Agentic AI in Engineering

Executive summary 

EDRMedeso frames agentic AI not as a standalone feature, but as a new operating model for engineering. Synopsys’ AgentEngineer technology applies AI agents and multi-agent systems to engineering workflows, with agents able to reason, plan, learn and execute tasks with increasing autonomy. The strategic emphasis is on workflow transformation, engineering productivity and better use of scarce expertise rather than on conversational AI alone. 

The approach is organized around an L1 to L5 maturity framework, analogous to levels of vehicle automation. It begins with assistive automation and advances through collaborative, partially autonomous agents toward highly autonomous, self-directed systems. This gives customers a practical language for discussing progress without implying that full autonomy is immediately available. 

Three ideas are particularly relevant to simulation and digital engineering: 

  • The unit of transformation is the workflow, not the individual software tool 
  • Agents become a new user type that must interact with solvers, data, APIs, licenses and governance controls 
  • Human engineers remain responsible for objectives, constraints, oversight, validation and sign-off as autonomy increases. 
IMPLICATIONS - The near-term opportunity is not to primose engineering. It is to help organizations identify high-toil, bounded workflows where expert knowledge can be captured, execution can be governed, and measurable value can be demonstrated before increasing autonomy

 Key conclusions 

  • Agentic AI should be positioned as a maturity journey, with clear boundaries between current capability and future direction 
  • Workflow redesign matters more than adding a chatbot to an engineering product 
  • Multi-agent orchestration is the differentiator: specialist agents can coordinate between complex, multi-step engineering tasks 
  • Engineering trust depends on traceability, constraints, validation and human accountability 
  • Value is derived by combining domain expertise, workflow discovery, integration, governance and reusable technical infrastructure 

1. Why Agentic AI Matters to Engineering 

Engineering organizations are facing a familiar and common set of pressures: increasing product complexity, constrained specialist capacity, fragmented toolchains and shorter development cycles. Traditional automation can remove repetitive steps, but it is usually brittle, narrowly scripted and difficult to adapt when objectives, constraints or data change. 

Agentic AI introduces a different model. Rather than following a predefined sequence of tasks, engineering agents extend automation by interpreting goals, developing plans, determining actions, coordinating resources and adapting to changing circumstances while remaining within approved governance controls. 

The significance of this capability is not technical. It is operational. 

For the first time, organizations can begin to think about automating engineering workflows rather than simply automating individual engineering tasks, the possibility of persistent workflow participants rather than isolated assistants. 

From task automation to workflow agency 

The critical shift is from automating a known sequence of events or tools to managing a goal-directed process. A conventional script follows predefined logic. An engineering agent can potentially determine which steps are required, use available resources, respond to exceptions and ask for human intervention when confidence or authority is insufficient.

A PRACTICAL DISTINCTION - Copilots help an engineer perform work. Agents take responsibility for bounded portions of the work. Multi-agent systems coordinate several specialized responsibilities across and end-to-end workflow.

2. The Synopsys model 

AgentEngineer as the organizing concept 

Synopsys describes AgentEngineer as an agentic AI framework for engineering automation. Its agents can work individually or in coordinated teams, tackling multi-step activities that previously required substantial manual effort and specialist attention. Public examples include RTL generation, verification planning and testbench creation. 

The model centers on four related capabilities: 

  • Reason: Agents interpret objectives, constraints, context and available resources to understand what must be achieved. 
  • Plan: Agents determine the sequence of activities required to achieve an objective and adapt plans when conditions evolve. 
  • Learn: Agents improve through feedback, outcomes and experience, enabling more effective future execution 
  • Execute and orchestrate: Agents perform tasks and coordinate activities across tools, systems and other agents. 

Together these capabilities create a foundation for progressively increasing engineering autonomy while maintaining accountability and control 

 

The L1-L5 maturity framework

Importantly, this framework provides organizations with a language for discussing progress without creating unrealistic expectations around immediate full autonomy. 

3. Strategic principles behind the approach 

Workflow-first, not tool-first 

Synopsys places workflows alongside compute and engines or solvers as a primary level of engineering optimization. This is important because customer value rarely comes from running a solver in isolation. It comes from completing a defensible chain of decisions, data preparation, analysis, interpretation and verification. 

Multi-agent specialization 

Complex engineering work is naturally decomposed into specialist roles. A multi-agent architecture can mirror this by assigning planning, setup, execution, checking, interpretation and reporting to different agents. Coordination then becomes a core capability, not an integration afterthought. 

Humans retain accountability 

Engineering autonomy must remain bounded by objectives, constraints, permissions and validation. The engineer provides intent and remains responsible for accepting results. This human-in-the-loop model is essential when outputs influence safety, compliance, product performance or substantial commercial decisions. 

4. Reference architecture for engineering agents 

A credible agentic engineering environment is broader than the language model. The most effective implementations combine six interconnected layers comprising domain knowledge combines domain knowledge, orchestration, deterministic engineering tools and governance. The following layers provide a useful reference model. 

5. Implications for simulation and digital engineering 

Simulation is one of the most compelling application areas for agentic AI. Simulation teams consistently face three challenges: 

  • Growing demand 
  • Limited specialist capacity 
  • Complex workflows 

Agentic systems provide a practical mechanism for addressing all three. 

 

Start with high-toil, low-risk work 

The strongest target use cases are repetitive, bounded and measurable. Examples include preparing standard analyses, generating study variants, monitoring runs, extracting results, applying known checks and assembling draft evidence packs. These applications reduce effort without transferring final engineering accountability to the agent. 

Expert amplification before broad autonomy 

Agentic AI can increase the leverage of scarce specialists. Experts define methods, constraints and acceptance criteria; agents then apply that knowledge consistently across a larger volume of work. This supports simulation democratization while avoiding the assumption that inexperienced users can safely replace specialist judgement. 

Governed access for wider user groups 

As workflows mature, simplified interfaces can make validated engineering capability available to product designers, system engineers and other non-specialists. The workflow must constrain allowable inputs, expose assumptions and provide clear escalation routes when a case falls outside its validated envelope. 

Cross-tool value 

The greatest value may emerge when agents coordinate across requirements, solution architecture, geometry, simulation, optimization, test and reporting. This aligns agentic AI with the digital engineering narrative: simulation becomes part of a connected decision process rather than a late-stage verification activity. 

These use cases create measurable efficiency gains while maintaining engineering oversight. 

Over time, organizations can expand agent responsibilities as confidence, governance and validation maturity increase, building trust incrementally through transparent, repeatable and demonstrably reliable outcomes. 

The goal is not autonomous simulation. The goal is enabling expert knowledge to be applied more consistently across a larger volume of engineering work. 

6. A pragmatic adoption roadmap 

Organizations should resist the temptation to start with autonomy. The most successful transformations begin with workflow understanding and engineering governance. 

Recommended success measures 

  • Engineering cycle time and queue reduction 
  • Expert hours released for higher-value work 
  • First-time-right rate and reduction in avoidable rework 
  • Consistency of method application and evidence capture 
  • Percentage of cases completed within the validated workflow envelope 
  • Number and severity of human interventions or escalations 
  • Reuse of workflow components across projects, teams or customers 
  • Commercial value created through services, software, support or license consumption 

 

The organizations that move fastest will not necessarily be those pursuing the highest levels of autonomy. They will be those developing the strongest trust architectures. 

7. Strategic opportunity 

Synopsys provides a strategic product portfolio, yet some customers may need support translating possibilities into their own engineering environment. Bridging the gap between ambition and operational deployment by combining simulation expertise, workflow design, integration and customer discovery is a must. 

Build the winning team 

  • Advisor: gauge readiness, select use cases and define governance and value measures 
  • Workflow architect: convert expert practice into repeatable, testable and maintainable processes 
  • Integrator: connect agents to engineering applications, data, cloud execution and enterprise controls 
  • Assurance partner: define validation envelopes, traceability and human approval mechanisms 
  • Capability builder: train teams to operate, supervise and improve agent-enabled workflows 
  • IP manager: develop reusable orchestration, connectors and domain workflow components, where cross-customer reuse is credible 

8. Risks and guardrails 

Conclusions

The strongest aspect of this approach to agentic AI is the clarity of direction. This paper describes the development of agentic AI workflows as the staged evolution of engineering work, gives multi-agent workflow orchestration a central role and avoids reducing the opportunity to a conversational conversation. The model also prompts important customer questions because agents will increasingly consume tools, data and engineering services alongside human users. 

The objective is to help customers identify where high-toil work, scarce expertise and repeatable methods create a credible case for governed agents. Early engagements should deliver measurable workflow improvements while building the architecture, controls and customer confidence required for greater autonomy later. 

This positions agentic AI as a practical extension of simulation democratization and digital engineering: not the replacement of engineering judgement, but the scalable application of expert knowledge through controlled, traceable and increasingly capable workflows. 

Will Marsden

Will Marsden

Field Chief Technology Officer
ajax-loader-image