• Written by Azhar Aliyar, Director

The production planning dashboard for a large packaging firm sits idle on a screen, its data untouched and unexamined, a testament to the complexities of managing a job-shop environment with heavy setup costs. The firm's printing and packaging operations are plagued by issues such as setup and makeready losses, quoting accuracy, and substrate waste, which can be exacerbated by the lack of effective workflow management.

Printing and packaging challenges

The printing and packaging industry is particularly susceptible to setup and makeready losses due to the high cost of switching between different jobs and the need for precise calibration of equipment. Quoting accuracy is also a challenge, as it requires taking into account various factors such as material costs, labor, and equipment usage. Furthermore, substrate waste is a significant concern, as it can result in substantial financial losses and environmental impacts. These challenges make it harder for firms to optimize their workflows and improve efficiency.

A shift supervisor at the firm checks the maintenance log and notices that the equipment is due for maintenance, which will require a significant setup time. This highlights the need for effective workflow management to minimize downtime and reduce setup losses. The supervisor must balance the need for maintenance with the production schedule, taking into account the constraints of job-shop economics.

Characteristics of agentic AI workflows

An agent is distinct from a chatbot in that it has the ability to use tools, maintain state, and work towards a goal. In the context of printing and packaging, an agentic AI workflow can utilize tools such as production planning software and equipment control systems to optimize workflows and minimize losses. The agent's ability to maintain state allows it to keep track of the production schedule, equipment status, and material usage, enabling it to make informed decisions and take actions to achieve its goal.

For example, an agentic AI workflow can be designed to automatically generate quotes based on historical data and current production costs, taking into account factors such as material usage and equipment downtime. This can help improve quoting accuracy and reduce the risk of errors. The agent can also use its tools to monitor production and adjust the schedule as needed to minimize setup losses and substrate waste.

Human approval and oversight

While agentic AI workflows can automate many tasks, human approval and oversight are still essential in certain situations. Removing the human approval step prematurely can lead to errors and unintended consequences. For instance, an agentic AI workflow may generate a quote that is based on incorrect assumptions or outdated data, which can result in financial losses if not caught and corrected. A human approver can review the quote and ensure that it is accurate and reasonable before it is sent to the customer.

Most people get wrong the idea that agentic AI workflows can replace human judgment and oversight entirely. This is incorrect, as human approval and oversight are necessary to ensure that the workflow is operating correctly and making decisions that align with the firm's goals and values. The production planning dashboard can provide valuable insights and data to support human decision-making, but it should not be relied upon as the sole decision-maker.

Failure modes and logging

An agentic AI workflow can fail differently from a model, as it is designed to take multi-step actions rather than simply providing a prediction or recommendation. The agent's ability to use tools and maintain state means that it can fail in more complex and nuanced ways, such as by generating incorrect quotes or schedules. To mitigate these risks, it is essential to log the agent's actions and decisions, including any errors or exceptions that occur. This can help identify the root cause of failures and improve the overall performance of the workflow.

The logging mechanism should be designed to capture relevant data, such as the input parameters, output results, and any errors or exceptions that occur. This data can be used to analyze the agent's performance and identify areas for improvement. For example, the log data may reveal that the agent is consistently generating incorrect quotes for certain types of jobs, which can indicate a problem with the underlying data or algorithms. By analyzing the log data, the firm can take corrective action to address the issue and improve the overall performance of the workflow. If you need help with implementing agentic AI workflows, you can contact us for more information.

Limitations and incorrect applications

Agentic AI workflows are not suitable for all situations, and there are certain scenarios where they may not be the best choice. For example, in situations where the workflow is relatively simple and does not require multi-step actions, a chatbot or other type of AI may be more appropriate. Additionally, agentic AI workflows may not be the best choice for situations where the goals and objectives are not well-defined or are subject to change. In these cases, a more flexible and adaptive approach may be needed.

Some firms may try to use agentic AI workflows to automate tasks that are better suited to human judgment and oversight, such as strategic decision-making or creative problem-solving. This can be a mistake, as agentic AI workflows are designed to operate within well-defined parameters and may not be able to handle complex or nuanced decision-making. By understanding the limitations and potential pitfalls of agentic AI workflows, firms can ensure that they are using these technologies in a way that is effective and efficient.