AI-generated illustration

AI Agents are Here: Personal Helpers, Perfect Productivity?

Next-generation AI systems promise to get the job done for you. How powerful are AI agents really? And where can they bring the biggest value?

Karsten Lemm

July 23, 2025

AI | Business | Leadership | Work

Wouldn’t it be nice to have a personal assistant? Someone who knows you well, understands what you want to achieve – and then just goes to work, finishing the job on your behalf?

That’s exactly what AI agents promise to do, far surpassing chatbots. Generative AI systems like ChatGPT, Claude or Gemini may seem surprisingly powerful already. But they are “fundamentally passive”, the authors of a new McKinsey report on “agentic AI” point out. “They do not act unless prompted and cannot independently drive workflows or make decisions without human initiation.”

What is it, then, that makes AI agents so much more capable? Do they deserve to be the new stars of the tech scene? And what can you reasonably expect from them?

The Making of an Agent

If you’re old enough to remember Clippy, Microsoft’s initial attempt to build a digital assistant for Microsoft Office users, back in 1995, you may wonder: “Why will people use agents?”, as even Bill Gates admits. But these new systems, he argues, will be “dramatically better”, much smarter, and able to understand their users better. “Clippy has as much in common with agents as a rotary phone has with a mobile device”, Gates writes.

What makes AI agents so different, and potentially extremely powerful, is their ability to take action – autonomously, without human input. “They can do more than just talk to you, they can complete tasks”, Llion Jones, founder of Sakana AI, explained at the DLD Munich 2025 conference. That’s literally what “agentic” means, he added. “You have agency in the world. You’re able to move things around.”

The Age of AI Agents Get insights from Llion Jones (Sakana AI), Nancy Xu (Moonhub AI), Andreas Hufenstuhl (PwC), and Azeem Azhar (Exponential View).

In a business context, the result is “a little bit like having a digital coworker or a digital employee”, suggests Nancy Xu, founder and CEO of Moonhub AI. The former Stanford researcher sees smart algorithms as a way to automate drudge work and free humans from repetitive tasks that are better left to machines.

“Most of you probably don’t like a couple of aspects of your job, and that’s where I think the opportunity for AI agents really will be coming”, she says. “It’s not only better, faster, cheaper, but the ability to help individuals become more creative in their day-to-day workflows.”

This is in line with findings of a recent Stanford study, co-authored by Erik Brynjolfsson, that showed “positive attitudes towards AI agent automation on certain occupational tasks”, particularly when they promise to free up time for more rewarding work. “The overall pattern suggests that AI agents could play a supportive role, enabling workers to offload low-value or burdensome tasks, rather than serving as replacements”, the authors write.

DLD Future Hub: Focus on AI

Two revolutions are happening at once: Business is becoming more sustainable, and the digital transformation is increasingly driven by AI. From September 9-11, 2025, the DLD Future Hub conference will bring together international experts across industries and disciplines to discuss pioneering solutions in the circular economy and what’s next in artificial intelligence. Join us for this unique event, co-hosted by BMW Group and the Bavarian AI network BAIOSPHERE!

Conference Details

Newsletter

From Words to Action

From a technical perspective, AI agents are based on Large Language Models, or LLMs, that also power chatbots. The big difference lies in their ability to understand context, analyze their work, and even learn from their own mistakes. “Within multi-step reasoning, LLMs enable agentic AI to break down complex problems into sequential tasks”, a UC Berkeley article explains. “This dynamic, context-aware approach allows agentic AI to adjust its reasoning as new data emerges, producing coherent, logical solutions to multi-stage problems.”

With agents, “the AI itself becomes ever deeper integrated in our everyday life”, says Björn Ommer, a leading researcher in the field. “It understands even better what we demand from it, what we need, what we don’t have so far. And with that it can provide us better services.”

Building a New Foundation Model LMU professor Björn Ommer highlights the newest developments in AI research.

Newer AI models go through “an internal monologue“, OpenAI’s Colin Jarvis explains. “They might come up with a hypothesis, test it, discard it, and then come up with a new hypothesis.”

Another important advancement is the ability to handle different kinds of data, such as audio, video and text, at the same time – and then draw conclusions from the combined information, an ability that AI researchers call multimodality. With models like OpenAI’s GPT-4, “the cool thing that you can now do is share your screen”, Jarvis says, and have the AI “interact with everything that you can see.”

Redefining Intelligence OpenAI’s Colin Jarvis explains why 2025 is shaping up to be the year of AI agents.

A Productivity Booster?

The big promise of AI agents is that they will make humans more productive. While chatbots could be seen as a “sort of auto-complete”, the newer, more advanced AI systems promise to do “work that may have taken you days, and you can do it within hours”, You.com founder Richard Socher told participants of a recent Udacity webinar. (The online learning platform was launched at DLD in 2009 by AI researcher Sebastian Thrun, a Stanford University professor at the time.)

Any task that comes down to a “sequence of clicks and actions” could at some point become automated through AI agents, Socher suggests. “Essentially whatever workflow you have, you can now explain that to an AI, create an agent for it, and then it does it for you”, he told journalist Jochen Wegner at DLD25.

Researching, summarizing, analyzing data – these are tasks where artificial intelligence can shine. “Our agents are not there to book your flight”, Socher emphasizes. “Our agents are there to help you with knowledge work.”

This makes it possible to automate a variety of tasks: “Part of your work for engineers, for sales, for service, for HR, recruiting, marketing; any kind of work where you need to understand both internal data as well as anything on the web – agents will automate these workflows.”

Answers. Agents. AGI Take a deep dive into the state of artificial intelligence with You.com founder Richard Socher in conversation with Jochen Wegner.

McKinsey put this into practice: The consultancy built an agent that automates the creation of a briefing deck on people participating in a meeting. “It’s fabulous. It gives you all the information the team would have pulled [together manually]”, McKinsey Digital’s Rodney Zemmel told the DLD Munich audience. “But it also has some interesting quirks”, he admitted, cautioning that AI agents require human oversight. “Used badly”, Zemmel said, they could be “the equivalent of the intern who drank too much the night before.” (Zemmel recently left McKinsey to join Blackstone.)

Rewiring for Impact Watch Rodney Zemmel discuss AI as a “tech transformation engine” with Axios reporter Ina Fried.

Business Impact

Surveys support the notion that agentic AI can boost the bottom line. “Of those adopting AI agents, two-thirds (66%) say that they’re delivering measurable value through increased productivity”, PwC reports.

But merely investing in technology is not enough, the consultancy warns. For best results, companies should move fast, educate their workforce, and “fundamentally rethink operating models” and processes around AI agents.

McKinsey similarly advocates moving “from scattered initiatives to strategic programs”. Currently, too many companies focus on pilot projects instead of implementing AI wherever possible, the authors of the report Seizing the Agentic AI Advantage write. “Unlocking the full potential of agentic AI requires more than plugging agents into existing workflows. It calls for reimagining those workflows from the ground up – with agents at the core.”

Illustration from McKinsey report “Seizing the Agentic AI Advantage” showing how a bank improved its productivity by 60% using AI agents.

Smart move: A retail bank saw a 20% to 60% productivity gain by using AI agents to “reinvent the process of creating credit-risk memos”, McKinsey reports.

Hubert Burda Media, one of Germany’s largest media corporations (and DLD’s parent company), went all-in by building its own AI platform, called AISSIST – and is seeing extremely positive results, according to Eli Varn, Co-CEO of BurdaVerlag, the company’s publishing unit.

“Currently we have about 1,800 of our employees working on this platform, and it ranges from corporate functions all the way to our editorial teams”, Varn says. “In terms of business impact, what makes a real difference is that it is deeply integrated into our production processes.”

For its popular children’s brand Lissy PONY, Burda partnered with Black Forest Labs to speed up the process of creating images with the help of generative AI. “What used to be a very lengthy back-and-forth process between our editorial departments and external illustrators now can be done in the time it takes to write a prompt”, Varn told the audience at DLD Munich. “The business impact of using AI is huge for us, and integrating agentic workflows into our everyday business processes is going to be even bigger.”

Agentic AI – What’s Next? Get real-life examples from Hubert Burda Media’s Eli Varn and Sebastian Küpers (Plan.Net Group) in this DLD25 session moderated by Andreas Liebl (appliedAI).

Your New Job: AI Manager

Agents become even more powerful when they connect and communicate with each other, forming a new kind of swarm intelligence: “The big thing”, Sebastian Küpers says, “is the understanding that you can basically build teams of agents with specialized role to work together.”

At his agency, Plan.Net Group, employees use agentic AI for various task, such as compiling insights and managing content, significantly speeding up their work. Importantly, “humans are basically coaching the team of agents”, Küpers notes, “helping them to build up their memory by constantly giving feedback.”

Not only does this cooperation foster acceptance of the new technology, it is crucial to making the systems reliable. “When you take multiple actions, your accuracy has to be insanely high”, Richard Socher explains, because errors quickly add up, rendering the systems unusable.

DeepMind founder and Nobel laureate Demis Hassabis, who first spoke at DLD in 2017, makes a similar point: “If your AI model has a 1% error rate and you plan over 5,000 steps, that 1% compounds like compound interest”, he told attendees of a Google event.

The future of work, at least in offices around the world, may well be humans managing algorithms, directing dozens – or hundreds – of AI agents to complete their tasks in the most efficient way. “Almost all of us who are currently doing individual contributor work are going to become managers”, Socher told the DLD Munich audience. “We will manage our AIs.”

This isn’t necessarily easy, he added. “You have to be very precise about your language and the corner cases of how a job is done correctly, and where it might fail.”

But it also opens up new opportunities for entrepreneurs – and could shake up markets by allowing smaller companies to compete with vastly bigger ones. “You can be a five or fifty-person, multibillion dollar company”, Socher said. “That’s not that crazy anymore if you use agents very well.”

More Expert Insights About AI

“Successful AI Means Responsible AI”

How can artificial intelligence advance humanity and benefit society as a whole? Google executive Ewa Duerr says responsible use of the technology is key.