OpenAI released a video that is quite possibly either a clever ruse or a company at the height of its arrogance.
In the video, a woman uses what OpenAI calls an Operator to book a trip.
She starts by picking her travel website of choice, Priceline. Then, she inputs a text parameter for her booking, such as price (refundable), food (free breakfast), and preferred location (NYC).
The AI Agent – sorry, Operator – “books the trip for her.”
OpenAI calls its tool “Operator” and only once references the term AI agent in a blog post from January 5, 2025. However, that blog also mentions AGI and superintelligence and makes other bold claims, so we won’t give them the benefit of the doubt.
OpenAI and everyone else is selling the idea that this is almost an autonomous algorithm (AI agent) that will transform the world by replacing human work.
Except, that is not true. As you can see in the video, the woman must still confirm everything: Her account, payment method, and final confirmation email. She will be responsible for any issues, such as cancellations.
She would have accomplished her task in less time by going directly to Priceline.com and using their user-friendly UX.
It gets better. Watch Sam Altman and his team fumble through ordering food. It really is garbage:
It can not book a trip or order food, so it looks like a worse and more convoluted version of Google or Bing.
This is just part of an AI bubble set to burst, right?
Hints at the World of Tomorrow
AI investments are likely in a bubble, but that does not spell doom for AI overall.
AI agents will shape our lives at some point in the distant future. But we should not worry about that – it may not happen in our lifetime.
We need to focus on what exists today. It is both incredibly exciting and also full of bullshit.
OpenAI is demonstrating the same rule-based automation we have used for over a decade, but with an added intermediary step that utilizes an LLM to handle unknown variables, such as text, voice, or image inputs.
Watch the following video to understand what is going on.
(I have zero affiliation with this person, but they are one of a thousand people all claiming to be utilizing AI agents)
In the video, David Ondre uses a substandard automation interface to create a workflow that accepts voice message input. That input generates a response to his email. In a later step, David instructs an LLM to read and summarize the emails in his inbox, check for conflicts with his calendar, and help him prioritize what he needs to respond to.
He is creating rules and parameters for the AI to work, but the LLM is “breaking free” and is being entrusted to make decisions, freeing up the human processes to be less involved.
It is cool, but it is also a problem.
The first problem is that “AI Agent” is the big buzzword of 2025, seeking to mislead or take advantage of job seekers, investors, and people concerned about losing their jobs.
The second is that people should absolutely be excited about and prepared to utilize tools similar to those shown in the video above.
Let’s take a second to differentiate what we are talking about:
- Automation with LLM (where we are today): A vending machine that dispenses drinks but can also understand you when you say, “This is diet, and I wanted regular.” The vending machine understands, issues a refund, and notifies the owner to put the cans in the correct location. Human involvement is very high.
- AI agent (what will happen far into the future): A robot evaluates local retail opportunities, opens a soda stand, greets customers based on account status or other identifiable attributes, handles reorders and accounting, manages VIP points via an app, calls maintenance, and adapts price and inventory based on factors such as demand to maximize profit and reduce waste. Human involvement is minimal or non-existent.
A vending machine that issues refunds is neat but not earth-shattering. Not being earth-shattering is a good thing. That shows us that AI can improve our world while being mundane and overlooked, just like vending machines.
The next five years of technology will be like this. People are figuring out how to integrate AI, which goes entirely unnoticed by 99% of people.
That is what will begin to change everything.
Eliminate Priceline and Change the World
Remember the Priceline lady?
For AI to get me the best price on a flight, car, and hotel, the tool has to navigate through Priceline and parse all of its data to retrieve the request.
Utilizing automation tied to an LLM, an AI tries to understand the intent of a request and uses its relative freedom to complete the task.
What will change is how American Airlines, Avis, and Marriott structure their data to work natively alongside tools like ChatGPT or whatever new AI product enters the market.
This will look different because when you ask the tool to help you book your trip, the tool will interact with those airlines, hotels, rental car companies, and local tourist spots directly through data, not an interface made for people.
That data will then flow into your AI tool, which will generate a UI with travel availability, price comparisons, promotions, and other perks that would persuade you to make a purchase.
When this happens, Priceline and Expedia are gone. Similarly, any website that aggregates data or acts as a middleman for a product, such as Cars.com, Fiverr, Zillow, or other middleman aggregation tools, is gone.
Apply the same thinking to buying a car online. You will not navigate to Cars.com.
If you are a freelancer, you will tell your AI tool about your skills, and eventually, that data will flow directly to a hiring manager’s AI tool.
You no longer navigate to websites, create accounts, check emails, or log into your bank. Instead, all interactions occur inside your AI tool and will be seamless.
A Mundane Future Rich With AI
To add color and perspective, let’s invent future situations to demonstrate how a seamless AI experience will solve problems.
Alexandra Needs a Vacation
Alexandra finished her two-hour presentation during her company’s quarterly planning session. Due to her long hours, she feels increasingly overworked and misses her daughter’s Thanksgiving play. Next year, a senior executive position opens, and she is being fast-tracked for the role. She knows her family is important, but her job provides everything for them.
At the end of the month, she has a window of opportunity to take a once-in-a-lifetime vacation, but her already limited time makes planning impossible. She mentions this to her secretary, who recommends an AI tool she has been using to simplify office management. Alexandra downloads the tool on her phone, syncs her calendar and contacts, and gives it access to her credit cards, email and communication tools, and social media.
She tells the tool what week to take off work and suggests a quiet, warm, and child-friendly trip. With so much data access, the AI tool takes the suggestions and knowledge of every online movement, her salary, and her travel points to provide a series of vacation options. Alexandra makes a few changes and then books a trip and excursions with a single press of the screen.
A notification is sent to her family and her daughter’s school. A PTO request goes to her boss. Travel points are deducted from her credit card.
During the trip, a client can not find an important financial document. She has no cell signal and does not receive the email. The tool notifies the client of the connection issue and then sends over a file on her behalf. The client confirms this is what he needed. When Alexandra returns to the hotel, the AI tool has summarized the situation. Comfortable knowing the AI tool is competent, she returns to enjoying her trip and does not think about work.
David is Stuck in Traffic
David works a flexible job to pick up his kids from work while his wife, a nurse, works another long shift at the hospital. A mile from his exit, a semi overturns on the highway. He is trapped. David turns to his phone and says, “Hey, Google, I am stuck in traffic on I-60 North before Exit 14A and need someone to pick up my kids.”
Google sifts his contact list, referencing GPS data and familiarity with each individual (via social media or frequency of interaction). Then, the AI tool sends a text to several people who can get to the school and avoid the interstate. Amber responds first. David thanks her. Google sends a follow-up message to everyone else, telling them to ignore the request.
When David arrives home, his kids are already eating pizza. Google knows nobody is home to cook dinner, so it orders food. A local pizza shop is sympathetic and throws in a free brownie. David is thrilled and is now a loyal customer.
Loral Donates During Christmas
Loral is at the grocery store and realizes she forgot her grocery list. She remembers the AI tool her daughter put on her phone and asks it to plan five days of dinners. She explains, “We need leftovers some nights to save money. Our budget is $200 a week. Also, my son needs sports drinks, and Abby has developed a mild intolerance to gluten.
The AI app reads several recipe databases and populates her screen with images and recipes, video instructions, and a list of ingredients. The AI even plans ahead by purchasing some things in bulk to be reused in next week’s menu to save money. Loral accepts the menu, and the AI organizes her shopping list by telling her exactly where each item is in the grocery store.
Several months later, around Christmas, Loral is in line to check out. She overhears a woman explaining that their local church does not have enough Christmas gifts for all of the children. Loral offers to help and asks the woman what else she needs. On the list are a doll, clothes, and a tricycle. The AI tool notifies Loral that it overheard the conversation and has already found the requested items at their lowest price. They can be shipped directly to the organization’s doorstep by tomorrow. Since using the AI tool to shop, Loral has come under budget each week, leaving enough extra money to pay for the Christmas gifts.
Complexity Begets Complexity
The previous examples are possible only after humans design and implement data standards and systems that integrate into AI tools.
Right now, we have relatively simple automation that matches an LLM. Over time, more complex automation combined with task-specific multimodal models will allow AI tools to take on more tasks autonomously with high accuracy. To the point that humans will say, “Why did I ever need to do that?”
This could be a privacy nightmare, for sure. However, convenience will be a factor in selling these tools.
We will need many people to build these systems, including researchers, designers, programmers, marketers, salespeople, analysts, and so forth.
This is all very new. Many companies will burn down, and new opportunities will arise from their ashes.
That work will be done alongside those AI tools. Things will happen much more quickly than ever before. AI will infiltrate almost every part of our lives, and we will not even notice.
In the future, truly autonomous AI will emerge, running its own vending machine business. By then, it is unlikely anyone alive will care too much. Society and our economic structures will have changed so much that it will be welcomed.
These things are frightening today because we do not feel prepared, and we aren’t. Not that there is any need to worry. We do not yet have AI agents. Just mediocre automation with some AI workflows tacked on.
That mediocrity is wonderfully exciting and will change how we all live in less than a decade, and you will not notice a thing.
