[Media Insight] The Biggest Misconception About Agents: They Can Solve Every Problem
AI agents are not master keys; IBM experts will take you to uncover industry misunderstandings.
Core content:
1. What can AI do and what can’t do?
2. Analysis of IBM's AI agent solution Watsonx Orchestrate.
3. How to identify "fake agents" in the market.
Editor's Note: The development of AI agents is already unstoppable, but there is often a gap between ideals and reality. At present, most agents can realize data analysis, trend prediction and a certain degree of workflow automation. In simple scenarios, the correct tool can be selected to complete tasks. However, in the face of complex scenarios, the technological maturity is still insufficient. In IBM's view, the essence of enterprise-level AI is not to show off skills, but to reconstruct business. On the Agent track, IBM is still using a combination of "full stack technology + industry Know-How + open ecology". This process does not pursue speed, but emphasizes more on technical depth and landing accuracy.
Based on this, we share with you the in-depth good article "The industry's biggest misunderstanding of Agent: It can solve all problems". The article was originally published on InfoQ, the details are as follows. Please be sure to indicate the author and source of the article when reprinting.
What is the biggest misunderstanding in the industry about AI Agent at present?
"IBM Wu Minda, a senior technical expert in data and artificial intelligence in the Greater China Science and Technology Department, answered this way at a media roundtable recently.
AI Agent is the "must-fight place" for almost all technology companies at present. During the Think 2025 conference, IBM also launched an upgraded version of the AI agent solution watsonx Orchestrate: it provides pre-built, out-of-the-box professional field agents (such as human resources, sales and procurement agents, etc.); supports enterprises to build their own AI agents within 5 minutes; through agent orchestration tools, it can achieve multi-agent and multi-tool coordination required for complex projects; and can provide observability for the entire life cycle of the AI agent, including performance monitoring, protection, model optimization and governance.
AI Agent The "turning point moment" for large-scale applications has arrived, which is the consensus in the industry. However, IBM also emphasizes that there is no need to over-“deify” AI . "The essence of technology depends on whether it can solve the real problems of the company, especially when it is bound to the core business, it must return to the business scenario to see whether the technology truly generates value." Zhai Feng, general manager and chief technology officer of IBM Greater China, said.
In other words, AI cannot solve all problems, and not all problems need to be solved with AI; the same applies to agents.
"true-false" agent
Different from traditional AI assistants (such as chatbots), AI Agent can not only understand instructions and generate content, but also independently plan task paths based on real-time data, call multi-system resources, and dynamically optimize strategies during execution.
These excellent qualities have made the industry flock to AI Agent, and of course, they have also caused many "old wine in new bottles" products on the market. Although they have been replaced by AI Agent packaging, the kernel is still traditional AI tools.
Wu Minda told reporters that it is not difficult to identify the so-called "false agent". "The computing power of pure 'old wine' is not used (used) when running. The Agent has an autonomous 'brain' and needs to think about things and calculate constantly. At this time, it needs to accumulate computing power. However, the previous automation processes or AI model calls were basically arranged in advance and calculated with historical data. There was not so much resource consumption during operation, and the required computing power was very small, so the CPU could run."
Through a unified platform portal called AskIBM, IBM is also using an AI Agent to empower employees. According to reports, AskIBM can automatically route to vertical agents such as HR, IT, sales, and procurement based on employees' intent to query, realizing the full process automation from problem analysis to system interaction.
In Wu Minda's opinion, building an AI agent is easy, but in the enterprise, I want to use AI. If you do a good job of application, you need to achieve scale, and this is difficult. First, behind Agent development involves different frameworks, applications, and manufacturers. How to connect with each other? Second, how can enterprises find high ROI and suitable scenarios? Third, how to manage the entire life cycle of Agent from construction, production, to operation and maintenance?
In response to these problems, Watsonx Orchestrate has a clear architectural design.
IBM has three core ideas:
The first layer is the vertical domain agent matrix that is out of the box. Including the first batch of 3 Artificial AI agents: human resources agents, sales agents and procurement agents. "For example, the training, ID, or permission application for new employees after joining the company does not even require HR intervention, and the robot behind it can answer 90% of the questions. It doesn't matter how many AI agents are behind this. From a business perspective, this doesn't have to be a bother at all." Zhai Feng said. At present, human resources agents have been officially launched, and sales and procurement agents are also planned to be opened for use in June.
"For domestic companies, these agents can be used as templates when using them, and then adjusted according to their actual needs." Wu Minda gave an example.
The second layer is multi-agent orchestration. Once the agent development is launched and deployed to the agent catalog (Agent Catalog), it can be opened to departments or other employees for use. This directory is similar to an agent warehouse, which supports classified retrieval, permission management and version control. Administrators can also set access rights and publish and share the agent through the approval process.
But as the number of enterprise agents increases from dozens to hundreds, the complexity of management will become higher and higher. In this regard, Watsonx Orchestrate also introduced a multi-agent orchestrate orchestration function to support cross-agent collaboration. For example, after the sales agent obtains new customer clues, it will automatically trigger the market agent to analyze the competitor dynamics, and then call the customer service agent to generate a personalized follow-up strategy. Moreover, whether it is a professional field agent built by the enterprise itself, a partner building or an open source community, it can realize information sharing and collaborate on complex multi-step processes.
The third layer is an open ecosystem and open-source collaboration. The front end is a unified entrance, and behind it is a very open, intelligent ecosystem. Watsonx Orchestrate integrates over 80 industry-leading enterprise-level application tools from companies such as Adobe, AWS, Microsoft, Oracle, Salesforce Agentforce, SAP, ServiceNow, and Workday. For example, an enterprise can choose to call Salesforce's sales forecasting agent directly in Orchestrate without repeatedly developing a docking interface.
Is the data from AI Ready?
Zhai Feng said that AI applications without data are empty talk. If enterprises want to implement AI, first ask themselves three questions: Is there any high-quality data? Are these data being used? Does it really work?
In other words, having data does not mean that data can be used well. "More than 90% of the enterprise data is actually unstructured data, but now people are paying more attention to structured data." Wu Minda emphasized, "So, helping enterprises improve their use of unstructured data is also the main focus of IBM."
"We believe that this method has higher accuracy than RAG because the documents in it are not directly vectorized, and there is an extraction process in the middle. Specifically, we use Watson.Data integration to process structured and unstructured data. For unstructured data, it will extract the entities and values in the process of vectorization, and then vectorize the document. In the future, when a large model is doing a knowledge base query, it will not only return similar vectors, but also return the relevant entities and values, and improve the accuracy through the assistance of entities and values." Wu Minda said.
Watsonx. Data integration is a comprehensive data integration tool, and Data Replication was provided by IBM in the past. Different data processing tools, Watsonx.Data integration can support both structured data and unstructured and semi-structured data.
From down, when the data is put into watsonx.data and integrated by watsonx.data integration, watsonx.data intelligence I started to "work", and its function was to provide unified data governance and data blood ties. Wu Minda gave an example, "The same batch of data has multiple ways to access, such as using the large language model knowledge base - RAG to Q&A, or using traditional SQL query - report query, and using machine learning to model and extract data to train a model, which is also a way. How to ensure that the permission control of different access methods is together? This situation can be controlled through watsonx.data intelligence."
At the same time, IBM also encapsulates the managed data into API interfaces or vector databases for real-time calls by the agent. For example, supply chain agents can directly access real-time inventory vector data and dynamically adjust procurement plans. This not only improves data availability, but also provides “nutrients” for the continuous evolution of Agent.
Looking at the entire link, if we compare the raw data in the enterprise source system to the raw materials of the manufacturing factory, then the function of watsonx.data integration is to manufacture and process the raw materials, and then put them in the watsonx.data warehouse, the asset directory is then managed through watsonx.data intelligence, and finally provided to the front-end AI and BI for use.
Is the process automated?
1. Each enterprise has an average of thousands of applications. How does an AI Agent connect and connect with these systems and applications?
The dilemma of heterogeneous systems exists at any stage of enterprise development. The interfaces between systems are different and the standards vary greatly. The problem in the past was how to connect different systems and how to break data silos, but the problem now is AI How to break the "dimensional wall" between the Agent and these systems, how to call the data in it and perform related tasks.
For example, if a company receives feedback from customers' quality issues, it will give it to the AI Agent Processing: The first step is that it needs to feed the problem back to the quality management system, and then call data from different software systems such as production, design, and process for analysis in order to locate the root cause of the problem; the second step is that after confirming the problem, the enterprise knowledge base needs to be updated to avoid the recurrence of similar situations; the third step is that it needs to notify the relevant person in charge, at this time it needs to call different communication applications such as email, DingTalk or Qiwei; finally, if it is a problem with external supplier parts, it involves external communication, so the information generation document may be required to be sent in EDI format.
"To string all these processes together, every step is not easy. The AI Agent must be effectively integrated with the existing system." Zhang Cheng, senior technical expert in automation at IBM Greater China Science and Technology Division, said that although integration is a relatively "old" concept, its importance and value are still increasing today. During the Think 2025 conference, the Hybrid Integration released by IBM is mainly to provide complete cross-platform integration capabilities on and off the cloud.
2. How to visualize multiple agent links and how to deal with possible errors during task execution?
For example, in a certain AI Agent After going online, if there are problems such as delay, network flash, or memory overflow, and downtime, how can you monitor the entire link in real time, actively detect it, and quickly diagnose, analyze and deal with it? This process relies on automated IT operation and maintenance capabilities.
At the operation and maintenance level, IBM proposed the concept of AgentOps, that is, it hopes to visualize the entire link from the construction, deployment and launch of AI Agent to the operation and maintenance, optimization and iteration. For example, through the Instana full-stack monitoring tool, the call link, resource consumption and decision accuracy of the Agent are tracked in real time. When the response delay of an agent exceeds the threshold, the system can automatically trigger the capacity expansion mechanism.
3. How to allocate resources reasonably to ensure efficiency while minimizing costs?
IBM believes that automation must be achieved at the infrastructure level so that AI can use basic resources highly automated and resiliently. Observability tools become critical to this, including discovering, managing, monitoring and optimizing the use of agents across the enterprise, ensuring efficient and responsible technology adoption.
In this regard, the Watson product portfolio provides a set of monitoring tools that can monitor AI performance and reliability, execute AI guardrails, and effectively use AI resources, for example, evaluate and select specific goals such as cost-effectiveness or performance AI model. In addition, IBM also announced the acquisition of HashiCorp last year, which is positioned to help enterprises achieve automated and on-demand full life cycle management from the underlying infrastructure level.
"Related data shows that on average, 27% of enterprises' cloud computing spending is wasted, and these wastes can be completely analyzed through the platform to better realize the deployment of underlying resources. This is IBM Automation software that focuses more on solving things." Zhang Cheng emphasized.
Return to the essence of the business to view technological value
When I wrote this, I don't know if you have noticed it, except for the AI Agent. In addition to this new concept, data, automation, etc, are actually still old topics that enterprises often talk about in the information and digital era. Ultimately, whether it is people who perform tasks, make decisions, or AI that performs tasks, makes decisions, a relatively complete and mature IT infrastructure is essential, and this is a "class" that enterprises must make up for.
Solved this problem, and then looked for the scene.
"Every enterprise development stage is different, and the bottlenecks encountered are also different. Enterprises must first think clearly about which place is the real pain point. Whether you want to reduce costs and increase efficiency or innovate your business, the enterprise's demands must be clear. It is not today's HR Agent. Here, you only have 3 HRs in your company and you need to use them," said Zhai Feng.
So, how can enterprises accurately grasp business needs and make internally built AI agents more targeted, high-value and continuously optimized? Zhang Xun, manager of the Garage Innovation Team of IBM Science and Technology Department, summarized that enterprises should do it through advanced paths and realize intelligence step by step through continuous iterative optimization.
"Enterprises must first ensure that investment is controllable, so our team usually chooses the most typical scenario with customers through POC (proof of concept) to deploy our ideas and products, and then verify its ROI. If it meets expectations, it will be deployed on a large scale." Zhang Xun told InfoQ reporters, "The entire POC process is about 30 days, but the plan is iterated every week. This process requires customers to fully participate, provide timely feedback, and verify whether we are on the right path. If it is not right, we need to adjust continuously."
Taking the manufacturing industry as an example, the IBM garage innovation team has summarized four scenarios that can bring the largest ROI to the company: R&D, production, supply chain, and finance. Take IBM itself for example. Why is it a priority to release three agents, HR, finance, and procurement? In fact, it is also a scenario that has been verified internally by IBM itself and has a better ROI.
In short, in IBM's view, the essence of enterprise-level AI is not to show off skills, but to business reconstruction. As IBM Chairman Arvind Krishna said, "The era of AI experiments has ended. The competitive advantage of enterprises depends on tailor-made AI applications and quantifiable business results." On the Agent track, IBM is still using a combination of "full-stack technology + industry Know-How + open ecology". This process does not pursue speed, but emphasizes more on technical depth and implementation accuracy.
For enterprises, they must realize that no matter how cool the technology is, they cannot solve the problem of the nature of the business. The more dazzling the technical concepts and products are, the more they must calm down and practice their "internal strength" and quickly make up for their IT infrastructure capabilities. The basic premise of AI is the express train.
