AI agents
AI agents are rapidly gaining in importance — and are changing decision-making processes in many industries. But what happens when they find their way into R&D and QA environments? With structured data and clear goals, they proactively analyze, make decisions and make recommendations — for more innovation, better quality and faster development cycles.
Why are AI agents on the agenda right now?
AI agents are regarded as a key development path for the next wave of technology. The consulting firm Gartner expects that around 15% of all business decisions will be influenced by agent-based systems by 2028 (report).
IBM also sees the concept of “Agentic AI” as a new phase of intelligent automation. Technical maturity, available data and cloud infrastructures are creating the conditions for their practical use today for the first time — even in an industrial environment, according to a trend report from IBM.
These technological advances open up entirely new opportunities in lab environments: processes can be optimized based on data, innovation accelerated, and quality assurance made smarter.
This is ensured by technological developments such as more powerful cloud infrastructures, larger amounts of available data and improved AI models. This opens up completely new opportunities in the laboratory environment: Processes can be optimized based on data, innovations can be accelerated and quality assurance can be made smarter.
What are AI agents?
An AI agent is an intelligent software program that independently pursues goals, analyses data, prepares decisions and derives concrete suggestions for action or process interventions from them. In contrast to classic AI, which usually works reactively and waits for specific queries, agents act proactively: They recognize what needs to be done — and do everything they can to efficiently achieve the defined goal. To do this, they need three key elements:
Input: The agent processes structured input data — for example laboratory data, project context, previous results, or clearly defined requirements.
Logic: Using decision rules, machine learning, or heuristics, the agent evaluates various options and selects the most appropriate strategy.
Output: The results are suggestions for action, automated recommendations or even direct process interventions — which are fed back to people for a decision.
A key characteristic of agents is continuous learning through feedback: They analyze which suggestions have been implemented and how successful they were — and improve their decision-making logic on this basis.
Especially in the laboratory environment, this means real development: AI agents help not only to manage material data, but also to use it intelligently — for more informed decisions, shorter development cycles and more efficient quality assurance. An example: When a customer needs a new coating for extreme temperatures, the agent can analyse historical recipes, requirements and test results — and within a very short period of time, suggest a suitable combination that meets the framework conditions.
How do AI agents work?
An AI agent processes information in three steps:
- First, he needs structured, digitally available data. Typical sources include LIMS systems, ERP databases, Excel lists or specialized material intelligence platforms such as LabV. Relevant data can include recipes, material properties, process parameters, or test results.
- In the second step, the agent recognizes defined goals, searches for patterns in the available data, evaluates relationships and develops concrete suggestions based on his decision logic.
- In the third step, the results are presented: The agent provides recommendations for action, optimized recipes, test plans or warnings in the event of deviations. People always remain the last resort, which reviews and decides on the proposals.
Practical examples
R&D paint development: From historical knowledge to a new recipe
A customer from the paint industry required a coating solution for extreme operating conditions. Instead of relying on traditional trial and error methods, an AI agent was used.
The process:
- Analyze historical data: The agent examines previous recipes and test results under similar conditions.
- Integrate framework conditions: Requirements such as temperature resistance, adhesion and regulatory requirements are taken into account.
- Create proposals for action: The agent develops optimised recipes and provides reasons for his recommendations.
- Selection and fine-tuning: The person makes the final selection and controls further fine-tuning based on feedback.
The result: Development time was significantly reduced, resources were used more efficiently and the probability of success increased significantly.
Quality assurance for recycled materials: Early identification of deviations
An AI agent also successfully supported the processes when it came to quality assurance of recycled materials. Material fluctuations could be identified at an early stage and compensated for in a targeted manner.
The process:
- Check input data: The agent analyses new batches of materials and compares them with historical quality data.
- Identify abnormalities: Deviations from defined standards are automatically detected.
- Derive recommendations for action: The agent provides suggestions for mixing with new products or for adapting process parameters.
- Integrate feedback: Test engineers evaluate the suggestions so that the agent can continue to learn from the results.
The result: The material quality could be stabilized, the scrap rate reduced and the testing effort reduced.
Requirements and limits
For AI agents to be fully effective, a number of requirements must be met. In particular, a structured, digitally available database is essential for successful use. Without high-quality input data, an agent cannot provide reliable analysis or recommendations. Clear target definitions are just as important. The agent must know what he is working towards — for example, developing an optimised recipe or identifying quality discrepancies.
Feedback mechanisms are another success factor. Only when agent suggestions are evaluated and feedback is systematically integrated can the agent learn and improve. Last but not least, people always remain the final decision-making authority. The agent makes suggestions — the responsibility remains with people. Agents are therefore tools and not replacement instances. Their strength develops in combination with human expertise.
Conclusion: Why connected data is important now
AI agents are not changing practice overnight — but they are opening up new perspectives for data-based decisions. The prerequisite for this is that the right foundations are laid today: clean data structures, networked systems and a well-thought-out use of digital knowledge.
Anyone who starts to consistently record their material data, link it sensibly and make it digitally usable is creating the conditions for the next step: the targeted use of intelligent agents. They can then automate processes, provide suggestions, or prepare decisions — embedded in a system that is understood and supported. We will discuss in detail the opportunities, challenges and specific use cases associated with this in the webinar “AI Agents in the Laboratory — Assistance Systems for R&D and QA.”
The first step is not agent technology — but the decision to use data strategically. Companies should work on this today so that they can really benefit from developments in artificial intelligence in the future.