• Everyday AI
  • Posts
  • ChatGPT Agent Mode Overview: Real use cases and 3 worthwhile tips

ChatGPT Agent Mode Overview: Real use cases and 3 worthwhile tips

U.S. Gov. unveils AI action plan, Windows 11 adds Copilot Vision update, Amazon shuts down its Chinese AI lab and more!

Author

Everyday AI
July 23, 2025

Outsmart The Future

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: Discover how ChatGPT Agent Mode completes complex tasks, real use cases for business, hands-on demo tips, and why this tool could transform your workflow. Give it a listen.

🕵️‍♂️ Fresh Finds: Google Photos gets new AI futures, YouTube Shorts adds image-to-video feature and a list of news sources that AI chatbots read. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: U.S. Gov. unveils AI action plan, Windows 11 adds Copilot Vision update and Amazon shuts down its Chinese AI lab. For that and more, read on for Byte Sized News.

🧠 Learn & Leveraging AI: Want to make the most out of ChatGPT Agent mode but don’t know where to start? We break down real business use cases. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about OpenAI and UK government join forces, Google eyeing news licensing for AI content, OpenAI and Oracle boosting power for Project Stargate and more. Check it here!

 ChatGPT Agent Mode Overview: Real use cases and 3 worthwhile tips 💡

ChatGPT Agent Mode is here.

If you're wondering what types of use-cases we're using internally and some tips to get you ahead of the curve....we gotchyu.

Make sure to join us as we put AI to Work this Wednesday.

Also on the pod today:

• AI Agents vs. Agentic Workflows 💼
• Pros and Cons of ChatGPT Agent Mode 🤔
• Agent Mode Security, Privacy, and Risks 🔒️

It’ll be worth your 47 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 – Toolable helps you build and share your own AI tools, Hedra takes your AI content game to the next level, Fynix helps you code faster with AI.

Google – Google Photos is adding AI features to ‘remix’ photos in different styles and turn pics into videos.

YouTube – Youtube Shorts is adding new AI features including an image-to-video tool.

AI Tech – Github is launching GitHub Spark, its AI app making tool.

AI in Government – The FDA’s AI tool has been hallucinating according to employees.

AI Search – This new report shows what news sources AI chatbots read.

AI Research - A new study reveals that AI models can pass biases or harmful tendencies to other models through generated data.

1. Trump Unveils AI Action Plan, Promising a Lighter Regulatory Touch 🇺🇸

President Trump’s administration has rolled out a new AI strategy aimed at removing federal hurdles, expanding AI data centers, and pushing U.S.-made AI technology onto the global stage, signaling a sharp shift from Biden-era restrictions.

The plan also targets so-called "woke AI" by proposing federal contracts only for developers who avoid ideological bias—though how that will be defined remains unclear. This move could reshape AI policy and international tech exports while sparking legal debates over government-imposed content standards.

2. Microsoft Unveils New AI Features in Windows 11, Including Copilot Vision 👁️

Microsoft is rolling out a suite of AI-powered tools for Windows 11, with the standout being Copilot Vision, which can scan and analyze everything on your screen, answering questions and assisting with tasks across apps and websites.

Snapdragon-powered Copilot Plus PCs get exclusive perks like an AI agent in Settings that can execute commands for you, plus advanced features like an AI sticker generator in Paint and a “perfect screenshot” tool.

3. Amazon Shuts Down Shanghai AI Lab Amid U.S.-China Tensions ❌

Amazon is closing its Shanghai AI research lab, a move tied to “strategic adjustments amid U.S.-China tensions,” according to an internal post shared by a lab scientist. This follows recent layoffs at Amazon Web Services, highlighting the company’s ongoing retrenchment in China after shuttering its e-commerce and Kindle operations there.

With geopolitical pressures mounting and trade restrictions on AI-related tech tightening, Amazon’s decision underscores the growing challenge for global tech firms navigating the U.S.-China divide.

4. Google DeepMind Launches AI to Decode Ancient Roman Inscriptions 📜

A new AI tool named Aeneas, developed by Google DeepMind in collaboration with historians, is set to revolutionize the study of ancient Roman texts by predicting their origins, dating them within 13 years, and suggesting missing words in damaged inscriptions.

Trained on a vast database of nearly 200,000 inscriptions, Aeneas helps scholars swiftly link related texts and offers deeper historical insights beyond simple word matching.

5. Google’s AI Search Summaries Change How Users Click Links 🔗

A new study from Pew Research Center reveals that Google’s AI-generated search summaries, now shown in about 18% of U.S. searches, are causing users to click on traditional search results far less—only 8% of the time versus 15% without AI summaries.

These concise AI overviews often satisfy users enough to end their browsing sessions early, which could challenge online publishers relying on traffic for revenue. Longer, question-based searches are more likely to trigger these AI summaries, which frequently cite Wikipedia and government sites.

6. Google Doubles Down on AI with Big Bets and Big Revenue 💰

Alphabet’s latest Q2 2025 earnings reveal AI is not just a buzzword but a major profit driver, with CEO Sundar Pichai confirming AI features like Overviews and AI Mode are performing strongly. The company is upping its capital expenditure to $85 billion this year, signaling an aggressive push to keep pace with rivals like OpenAI and Meta in data center expansion.

Despite AI helping users get quick answers, Pew research indicates it may reduce traffic to individual websites, raising questions about its broader impact on the web economy.

🦾How You Can Leverage:

ChatGPT just released real agents that can actually complete projects from start to finish. 

Millions of users and most companies have no idea they're about to get steamrolled.

This isn't another AI tool update.

Nope. 

For the first time ever, a major AI company shipped an actual AI agent that can handle ENTIRE projects from start to finish, not just bite-sized tasks your team can tie together with multiple AI-powered tools and workflows. 

We're talking terminal access, API connections, and the ability to go completely off-script when your instructions suck.

OpenAI is the first. The other big tech players will shortly follow. 

This isn’t just another AI update. This is a line in the sand from human-powered AI work to agents working with human oversight. 

Big different shorties. 

So on today's show, we're breaking down why most teams think they've used agents but haven't, plus the brutal three-run strategy that separates winners from the companies still manually clicking through spreadsheets.

Let’s goooooooooo. 👇

1 – Real Agents vs Expensive Workflows 🤔

Most teams call custom GPTs and Zapier automations "agents."

Wrong.

Those follow predetermined paths like obedient robots.

A true agent makes decisions you never programmed it to make.

Terminal access. API connections. Complete autonomy when your instructions are trash.

OpenAI slapped their first-ever "high risk" label on this mode because it can literally do anything an intern could do, including the dangerous stuff that keeps IT departments awake at night.

The companies winning right now understand that real agents handle ENTIRE projects from research to final presentation while their competitors are still breaking everything into bite-sized AI tasks.

Try This: 

Audit every tool you're calling an "agent" right now.

Can it log into your business systems without predetermined workflows? Can it make decisions you didn't program? If not, you're playing with toys while competitors deploy digital employees that work 24/7.

Stop calling workflows "agents" and start identifying actual repetitive projects that span multiple platforms.

Also, if you missed our first episode covering OpenAI’s Agent Mode, here’s your refresher. 

2 – Four Use Cases Replacing Job Functions 💼

We dished some real business use-cases on today’s show:

Executive dashboards that automatically pull metrics from fifteen different platforms your team manually logs into every week.

Meeting prep automation that researches attendees, companies, recent news, and creates talking points while you sleep.

We discussed how one person's ENTIRE job used to be logging into marketing systems, pulling data, creating spreadsheets, then building presentations.

That job won’t exist in the same way anymore at smart companies.

Vendor research and CRM data enrichment that transforms manual clicking into automated intelligence gathering.

These aren't party tricks. They'll replace current jobs, so plan accordingly.

Try This: 

Pick your team's most brain-numbing weekly task that involves multiple websites and manual clicking. Calculate the hourly cost, multiply by 52 weeks.

That number should terrify you because most executives discover their teams spend 40% of time on work a digital intern could handle better and faster.

3 – The Three-Run Optimized Strategy 🧠

Take over browser control immediately and log into your systems before starting the agent.

Use the full context window by uploading files and connecting Gmail, Google Calendar, Google Drive since this runs inside ChatGPT's entire ecosystem.

Follow the three-run rule: first run means observe without interrupting, second run involves documenting failure patterns, third run is when you start giving corrections.

Most people quit after the first failed run when real learning starts at run three.

Try This: 

Pick your most repetitive weekly task and commit to three runs over three weeks. Document failures in run one, track patterns in run two, only iterate in run three.

Smart teams understand that breakthrough happens exactly one run after most competitors give up and go back to manual clicking.

Reply

or to participate.