After Deep Research comes deep research

In just a few weeks, OpenAI released deep research, Perplexity launched Deep Research, and most recently, Grok introduced DeepSearch. Even Google Gemini launched its own take on Deep Research, called … Deep Research, at the end of last year.

“A lot of companies create their versions [of Deep Research] to follow the broader trend and focus on reasoning models that have taken the world by storm,” observes Kate Soule, a Director of Technical Product Management at IBM Research, in a recent episode of the Mixture of Experts podcast. The DeepSeek moment, in particular, demonstrated how open source can drive innovation. “We’re seeing a rich ecosystem emerge.”

Features like deep research and agentic search are reshaping how people find and consume information online. While traditional search engines rely on indexing and ranking pages, AI-driven agents directly read, synthesize and deliver structured answers.

“Most of these deep research models will start first by creating a plan,” explains Shobhit Varshney, a VP and Senior Partner at IBM.  “Then it fires off and starts to crawl the web … to find all the websites that are relevant.”

In this context, agents aren’t just answering simple questions—they’re searching the web or private data, gathering facts, fact-checking and breaking down problems into smaller components. Agents follow a chain of thought, double-check their work, and produce structured research or analysis.

The objective evaluation of these deep research models’ output remains, however, a limit. “I don’t know what good looks like,” says Varshney. “It's a problem that is an order of magnitude tougher than writing code or solving math, where I can deterministically tell you whether the answer is correct or not,” he notes.

But search is only one of many applications of agentic AI.

“This is just one of those early use cases that we've identified where there's some clear demonstrable value that the reasoning is bringing,” says Soule.

The shift to agentic AI is already transforming industries, forcing businesses to rethink how they operate. “I haven’t met one founder who doesn’t work on this question—and I meet with 500 founders every year,” says Frank Desvignes, a Founding Partner at True Global Ventures in Silicon Valley.

One of the industries already seeing this shift is travel. Sarah Eley, Co-Founder and COO of Bookit N Go, a B2B and B2C travel booking service, says agents can research, offer personalized recommendations and book trips on their users’ behalf. “It seems so obvious to me that this is the logical next step when it comes to travel because it’s so difficult today to plan and book,” she says.

Beyond marketing and customer support, every aspect of business—from the supply chain and finance to human resources and operations—will be transformed by agentic AI.

“Agentic AI is causing a paradigm shift,” says Karen Butner, a Global Research Leader at IBM’s Institute for Business Value.

Instead of people doing the individual tasks that are involved, the agents will be doing the tasks, Butner explains. People will be managing the AI agents to make sure they work together smoothly to achieve business goals.

“These systems don’t just learn—they adapt. I call the people managing them ‘AI conductors’ because they orchestrate everything,” says Butner.

In an agentic world, operations will run 24/7, and humans will be reskilling. AI agents’ value won’t be in operational tasks but in their ability to understand and analyze the environment. The opportunity will be massive.

What can we expect from deep research as it pertains to these agentic systems? It’s one of the more practical use cases for reasoning, says Soule. What’s more, once these reasoning models start layering on enterprise data, the possibilities multiply, Varshney believes: “I think that’s the company that’s going to make billions of dollars.”