Microsoft's homegrown AI debut

PLUS: OpenAI launches gpt-realtime for voice agents

Good morning, AI enthusiasts. For years, Microsoft's AI strategy has been synonymous with OpenAI — but that narrative just got complicated.

The company's new MAI-Voice-1 and MAI-1-preview models mark its first homegrown AI, signaling a shift that could throw yet another wrench into the AI world’s most-watched partnership.

Reminder: Our next live workshop is today at 4 PM EST with The Rundown’s AI Educator, Nate Grahek — join and learn all the latest tips and tricks for getting the most out of ChatGPT. RSVP here.

In today’s AI rundown:

  • Microsoft releases homegrown AI

  • OpenAI’s gpt-realtime for voice agents

  • Create an AI agent to handle email support

  • Cohere’s SOTA enterprise translation model

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

MICROSOFT

Image source: Microsoft

The Rundown: Microsoft just introduced MAI-Voice-1 and MAI-1-preview, marking its first fully in-house AI models and coming after years of relying on OpenAI's technology in a turbulent partnership.

The details:

  • MAI-Voice-1 is a speech generation model capable of generating a minute of speech in under a second, already integrated into Copilot Daily and Podcasts.

  • MAI-1-preview is a text-based model trained on a fraction of the GPUs of rivals, specializing in instruction following and everyday queries.

  • CEO Mustafa Suleyman said MAI-1 is “up there with some of the best models in the world”, though benchmarks have yet to be publicly released.

  • The text model is currently being tested on LM Arena and via API, with Microsoft saying it will roll out in “certain text use cases” in the coming weeks.

Why it matters: Microsoft's shift toward building in-house models introduces a new dynamic to its OAI partnership, also positioning it to better control its own AI destiny. While we await benchmarks and more real-world testing for a better understanding, the tech giant looks ready to pave its own path instead of being viewed as OAI’s sidekick.

TOGETHER WITH AUGMENT CODE

The Rundown: Augment Code is bringing the power of its AI coding agent and context engine right to your terminal with Auggie CLI, now generally available.

From standalone terminal sessions to every piece of your dev stack, with Auggie CLI, you can:

  • Build features and debug issues

  • Get instant feedback suggestions for your PRs and builds

  • Triage customer issues and alerts from your observability stack

  • Build with the AI coding platform that gets you, your team, and your code

OPENAI

Image source: OpenAI

The Rundown: OpenAI moved its Realtime API out of beta, also introducing a new gpt-realtime speech-to-speech model and new developer tools like image input and Model Context Protocol server integrations.

The details:

  • gpt-realtime features nuanced abilities like detecting nonverbal cues and switching languages while keeping a naturally flowing conversation.

  • The model achieves 82.8% accuracy on audio reasoning benchmarks, a massive increase over the 65.6% score from its predecessor.

  • OpenAI also added MCP support, allowing voice agents to connect with external data sources and tools without custom integrations.

  • gpt-realtime can also handle image inputs like photos or screenshots, giving the voice agent the ability to reason on visuals alongside the conversation.

Why it matters: The mainstream adoption of voice agents feels like an inevitability, and OpenAI’s additions of upgraded human conversational abilities and integrations like MCP and image understanding bring even more functionality for enterprises and devs to plug directly into customer support channels or customized voice applications.

AI TRAINING

The Rundown: In this tutorial, you will learn how to build an AI agent that automatically triages incoming emails, tags team members in Slack, and drafts professional responses, turning your overwhelming inbox into an organized workflow.

Step-by-step:

  1. Go to Zapier Agents, click "New Agent", name it "Email Triage Assistant", and set it to run daily at 9 AM (batch processing saves Zapier calls)

  2. Click Copilot and paste: "Every day at 9 AM PST, retrieve all emails from the last 24 hours. Classify as: Spam, Auto-replies, PR/Marketing, Customer Support, Feedback, or General Inquiry"

  3. Add team tagging rules customized for your team members to funnel to specific departments or responsibilities

  4. Click "Add tools" and connect Gmail, Slack, and your FAQ URLs — grant full permissions for autonomous operation

  5. Test with your current inbox, verify categorization accuracy, then enable the daily schedule

Pro tip: Feed your agent FAQ URLs, Notion docs, and previous support threads in the instructions. The more context you provide, the better it handles edge cases and knows exactly who to loop in.

PRESENTED BY STACK AI

The Rundown: Deploy 10 AI agents that actually drive ROI on StackAI—the secure enterprise AI toolkit trusted by finance, legal, ops, & IT teams who move 80% faster than the rest.

With StackAI’s toolkit, you’ll get:

  • Drag and drop platform + ship as chatbots, forms, apps

  • Built-in PII protections, guardrails, audit trails, SSO, and compliance

  • Seamless integrations with 100+ tools you already use

COHERE

Image source: Midjourney

The Rundown: Cohere introduced Command AI Translate, a new enterprise model that claims top scores on key translation benchmarks while allowing for deep customization and secure, private deployment options.

The details:

  • Command A Translate outperforms rivals like GPT-5, DeepSeek-V3, and Google Translate on key benchmarks across 23 major business languages.

  • The model also features an optional ‘Deep Translation’ agentic workflow that double-checks complex and high-stakes content, boosting performance.

  • Cohere offers customization for industry-specific terms, letting pharmaceutical companies teach their drug names or banks add their financial terminology.

  • Companies can also install it on their own servers, keeping contracts, medical records, and confidential emails completely offline and secure.

Why it matters: Security has been one of the biggest issues for companies wanting to leverage AI tools, and global enterprises face a choice of uploading sensitive documents to the cloud or paying for time-consuming human translators. Cohere’s model gives businesses customizable translation in-house without data privacy risks.

QUICK HITS

  • 🎥 Google Vids - Create and edit videos with AI-powered tools

  • 🔊 MAI-Voice-1 - Microsoft’s new in-house voice generation model

  • 🗣️ gpt-realtime - OpenAI’s new advanced speech-to-speech model

  • 🥁 HunyuanVideo-Foley - Open-source model for professional-grade audio

Free Event: The Future of AI Agents in Coding with Guy Gur-Ari & Igor Ostrovsky, co-founders of Augment Code. Ask them anything today in r/webdev.*

xAI released Grok Code Fast 1, a new advanced coding model (previously launched under the codename sonic) that features very low costs for agentic coding tasks.

Anthropic published a new threat report revealing that cybercriminals exploited its Claude Code platform to automate a multi-million dollar extortion scheme.

OpenAI rolled out new features for its Codex software development tool, including an extension to run in IDEs, code reviews, CLI agentic upgrades, and more.

Krea introduced a waitlist for a new Realtime Video feature, enabling users to create and edit video using canvas painting, text, or live webcam feeds with consistency.

Tencent open-sourced HunyuanVideo-Foley, a new model that creates professional-grade soundtracks and effects with SOTA audio-visual synchronization.

TIME Magazine released its 2025 TIME100 AI list, featuring many of the top CEOs, researchers, and thought leaders across the industry.

*Sponsored Listing

COMMUNITY

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Scott M. in Franklin, TN:

"My client was using a legacy version of QuickBooks Desktop, which lacked the feature for sending automated follow-up emails for overdue invoices. To address this, I built a custom automation using Zapier AI: the workflow logs into the accounting email, IDs invoices that are more than 60 days past due, and follows the invoice link to verify whether it's been paid. If payment has not been made, the automation sends a reminder email stating that the invoice is late and includes the original payment link. Every communication includes the accounting department, ensuring they stay informed about delinquent payments."

How do you use AI? Tell us here.

That's it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer—the humans behind The Rundown

Reply

or to participate.