• Written by Azhar Aliyar, Director

Can we really automate the entire workflow for resolving supplier escalations in our automotive manufacturing process, or are there still too many judgement calls that require human intervention? This question is harder to answer than it seems, because the automotive industry has several characteristics that make automation more complicated.

What automotive makes harder

In the automotive industry, supplier escalations, engineering change propagation, and warranty root cause analysis are all critical workflows that can be difficult to automate. Supplier escalations, for example, often require a deep understanding of the supply chain and the ability to negotiate with suppliers. Engineering change propagation involves updating multiple systems and stakeholders, which can be a complex and time-consuming process. Warranty root cause analysis requires a thorough understanding of the vehicle's design and manufacturing process, as well as the ability to analyze large amounts of data.

These challenges are exacerbated by the tiered supply chain and tight takt times that are typical in the automotive industry. Tiered supply chains mean that there are multiple layers of suppliers, each with their own systems and processes, which can make it difficult to track and manage inventory and production. Tight takt times mean that production must be carefully planned and executed to meet customer demand, which can leave little room for error or delay.

What separates an agent from a chatbot

An agent is a type of AI system that is capable of taking multi-step actions to achieve a goal, rather than simply answering a single question. This is in contrast to a chatbot, which is typically designed to provide a simple response to a user's query. An agent, on the other hand, can use tools, maintain state, and work towards a goal, making it a more powerful and flexible system.

A good example of an agent in action can be seen in the work of a production planner, who uses a system like AsscherAi to analyze production schedules and supplier lead times. The planner can ask the system to identify potential bottlenecks and propose solutions, which the system can then implement by updating the production schedule and notifying the relevant stakeholders.

Where a human approval step belongs

While agents can automate many tasks, there are still some situations where human approval is necessary. This is particularly true in situations where the agent is making a decision that has significant financial or safety implications. In these cases, it is often premature to remove the human approval step, as it provides an important check on the agent's decision-making process.

For example, a shift supervisor may use an agent to analyze the maintenance log and identify potential issues with the production equipment. However, before the agent can implement any repairs or maintenance, the supervisor must review and approve the plan to ensure that it is safe and effective. This approval step provides an important safeguard against errors or unexpected consequences.

How an agent fails differently from a model

An agent can fail in different ways than a model, because it is taking actions in the world rather than simply making predictions. If an agent fails, it may cause real-world consequences, such as delays or defects in production. This means that it is especially important to log and analyze the agent's actions and decisions, in order to understand what went wrong and how to improve the system.

By logging the agent's actions and decisions, organizations can identify patterns and trends that may indicate a problem. For example, if the agent is consistently proposing solutions that are not effective, it may indicate a problem with the system's training data or algorithms. By analyzing these logs, organizations can refine the agent's performance and improve its decision-making abilities. To learn more about how AsscherAi can help with this process, visit our website at www.asscherai.com.

What this does not do

While agentic AI workflows can be a powerful tool for automating complex processes, they are not a solution to every problem. In some cases, the workflow may be too simple or too well-defined to require an agent, and a simpler system may be sufficient. In other cases, the workflow may be too complex or too dynamic to be automated at all, and human judgement and decision-making may be necessary.

Before implementing an agentic AI workflow, organizations should carefully consider whether it is the right tool for the job. They should also consider the potential risks and challenges associated with automation, such as the need for ongoing maintenance and support. To get in touch with our team and discuss how AsscherAi can help with your automation needs, visit our contact page.