In fact, we're not yet ready to face the actual actions of AI agents.

I originally thought I understood the development direction of artificial intelligence. Until I saw what Klarna had achieved with the work of 700 employees. An event that occurred in early 2024 made me rethink all my previous understandings of artificial intelligence.

The Swedish fintech company Klarna deployed an AI assistant to handle customer service. Not to assist customer service, but to truly take charge of it. In the first month, the system handled 2.3 million conversations, equivalent to the workload of 700 full-time customer service representatives.

When I saw that headline, I immediately thought: Here we go again. Another company hyping up AI technology to boost its stock price. Another so-called "revolutionary" chatbot that would collapse as soon as users were dissatisfied.

Then I delved into the actual results.

The problem-solving time was reduced from 11 minutes to less than 2 minutes. The volume of repeat inquiries decreased by 25%. And what really made me stop and read carefully was that the customer satisfaction rating was on par with that of human customer service. Klarna expects this to bring a profit increase of $40 million to the company in 2024.

That's when my thinking changed. This is no longer about technological progress, but a fundamental transformation in the way we work.

The Disturbing Truth I Keep Mentioning

I've been in the tech industry for so long that I've witnessed countless hype cycles destroy once-promising ideas. Blockchain was once highly anticipated to revolutionize everything - now it's a laughingstock. The metaverse was once seen as the future, but now it's a costly mistake for Facebook. As for NFTs... well, we all know how that ended.

So when people started touting "AI agents" as the next big thing, I rolled my eyes. Just another buzzword. Consulting firms use it to charge companies millions of dollars while providing mediocre solutions.

But I can no longer turn a blind eye to what's happening.

Salesforce has just launched their AI agent platform, Agentforce. In just a few months, they've announced amazing results: The average case processing time for Engine has been reduced by 15%; 70% of chat inquiries at 1-800Accountant were automatically resolved during the busiest tax week; Grupo Globo's subscriber retention rate increased by 22%.

These aren't small companies conducting experiments. They're established enterprises with real customers and real stakes. They're not just testing AI agents - they're betting on their operations.

Salesforce announced that it has 12,000 Agentforce customers. Its CEO, Marc Benioff, calls it "the third wave of AI - beyond assisted driving and into a new era of highly accurate, low-interference intelligent agents that will actively drive customer success."

What shocked me the most is that their vision is to empower one billion agents by the end of 2025. One billion autonomous AI systems will be making decisions in thousands of companies.

That number has been haunting me. It both comforts and terrifies me.

How Is It Different from All Other AI Tools?

I need to explain something that took me months to truly understand: AI agents are not chatbots, and they're not even "intelligent assistants."

When most people think of AI, ChatGPT comes to mind. You enter a question, and it answers; you ask it to write something, and it does. You're always in control, giving instructions and prompting each step.

AI agents work completely differently.

They observe their surroundings, make decisions autonomously, and take action. According to Salesforce, an agent can receive a high-level request, then plan to obtain data, analyze it in a spreadsheet, create a chart, and send a report via email - all in one go, without needing to be re-prompted.

Read that again. No repeated prompting.

You don't need to tell it how to do each step. Just tell it what you need to accomplish, and it can arrange the entire workflow on its own. It will decide which data to extract, how to analyze it, what format to use, and who needs to see the data.

That's not a tool. That's a colleague who never sleeps, eats, or takes vacations.

The Moment I Realized It Was Real

I'm naturally skeptical and question everything. I assume companies exaggerate their achievements until proven otherwise. But then I noticed the same pattern across different companies, industries, and use cases.

Klarna's AI assistant got off to a strong start - handling 2.3 million conversations in the first month. But what really caught my attention is that by May 2025, they adjusted their strategy. They realized some important factors that pure efficiency metrics couldn't capture.

Trust and satisfaction aren't just about transactions. They also involve emotions. Sometimes, customers need to talk to a human, not because AI can't technically solve their problems, but because they need someone to care about their issues.

So, Klarna adopted a approach where human customer service handles complex issues, while AI is used for routine tasks. You know what happened? The system became more efficient. Now, the AI can support the workload equivalent of 800 full-time employees - more than when relying solely on AI.

This development makes me think this isn't just hype. Companies are learning, adjusting, and finding what really works. They're not blindly applying AI everywhere - they're exploring where AI can truly create value and where humans are still irreplaceable.

Why I'm Paying Such Close Attention to Salesforce

Salesforce is important because it's not a startup with no track record, just talking big. It's a company with a market value of over $250 billion and more than 150,000 customers who rely on its platform for their critical business operations.

When they go all in, they're risking their relationships with enterprise customers, who can't tolerate failed experiments. They can't afford to hype ineffective technologies. They're fully committed to the Agentforce project.

The evolution of this platform is incredibly fast, which is rare in the enterprise software field. Agentforce 3 was released in June 2025. Compared with the version in January 2025, the latency was reduced by 50%. It supports using web search as a data source and added a command center for full observability.

The last point is crucial. The command center allows you to clearly see every action, every decision, and every interaction of the AI agent. This isn't just a feature built for show. It's to meet the needs of customers to deploy these systems on a large scale and build trust.

Moreover, its functions are constantly expanding. By mid - 2025, Agentforce agents will be able to autonomously perform operations such as creating or updating CRM records, generating new data entries (such as leads or cases), and making decisions that previously required human judgment.

The Part That Keeps Me Awake at Night

I've been thinking about one thing: We're developing much faster than we're prepared.

The global intelligent agent AI market is expected to explode from $7.28 billion in 2025 to over $41 billion by 2030. Some analysts predict that by 2030, AI agents will manage 80% of digital workflows in customer service, IT, human resources, and sales.

Five years. In five years, most digital work in business functions will be done by autonomous AI.

I think most companies, or rather, most employees, don't realize how fast this is happening.

Look at what's already happening. Companies implementing AI agents report a 7.8% increase in productivity and a 30% reduction in the time spent on repetitive tasks. Some organizations claim that up to 65% of internal tasks (such as IT support tickets or routine customer requests) are already automated.

These aren't predictions. They're happening right now, in 2025.

And this technology is accelerating. Every few months, these systems become more capable, can handle more complex tasks, and require less human intervention. The update of Agentforce 3 reduced the latency by 50% in just five months. This isn't a gradual improvement; it's exponential growth.

What I'm Really Worried About

I'm not worried that AI will replace all jobs. That's a thought that came from a moment of confusion and completely ignores what's happening. I'm worried about the gap between what these systems can actually do and our understanding of how they work.

Klarna realized that pure AI isn't enough - they need humans involved when dealing with complex and emotional situations. But how many companies will learn this lesson the hard way? And how many companies will deploy AI without understanding its strengths and weaknesses?

I'm worried that the speed of technology adoption will outpace the growth of people's wisdom.

Salesforce talks about empowering one billion agents by the end of 2025, and I believe they can do it technically. But are we ready for one billion autonomous systems making decisions that affect people's real lives, finances, healthcare, and legal issues?

Technology is advancing at an astonishing pace, but our understanding of how to manage it, when to trust it, and where to limit it is progressing slowly.

This Is More Important Than You Think

I keep thinking about what Marc Benioff of Salesforce said, that this is "the third wave of AI." It's not about better tools or smarter assistants; it's a fundamental transformation in the way we work.

The first wave of AI was AI that could recognize patterns - image recognition, speech - to - text, etc.

The second wave was generative AI - systems that can create content, write code, and answer questions.

The third wave refers to intelligent agents that can observe, make decisions, and take action autonomously. These systems not only help you work but can also do the work for you.

This isn't a gradual change; it's a stage - by - stage transformation.

Moreover, this wave is coming faster than any previous one. ChatGPT reached one million users in just five days. In just a few months, thousands of enterprises have deployed these AI agents.

The adoption curve is vertical.

What I'm Going to Watch Next

I'm keeping an eye on three things that will tell me where things are really going.

First: How companies deal with failures. AI agents will make mistakes. They'll misunderstand context, make wrong decisions, and cause problems. Companies that can detect these failures early, learn from them, and establish better safeguard mechanisms will ultimately succeed.

Second: Where are the real limits? Currently, everyone is looking for easy breakthroughs - customer service, data entry, scheduling. But where will AI agents hit a bottleneck? What tasks require human judgment that can't be automated? Finding these boundaries is more important than celebrating success.

Third: How employees adapt. This isn't about job loss; it's about job transformation. The most successful teams will be those that know how to cooperate with AI agents, rather than compete with or resist them. But this requires training, a change in mindset, and organizational restructuring, and most companies haven't even started planning for these.

Why I'm Writing This Article

I'm not trying to hype up AI agents. To be honest, it's the opposite. I'm trying to cut through the fog and see the truth. Real companies with real customers are deploying autonomous AI on a large scale and achieving tangible results. This isn't speculation; it's reality.

But I also want to sound the alarm. We're entering uncharted territory, and it's happening very fast. The development of this technology far exceeds our ability to understand its impact, regulate its use, or cope with its consequences.

Klarna's experience tells us that just pursuing efficiency isn't enough - you also need to understand when humans are still crucial. Salesforce's rapid development shows how fast these systems are improving. Market forecasts indicate that this trend is about to sweep across all industries.

The question isn't whether AI will come. They're already here.

The Question Is: Are We Paying Attention?

Whether we're ready or not, the way we work is changing. Unlike previous technological changes that took decades to gradually implement, this change could happen in just a few years - or even a few months.

I don't know if this excites or scares you. To be honest, I feel both.

But I know: Ignoring it is no longer an option.

This article is from the WeChat official account “Data - Driven Intelligence” (ID: Data_0101). Author: Xiaoxiao. Republished by 36Kr with permission.