OpenAI's “o3-mini” and “deep research” Could Disrupt Knowledge Work Faster Than We Thought

Just days after Chinese AI lab DeepSeek shook markets, OpenAI has responded—big time.

They’ve dropped o3-mini, a brand-new reasoning model that excels at science, coding, math, and more, while allegedly running faster and cheaper than previous generations.

But that’s not all: OpenAI also quietly unveiled a powerful new feature called “deep research,” which transforms ChatGPT into an autonomous research agent for high-level tasks.

And buckle up, because both releases have big implications for knowledge work.

On Episode 134 of The Artificial Intelligence Show, I got the scoop on what to expect from Marketing AI Institute founder and CEO Paul Roetzer.

o3-mini: A Rapid Response to DeepSeek 

DeepSeek’s breakthrough last week triggered an overreaction in the markets, wiping out billions in stock value overnight. Some speculated that its open-weight models spelled trouble for OpenAI and other frontier labs. Now, o3-mini is here—and challenging those assumptions.

o3-mini is designed for STEM tasks (like coding, math, and science) and claims to match or exceed previous models’ performance. It offers responses 24% faster than older models, and—unlike many other releases—OpenAI made it available to free ChatGPT users. Pro subscribers get a “high” variant (o3-mini-high) that produces even better answers by “thinking harder.”

Said Sam Altman about the release:

The only challenge? Navigating the ever-growing list of available models. Roetzer jokes that ChatGPT is “starting to get in its own way” by offering too many versions for non-developers.

“People just want ChatGPT to work,” he says, “not to pick from seven different models.”

Say Hello to “deep research” 

Right on the heels of o3-mini, OpenAI introduced “deep research,” a new feature in ChatGPT that acts as an autonomous research agent. 

If you’re confused by the name, you’re not alone. This is the same name as Google’s Deep Research capability, which also acts as an AI research assistant, just with some different capitalization.

Think of OpenAI's deep research like a specialized research analyst living inside ChatGPT. You ask a complex question or upload documents, and it goes off to study them—sometimes for several minutes—before returning thorough findings.

In the process, deep research displays how it navigates sources, offering a sidebar that cites exactly where it pulls information from, so you can track its entire “thinking” process.

It already supports text, images, and PDFs. And in the near future, OpenAI plans to add visual data and images to the final research outputs.

Right now, deep research is initially limited to ChatGPT Pro subscribers who get 100 queries per month, since it’s computationally expensive. Team and enterprise plans will likely see it later.

A Glimpse Into the Near Future

Deep research has already wildly outperformed other models on “Humanity’s Last Exam,” a benchmark test used to evaluate super-strong AI.

Roetzer tested it by digging into how o3-level models might impact the job market.

“It spent five minutes working on my question,” he says, “and you could watch it going back and forth, reevaluating sources in real-time.”

He was, to put it mildly, impressed. Out of the box, deep research is capable of sophisticated research, reasoning, and analysis that you’d expect from a talented knowledge worker.

“My initial reaction to this was that the AI timeline is accelerating,” he says. “The delta between what these models are capable of, and society's understanding and preparedness grew again last night.”

Sam Altman, CEO of OpenAI, isn’t shy about what’s happening here. He estimates that deep research alone “could do a single-digit percentage of all economically valuable tasks in the world.” Even if it’s only “single digits,” that’s a lot of work. And it’s just the beginning.

Capabilities like the ones now present in deep research could have a widespread disruptive impact on any knowledge worker who deals heavily in text, data, or numbers. (Think anyone from financial analysts to lawyers to marketers.)

Put this together with some other signals we’re seeing, says Roetzer, and you start to get a glimpse into where this is all going…

Startup incubator Y Combinator recently put out a call for AI companies they’d like to fund. These include a heavy dose of companies building AI agents in many different categories and industries. And, when you go down the list of company ideas, a theme becomes clear:

These aren’t startups that build AI to compete with existing software. They’re startups that build AI to fully replace entire categories of knowledge work.

At one point, Y Combinator writes:

“The value prop of B2B SaaS was to make human workers incrementally more efficient. The value prop of vertical AI agents is to automate the work entirely.”

This points to a harsh truth that every knowledge worker needs to start planning for:

There is a massive amount of incentive among many investors to fund AI technology that makes businesses more efficient by reducing the amount of money spent on human labor.

“Try and step back and be realistic about what is happening, because it's going to happen really fast,” says Roetzer.

Calls to build AI that “fully replaces” certain jobs are getting louder—from Y Combinator, from Sam Altman, and from the market itself. That future is coming faster than many realize.

“This is real,” says Roetzer. “The timeline is accelerating.”