AI agents are the hype of the year 2025, but from a technological perspective, we have not yet reached the point where the big promises will pay off. However, it is undisputed that AI agents are coming and will deliver corresponding added value. This blog article explains what a sensible use of AI agents could look like.
Table of Contents
What are AI agents?
We have already written our own blog articles about AI agents and multi-agent systems. In a nutshell: AI agents are autonomous systems that can independently find a solution to a task and seek assistance where necessary. This means they do not need a predefined workflow and can therefore deal with challenges much more flexibly. They can also initiate processes independently if required.
Basics for AI agents
For AI agents to work properly, they must have the right technical functions on the one hand and the relevant knowledge to make the right decisions on the other. Many companies are currently building AI agents in and around their own ecosystem: Salesforce with its Agentforce for customer service and sales, DMS providers within DMS systems, etc. In the medium term, this will mean that there will be various AI agents that specialize in certain areas, departments or industries. We will explain the implications of this for enterprise architecture in detail later on.
Technical foundations for AI agents
AI agents will only be useful if they can interact with other systems. To do this, they need interfaces to other systems in order to trigger processes or obtain information. This blog article will not go into the technical aspects in detail; it is much more about the larger, strategic issues.
Contextual knowledge as the basis for good decisions for AI agents
Contextual knowledge is required for an AI agent to make good decisions. An example:
- A customer support agent receives a customer question in the ticket system and has to answer it.
- In order to answer it, it needs contextual knowledge – e.g. invoice data from the ERP or a guideline on how to handle certain information on a topic.
- An AI agent must be able to obtain this contextual knowledge somehow.
Of course, there are also use cases where it is sufficient for an AI agent to obtain the information from a system. In the medium term, however, information from different systems will be required for the major use cases in order to be able to solve the tasks.
Since we have already established in the section “Basics for AI agents” that there will be AI agents from various providers, this would mean that all DMS, CRM and other providers would now have to start writing their own connectors to the various systems. In practice, however, there are a few things that speak against this:
- Developing connectors that can process really large amounts of data is extremely challenging
- Such connectors would have to be able to adopt existing access rights and also take any changes into account
- Customers use so many different solutions that an extremely large number of connectors would have to be developed.
It therefore makes no sense for DMS, CRM and other providers to develop their own connectors on a large scale. Instead, such providers will rely on enterprise search solutions such as amberSearch, which have already solved the aforementioned problems.
Enterprise search as the basis for AI agents
An enterprise search should be understood as a cross-system search that enables employees to access internal know-how quickly and easily. Modern solutions such as those from amberSearch have been relying on vector-based indices since 2020 in order to process information with the help of LLMs. This is not directly possible with classic keyword-based indices, which were still used until the 2010s, for example.
So if an AI agent requires specific contextual knowledge, it makes much more sense in terms of IT architecture if it obtains this knowledge via an enterprise search. For example, a query could be made in this style:
“Search policy X, taking into account the access rights of person A and all information accessible to all employees.”
This would enable the enterprise search to provide the AI agent with all relevant information – regardless of whether the relevant information is in the DMS, SharePoint or on the drives.
Future outlook for AI agents
Artificial intelligence continues to develop at an unstoppable pace, which means that more and more AI agents will come onto the market for more and more use cases, capable of solving increasingly complex tasks. One possible scenario of how AI agents could be used in the future is shown in this video:
IT security & AI agents
Anyone who reads through this blog article and is a little creative will understand what is potentially possible with AI agents. A lot can be automated without traditional workflows – provided the relevant systems have the necessary interfaces.
Such systems will inevitably lead to IT security issues. To reduce the risks of AI agents, it is better to use several specialized AI agents with specific knowledge. AI agents with limited rights should correct and check each other in order to work together on solutions so that an AI agent does not have too many rights. Where appropriate, there should be a “human-in-the-loop” – e.g. before sending critical emails, contracts, orders, etc., such steps should be explicitly approved by humans. In addition, AI agents will need to be able to take existing access rights into account. However, this would be a challenge that is covered by enterprise search solutions.
Ultimately, we are – as of 2025 – still very early on with the topic of AI agents and realization on a larger scale will still have to develop over the next few years. Technological standards still need to be developed and technological solutions still require some optimization. However, it is important for companies to have a target picture so that they know what they need to prepare for and what their own IT infrastructure could look like in just a few years’ time.
AI agents for private use
At amberSearch, we focus on AI agents for companies. Nevertheless, the question of how AI agents can be used in the private sphere comes up again and again. Basically, there are 3 major use cases:
1. Research
AI agents can be used by private individuals to carry out detailed searches on the internet and on various websites. However, the same applies here: Such systems will only search information that is truly public. All other systems would have to be connected separately.
2. Consumption
AI agents can be used by private individuals to support general consumption. For example, tasks could be: Order spare part X of product Y to my home. Or plan and book me a vacation to Scotland for 10 days at the beginning of September.
3. Automation
AI agents could use this to control certain things in the private sphere – e.g. the smart home, general household management or personal assistance that coordinates appointments with the doctor. It remains to be seen to what extent such systems will create corresponding interfaces and how well these use cases will be accepted.
Conclusion: AI agents as a future technology with strategic potential
AI agents are still in the early stages of development, but their potential is enormous. Companies that get to grips with the integration of this technology at an early stage can achieve considerable efficiency gains in the long term. The key to success lies in the right strategic planning: a well thought-out IT architecture, the use of specialized agents with clearly defined authorizations and access to high-quality contextual knowledge are essential.
While the technical implementation will continue to mature in the coming years, companies should develop a target image for the sensible use of AI agents now. This will enable them to optimally prepare for the coming changes and benefit from the advantages of this technology in the long term.