What Are Agentic Engineering Systems?

Three Words That Are Not Synonyms

Automated, autonomous, and agentic are used interchangeably in marketing materials and even in some technical discussions. In engineering practice, the distinctions matter because they determine the level of trust required, the governance needed, and the role of the human in the loop.

Automated means following a predefined script. An automated pipeline runs the same sequence of steps every time it is triggered. A parametric CAD model updates geometry when input parameters change. A CI/CD pipeline runs tests when code is committed. The system does exactly what it was programmed to do, in the order it was programmed to do it, with no deviation. If the script does not cover a situation, the system stops or fails.

Automation is powerful and well-understood. The entire digital engineering pipeline (DGTLENG 302 in the previous curriculum) is built on automation — connecting tools, running simulations, checking consistency. The human defines the workflow; the system executes it.

Autonomous means operating independently within a defined domain, making decisions based on rules or models without human intervention for each decision. An autonomous vehicle navigates traffic. An autonomous satellite adjusts its orbit. An autonomous manufacturing cell adapts its process parameters based on sensor feedback. The system perceives its environment, decides on an action, and executes — without waiting for human approval.

Autonomy requires more trust than automation because the system makes decisions the human did not explicitly pre-approve. The decisions are bounded by the system's programming, but within those bounds, the system acts on its own judgment (encoded in rules, models, or learned policies).

Agentic means taking initiative within constraints, reasoning about goals, decomposing problems into subtasks, and adapting strategy based on intermediate results. An agentic system does not just execute a workflow or follow a policy — it reasons about what to do next given the current situation and its objectives. It can delegate subtasks to other tools or agents. It can ask for clarification. It can recognize when it is stuck and request human intervention.

The difference between autonomous and agentic is subtle but important. An autonomous system has a fixed decision policy — it maps situations to actions in a predetermined way. An agentic system has goals and the ability to plan — it determines what actions to take to achieve those goals, potentially in ways that were not anticipated by its designers. The agent has more flexibility and more potential for both helpful and harmful behavior.

Agentic Behavior in Engineering Context

In engineering, agentic behavior means an AI system that can:

Decompose a goal into subtasks. Given an objective like "find a bracket design that meets structural and thermal requirements within 2 kg," the agent breaks this into: generate candidates, evaluate structural performance, evaluate thermal performance, filter by mass constraint, rank by multi-objective criteria, and present the shortlist. The decomposition is not hard-coded — the agent determines the steps based on the goal and available tools.

Select and use tools. The agent has access to a set of engineering tools — CAD parametric models, FEA solvers, surrogate models, optimization algorithms, database queries — and chooses which tools to use for each subtask. A structural evaluation might use a surrogate for initial screening and full FEA for the shortlisted candidates. The agent makes this decision based on the accuracy requirements and the computational budget.

Adapt based on intermediate results. If the initial design space exploration reveals that all candidates violate the thermal constraint, the agent can adjust its strategy — expand the material options, relax a less critical constraint and report the trade-off, or focus exploration on a specific region of the design space that shows thermal promise. This adaptation is not scripted — it emerges from the agent's reasoning about its goals and the current situation.

Collaborate with humans. The agent recognizes when human input is needed — when constraints are ambiguous, when trade-offs require value judgments, when the situation falls outside its competence. It presents information to the human in a useful format, explains its reasoning, and incorporates human guidance into its subsequent actions.

Maintain context across interactions. The agent remembers previous decisions, understands how they constrain current options, and can explain the chain of reasoning that led to its current recommendation. This context persistence distinguishes agents from stateless tools that start fresh with each invocation.

Where Agentic Systems Appear in Engineering

Agentic engineering systems are emerging across the engineering lifecycle:

Design exploration agents (covered in Lesson 2) search design spaces, generate and evaluate candidates, and present curated shortlists to engineers. These agents combine generative design (DGTLENG 301 Lesson 2), surrogate modeling (DGTLENG 301 Lesson 3), and optimization to explore far more of the design space than manual methods allow.

Verification and validation agents (covered in Lesson 3) generate test cases, identify coverage gaps, run model checks, and flag inconsistencies. Instead of a human tester deciding which tests to run, the agent determines where the verification coverage is weakest and prioritizes testing resources accordingly.

Requirements analysis agents examine requirement documents for ambiguity, inconsistency, incompleteness, and testability. They can cross-reference requirements against the system model (DGTLENG 101, DGTLENG 201) to identify gaps and conflicts.

Manufacturing planning agents determine process sequences, tooling selections, and scheduling to meet production requirements. They adapt plans when constraints change — a machine goes down, a material is delayed, a priority shifts.

Maintenance planning agents combine predictive maintenance models (DGTLENG 206 Lesson 3) with logistics optimization to schedule maintenance activities, order parts, and coordinate crew availability.

Each of these represents a different point on the automation-to-agency spectrum, and each raises different questions about trust, validation, and governance.

Explore each system type. The critical insight is that the three types form a spectrum of increasing flexibility and increasing governance requirements. Automated systems are predictable but rigid. Autonomous systems are adaptive but policy-bounded. Agentic systems are flexible but require the most careful oversight because their behavior is goal-directed, not pre-scripted.

Assessment