compared to a general model?","acceptedAnswer":{"@type":"Answer","text":"MAI Decision 1 is described as 10 times faster and 200 times cheaper than a general model. It is available on Microsoft Foundry, OpenRouter, and other providers, with more details on the related MAI Code 2 model still to come. Teams comparing model costs and latency can follow MAI Decision 1 updates on daily.dev."}},{"@type":"Question","name":"Why did Microsoft offer voluntary exit packages to its software engineers?","acceptedAnswer":{"@type":"Answer","text":"Microsoft offered voluntary exit packages to most of its software engineers for the first time in company history. Satya Nadella gave a diplomatic answer when asked directly about this and whether layoffs could follow, without a fully direct explanation of the reasoning. Engineers tracking shifts in Microsoft's workforce strategy can follow the story on daily.dev."}}]}
```

