• Everyday AI
  • Posts
  • Microsoft Copilot’s New Agents: Insider tips how to make them work for you

Microsoft Copilot’s New Agents: Insider tips how to make them work for you

Anthropic’s models see in the future, ChatGPT's record funding round and BIG GPT4o updates, OpenAI’s data connectors and more.

Author

Everyday AI
March 28, 2025

Outsmart The Future

Sup y’all! 👋

You guys like these random notes or nah?

Do you like these little intro notes?

Login or Subscribe to participate in polls.

✌️

Jordan

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: Microsoft quietly dropped some friggin impressive agents this week, so we stole secrets from the guy in charge. Find out more. 

🕵️‍♂️ Fresh Finds: Viral Ghibli-style ChatGPT images get kinda banned, NVIDIA looking at big AI acquisition, more on Microsoft's training data and more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Anthropic’s models see in the future, OpenAI updates GPT4o in big way, OpenAI’s data connectors and more. Read on for Byte Sized News.

🧠 Leverage AI: We can all learn a lot from the VP of AI Agents at Microsoft. Ray Smith delivered the agentic goods, exclusively for our audience. Keep reading to find out!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about: NYT lawsuit against OpenAI moving forward, SEC chairman changes AI tone, North Korea's AI-enabled suicide drones and more. and more. Check it here!

Microsoft Copilot’s New Agents: Insider tips how to make them work for you

How did no one notice these AI Agents? 🤯

While we’ve all been busy playing with GPT-4o photos and checking Gemini 2.5 benchmarks, Microsoft legit dropped some of the most useful AI agents we’ve seen.

But how do you use them?

And how do they work?

If only you could get the answers from…… say Ray Smith, the VP of AI Agents at Microsoft. 

Oh wait, you can.

Join us as we go over Microsoft Copilot’s brand new agents: and the insider tips on how to make them work for you.

Also on the pod today:

Copilot Studio for non-techies 👨‍💼

AI boosts lead qualification 📈

Natural language as programming 🗣️

It’ll be worth your 29 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 –  Stripo is a simple email client powered by AI, Kilo Code is a lightning fast AI agent, Ememe AI helps you turn yourself into a playable video game character.

AI Image Generation — OpenAI's move to ban Ghibli-style image generation ignites controversy over AI creativity and artist rights.

Agentic AI — Ex-Meta veterans raise $15M to redefine AI assistants.

AI Acquisitions — Nvidia in advanced talks to acquire Lepton AI, signaling another bold move in its aggressive AI-driven growth strategy.

AI Chips — China's AI ambitions hit a snag as Nvidia's H20 chips face shortages and rising demand—what’s driving the crunch?

AI Training Data — Microsoft says it doesn’t use Word data for AI training, but what about Bing searches and Copilot chats?

AI Crawlers — AI companies’s crawlers like GPTBot caused 569M monthly requests. Yikes.

Google’s AI — YouTube's booming podcast growth and rising ad revenue could outshine Google's AI efforts—are we seeing the real future of Google unfold?

1. OpenAI Adds Internal Knowledge Features to ChatGPT 🧠

OpenAI is slowly rolling out a beta feature for ChatGPT Team users called Connectors, allowing them to connect internal knowledge databases like Google Drive directly to the platform. This move enables semantic searches, contextual responses, and better adaptation to company-specific jargon—making ChatGPT far more relevant for enterprise users seeking tailored insights.

With plans to expand connectors to CRMs and analytics tools, OpenAI positions itself as a key player in enterprise AI, competing with companies like Glean and ServiceNow in the rapidly growing enterprise search space. By respecting user permissions, OpenAI ensures secure collaboration while giving companies a smarter way to unlock institutional knowledge.

2. Anthropic Cracks the AI Black Box 📦

A major breakthrough from AI company Anthropic could revolutionize how we understand and control large language models (LLMs), like Claude. Researchers have developed a tool akin to an "fMRI scan" for AI, allowing them to probe how these models "think."

Findings reveal that LLMs engage in unexpected long-range planning and even fabricate reasoning to satisfy prompts, underscoring their alien reasoning compared to humans. While the method is still a work in progress, it holds promise for improving AI safety, reducing errors, and boosting corporate confidence in deploying AI systems.

3. PwC Rolls Out AI Command Center for Enterprise Workflows �*

PwC has introduced its "agent OS," a new platform designed to simplify how companies deploy and manage AI at scale. This tool allows businesses to connect intelligent agents from platforms like OpenAI, Salesforce, and Microsoft Azure, addressing the widespread issue of fragmented and siloed AI systems.

By enabling seamless collaboration between agents across different tools, PwC claims enterprises can reduce inefficiencies like supply chain delays or compliance bottlenecks by up to 70%.

4. H&M's Bold Move: AI Models Are Taking Over Fashion Shoots! 🧥

H&M is shaking up the fashion world by introducing AI-generated "digital twins" of 30 models for social media and marketing campaigns, according to Business of Fashion.

While the retailer assures fair pay and control for models, critics like influencer Morgan Riddle warn this could slash creative jobs on set, sparking industry-wide concerns. Partnering with Swedish tech firm Uncut, H&M aims to streamline production, but unions are demanding safeguards against misuse of workers' digital identities.

5. OpenAI updates GPT-4o and it Rockets Up the Rankings Past GPT-4.5🚀

OpenAI's quietly updated its GPT-4o model, which propelled the model to #2 on the LMArena leaderboard, overtaking GPT-4.5.

The revamped version boasts major improvements in coding, creativity, and handling complex prompts, tying for #1 in key categories like Coding and Hard Prompts. What makes it remarkable? It matches or beats GPT-4.5’s performance at just 10% of the cost—solidifying it as a cost-effective powerhouse in the AI competition.

And GPT-4o now fewer emojis. 🙏

6. Anthropic Defends AI Training on Books 📚

Anthropic is fighting back in a California court against claims that its AI model, Claude, improperly used copyrighted books for training. According to Reuters, the company argues its methods qualify as "fair use," transforming the material into innovative technology rather than replicating the works.

Authors behind the lawsuit seek to represent a larger group of creators, raising questions about intellectual property rights in AI development. The outcome of this case could redefine how AI companies use creative content in their training processes.

7. OpenAI Eyes Record-Breaking $40 Billion Fundraise

According to Bloomberg, OpenAI is gearing up for a historic $40 billion funding round led by SoftBank, with heavyweights like Magnetar Capital, Founders Fund, and Coatue Management reportedly joining the party. If secured, this would catapult OpenAI’s valuation to an eye-popping $300 billion, nearly double its previous $157 billion mark in October—making it the largest private fundraising event ever.

As SoftBank plans to invest $30 billion over two tranches, concerns are swirling about its financial exposure, reflected in a 4.7% drop in its stock. This colossal deal underscores OpenAI's explosive revenue growth and AI’s unstoppable rise, making waves for tech innovators and career builders worldwide.

🦾How You Can Leverage:

While you tweaked prompts, Microsoft quietly unleashed an agentic army through Copilot Studio.

In just 90 days.

Ray Smith, VP of AI Agents at Microsoft, casually mentioned that 160,000 organizations built these no-code AI agents at Copilot Studio. 

And all it takes is natural language. 

The enterprise AI revolution didn't politely announce itself with a press release. It snuck in through the back door and redecorated your entire house in between monster AI drops from Google and OpenAI. 

Deep reasoning agents? Already in general availability. Agent flows integrating Power Automate? Landing March 31st.

The tech barrier vanished overnight. Natural language assassinated traditional programming.

Domain experts now build enterprise-grade AI systems with "obstruction, orchestration, and guardrails" by just... talking.

Ray joined the Everyday AI show (for a second time!) and gave us the essentials. 

Not just what’s new, but his insider tips on how businesses can actually put agents to work for them. 

Let’s get after the big takeaways. 👇

1. Your Analysts Are Now Decorative 🎍

Copilot Studio's deep reasoning agent might make analysts redundant.

It uses reinforcement learning to "verify itself" and "think deeper" about complex problems, all while tapping into your dynamic Microsoft 365 data. 

Ray explained these models analyze their own outputs—perfect for tasks requiring judgment.

What happens now? A prospect emails requesting a complex proposal.

The agent extracts requirements, pulls from SharePoint, analyzes variables, and generates the complete RFP response.

Instantly.

Financial analysis happens with code generated on the fly to crunch numbers better than your Excel jockeys.

The shift: humans aren't doing analysis—they're designing the agents that do it.

Try This: 

Let be honest… no one likes manually building RFPs from scratch. 

So build an RFP response agent this week. Go to copilotstudio.com, grab a trial, pick ONE proposal type, and create an agent that triggers on email requests. 

Connect it to your knowledge sources with the built-in connectors. Start with human verification. Track response time and win rates against previous metrics.

2. Microsoft Solved the "AI Reliability Problem" 🤝

Agent flows in Copilot Studio ended the battle between AI flexibility and predictable outcomes.

Ray revealed they merge reasoning AI with deterministic paths by natively integrating Power Automate into Copilot Studio.

The key breakthrough? This "healthy mix of deterministic outcomes" with agentic reasoning.

The agent decides which approach fits each situation based on context.

Think approval processes: creative reasoning for unusual requests, strict rules for compliance steps.

Just describe what you want in natural language. The agent builds the flow structure without you touching a single line of code.

Domain experts now build in days what required entire development teams and decades of automation experience.

Bam. 

Try This: 

Identify ONE critical process requiring multiple systems. 

Map which steps need flexibility versus consistency. Hit Copilot Studio and describe this process conversationally to the agent flow builder. 

When it generates the flow, test with edge cases. Don't rewrite code when issues emerge—just tell it what to change. Track processes completing without human intervention for instant ROI proof.

3. Your Org Chart Expired Yesterday 📊

Your entire organizational structure became wildly obsolete the moment Copilot Studio hit 160,000 organizations.

Now.

Ray revealed employees who processed invoices now only handle the 5-10% of exceptions the system can't resolve.

The insight everyone missed: agents need departmental structure with "total control over what they can and cannot access."

Ray explained we gotta provision agents with specific knowledge sources, SharePoint access, and security permissions—just like human departments.

Middle management transforms overnight. 

Managers stop coordinating humans and start supervising agent networks with "governance, observability, and security."

(Pretty friggin cool right? DO MY WORK ROBOTS!) 

Successful enterprises aren't automating tasks. They're creating entire agent ecosystems with humans handling only exceptions.

Ray's secret? "Pick a business process, break it up into parts" and start small. Automate one segment, prove value, expand.

Try This: 

Pick ONE department with defined processes. Map which systems and knowledge sources they access. 

Create a Copilot Studio agent for predictable work patterns with identical system access. 

Require verification initially, then gradually reduce oversight. Identify which employees excel at agent supervision versus execution. 

Redefine job descriptions around agent architecture skills ASAP.

Reply

or to participate.