• Everyday AI
  • Posts
  • Ep 621: Microsoft’s new Agent mode: Microsoft marketing or Future of Work in Windows?

Ep 621: Microsoft’s new Agent mode: Microsoft marketing or Future of Work in Windows?

OpenAI launches Sora 2, Microsoft's Agent Mode deep dive, Claude 4.5 Sonnet lands, ChatGPT adds instant shopping and more.

Sup y’all! 👋

Wild day of AI news. Who saw Sora 2 coming? Anywhos, I’m planning my Wednesday show for ‘Putting AI to Work’ on Wednesdays.

What do you wanna learn more about tomorrow?

AI in Google Sheets — This new function is a powerhouse, and we touched on it briefly here. Worth digging in deeper?

Agent Mode in Excel — This is the only new Microsoft agent I’ve got access to, but I’ve been pretty impressed by it. Should we get dorky?

ChatGPT Pulse — This is for Pro users only now, but this ‘daily AI digest’ could be the future of proactive AI assistants and personalized ads. Wanna see how it works?

Sora 2 — IF we get access in time. Could Veo 3 have a legit contender in the AI video space?

What do you wanna learn more about tomorrow for 'AI at Work on Wednesdays'?

🗳️ Vote to see live results 🗳️

Login or Subscribe to participate in polls.

✌️

Jordan

(Let’s connect on LinkedIn. Just tell me you’re from the newsletter!)

Outsmart The Future

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: Microsoft went full agentic with its updates yesterday. We gave you the 101. Give it a watch/read/listen.

🕵️‍♂️ Fresh Finds: NotebookLM and Nano Banana collab, Meta’s new $14 billion bet, JPMorgan Chase’s blueprint to be AI-enabled. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: OpenAI reportedly launching Sora 2 AND social network, Claude 4.5 Sonnet lands, ChatGPT adds instant shopping and more. Read on for Byte Sized News.

🧠 AI News That Matters: Will Microsoft’s Agent mode change the Windows works? Or is it just shiny marketing?  Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about Microsoft releases Copilot Agent Mode, Lovable’s big AI and Cloud vibe code launch, why SAP says AI will shrink head count and more! Check it here!

Ep 621: Microsoft’s new Agent mode: Microsoft marketing or Future of Work in Windows? 🪟

Is this the AI agent we've all been waiting for? 🤔

Maybe

Capabilities? Through the roof. 

Execution, rollout and availability? ummmm......

Join us as we cut through the fluff on these new AI agents from Microsoft and separate the game-changing features from the shiny marketing. 

Also on the pod today:

Copilot gets true multi-step orchestration 🔄
Office agent: personal plans only?! 🚫 
Claude Sonnet models power Office Agent 🧠

It’ll be worth your 40 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Floutwork is a new browser for work with built in AI, CrePal helps you create video shorts with AI, Dex is an AI-powered recruiter to help you find the best candidate.

AI and Safety — Amazon is using doorbells and AI to fight crime.

NotebookLM Updates — NotebookLM may be bringing customizable Infographics powered by Nano Banana.

AI Partnerships — CoreWeave and Meta get official with a $14 billion computing power deal.

AI in Banking — JPMorgan Chase wants to be the first fully AI-enabled bank. Here’s their blueprint.

AI Startups — Members from OpenAI, Meta and Google teamed up to build new AI models.

AI Models — Anthropic’s new Claude model can reportedly work for 30 hours on its own. Yes…. hours.

AI and SecurityGoogle’s launched an AI ransomeware defense, but it has its limits.

 

1. OpenAI releases Sora 2 and iOS app in surprise announcement

OpenAI has launched a new video model with matching dialogue called Sora 2. They launched it via an invite-only Sora app that lets users create and share 10-second AI videos featuring authenticated “cameos” of themselves and friends, powered by the new Sora 2 model with synchronized dialogue and improved physics.

Access will prioritize heavy Sora users and Pro subscribers, with Plus, Team and eventually free users to follow, and creations carry visible watermarks and digital credentials while copyright policing is left to rights holders. This move puts OpenAI in direct competition with Meta’s recently released Vibes app and signals a broader shift to social, remixable AI creativity instead of passive feeds, potentially giving creators and startups a faster way to prototype branded content and social campaigns without a production team.

2. Claude Sonnet 4.5 lands with major coding and agent upgrades 💻

Anthropic released Claude Sonnet 4.5, adding developer-ready features like checkpoints, code execution, file creation, and a refreshed terminal—plus longer-running, more capable agents via the Claude API and Agent SDK.

The company says the model is “most aligned” to date, with improvements against sycophancy, deception, and prompt injection; OSWorld scores jumped to 61.4% from 42.2% in four months, signaling real gains in practical computer tasks. For teams, this could mean 30+ hours of autonomous coding that keeps coherence across large codebases, freeing engineers to focus on architecture and shipping product faster.

3. ChatGPT rocks ecommerce by adding “Instant Checkout” with Stripe 🛍️

OpenAI is rolling out a new Instant Checkout in ChatGPT for U.S. users, starting today with Etsy sellers and adding over a million Shopify merchants “soon.” The feature lets you browse and buy without leaving the chat, powered by an open‑sourced Agentic Commerce Protocol co-developed with Stripe (built to work across AI platforms and payment processors, and supporting Anthropic’s MCP).

OpenAI says product picks are organic and unsponsored, while merchant ranking factors include availability, price, quality, primary seller status, and whether Instant Checkout is enabled—OpenAI takes a small merchant fee per purchase. If this sticks, it could reshape e‑commerce and chip away at Google’s referral model

4. Google rolls out visual-first AI Mode in Search 🖼️

Google is releasing a major update to AI Mode in Search is launching in English in the U.S. this week, bringing conversational, visual-first results and shoppable experiences powered by Gemini 2.5 and the Shopping Graph. Users can describe goals in plain language (“barrel jeans that aren’t too baggy”) or upload an image, then refine naturally with follow-up prompts—no fiddly filters required.

The system’s new “visual search fan-out” analyzes subtle image details and runs multiple queries behind the scenes to surface richer, more relevant results, with links out to retailers and sites.

5. California sets first-in-nation AI safety transparency law 🏛️

California Gov. Gavin Newsom just signed SB 53, the Transparency in Frontier AI Act, marking the first U.S. law to require top AI companies to disclose safety practices and report serious AI incidents.

With 32 of the world’s top 50 AI firms based in California, the move could shape global norms and push companies toward clearer risk reporting while protecting whistleblowers. Anthropic backed the bill; Meta and industry groups warned about innovation headwinds, signaling a tug-of-war as Congress considers a parallel federal evaluation program proposed by Sens. Hawley and Blumenthal.

🦾How You Can Leverage:

Then kinda locked out orgs that might need it most. 

In short: after two-plus year of kinda lackluster Copilot offerings, we now have Microsoft agents that can build PowerPoints based on your data and an agent that speaks Excel. 

Microsoft’s new Agent mode and Office Agent just dropped with autonomous multi-step orchestration that actually works. 

Except…… the rollout is confusing, convoluted and kinda head-scratching. 

So on today's show, we dug into whether these agents actually transform work or just prove Microsoft still can't ship without confusing everyone.

Spoiler alert: if you have a bit of patience and the right access, Microsoft’s newly released agents are legit packed with power. 

Let’s get it. 

1 – The Agent access rollout is a head scratcher 🗿

We've talked to hundreds of companies paying for Copilot licenses nobody can figure out.

Fortune 500s with full Microsoft 365 are using ChatGPT instead because Copilot's permissions are too confusing.Now Microsoft just made it a bit more confusing.

Office Agent (for now) only works for non-enterprise subscribers even though it's literally called office agent.

Three agentic features, three totally different access methods, zero clarity on who gets what on the surface without reading the dang fine print.

The tech actually works. Autonomous multi-step orchestration is fundamentally better than old single-turn Copilot.

But if teams can't find features or understand what license unlocks what, the technology doesn't matter.

You get one launch on something as big as Agent Mode.

Microsoft kinda fumbled it on staggered rollout that guarantees Copilot confusion.

Try This:

Here’s the tl;dr on the new agent modes, how to access them, and who they’re available to. 

Agent Mode (In-App Iteration in Excel and Word)

  • Access: Available to users with Microsoft 365 Copilot licenses (Enterprise) or Microsoft 365 Personal/Family subscriptions

  • Program: Requires participation in the Frontier early access program

  • Platform: Currently available on web versions of Excel and Word only; desktop support is "coming soon"

  • Setup: Agent Mode in Excel requires installing the Excel Labs add-in

  • Geographic: Global availability

Office Agent (Chat-First Document Creation in Copilot Chat)

  • Access: Limited to Microsoft 365 Personal/Family subscribers only initially (Enterprise access coming later)

  • Platform: Available via web-based Copilot Chat

  • Restrictions: Available in the U.S. and English only at launch

2 – Excel's agent mode handles real work 📊

Agent mode in Excel doesn't collapse under pressure like old Copilot.

We threw 50,000 rows at it. GPT-5, Claude 4.5, and Gemini all choked while Excel's agent processed everything.

Agent Mode in Excel uses OpenAI’s latest models to think through complex data requests step by step. It then plans analysis, executes formulas, validates results, fixes errors automatically, then iterates until everything works.

Before you'd prompt Copilot, get mediocre results, start over from scratch.

Now just say "recalculate using 15% growth" and it adjusts without rebuilding.

Also it randomly merges cells and takes five minutes to think through complex tasks. Still beats Googling Excel formulas while your soul dies though TBH.

Try This: 

Open Excel on web, add Excel Labs from add-ins toolbar, launch agent mode.

Drop your messiest sales data and say "build revenue analysis by region with predictive charts and don't merge cells."

That last part matters because the bug is real.

Tell it what to change when something's off instead of re-prompting everything.

Your 30 minute spreadsheet nightmare becomes five minutes of strategic tweaking.

3 – Office Agent: What the enterprise really wants (but can’t yet have) 🎯

Ask for a competitive analysis presentation. The Office agent asks clarifying questions with clickable checkboxes instead of long prompts. Choose one of five focuses, pick charts or text, and set slide count from brief to comprehensive.

It then runs multi-pass web research using Anthropic’s Claude Sonnet 4 and Opus 4.1, sources images, plans slides, and shows live previews you can edit with natural language while it builds. Tell it “remove the image from slide six and turn those bullets into a diagram” and it updates without restarting.

It automates consulting basics: research, clarify, and build a polished deck in five minutes instead of hours.

(Disclaimer…. human in the loop yo! Don’t just send that to your boss right away.)

Try this:

If you have a personal or family Microsoft 365 Copilot license in the U.S., open Copilot chat, pick Office agent, and ask it to “build a 15-slide competitor analysis deck comparing us to the three biggest players with feature breakdowns and market positioning backed by current research.”

Use the checkboxes, watch multi-pass crawls and live previews, and make in-line edits as it works.

Yeah. This is what we wanted all along BIG AI!

Reply

or to participate.