What AI Agents Can Do Inside MATLAB and Simulink - Tianyi Zhu | Podcast #173
Tianyi Zhu from MathWorks explains the key differences between AI agents and chatbots, highlighting how agentic AI acts as a powerful amplifier for engineers. The discussion covers practical use cases in MATLAB and Simulink, measurable ROI in automotive workflows, and strategies for safely integrating non-deterministic AI into high-stakes engineering environments.
The defining characteristic of an AI agent, compared to a standard generative AI chatbot, is its ability to use tools and take direct action on behalf of a user. A chatbot, like ChatGPT, can provide detailed steps on how to build a Simulink model, but it cannot create the model itself. An AI agent, however, can be connected to the engineering environment via a toolkit. It can then interpret requirements, select the appropriate blocks, wire components, run simulations, and deliver a functional model, sometimes complete with simulation results.
AI as an Amplifier for Engineers, Not a Replacement
The narrative that AI will replace engineers is misguided. Instead, AI is set to amplify their creativity and productivity, enabling them to achieve more than ever before. AI agents take on the more “mechanical” parts of the job, freeing up engineers to focus on higher-level, creative tasks:
- Exploring new ideas and experimenting more frequently.
- Widening the design space exploration.
- Focusing on core creative work, such as deciding what to build, how to verify its functionality, and determining when a product is ready to ship.
This shift allows engineers to concentrate on the aspects of their work that require human ingenuity and critical judgment.
Integrating Agentic AI into Engineering Workflows
Adopting AI in engineering is not an “all or nothing” proposition but rather a “when and how” question. A thoughtful approach is crucial for success.
- Start Small and Scope Carefully: Instead of trying to automate an entire system in one go, engineers should identify small, repetitive, and manual parts of their workflow to automate first. This allows for quick wins and iterative learning.
- Maintain Traceability: In engineering, documentation and traceability are paramount. AI can assist in maintaining the link between requirements and engineering artifacts, and it can help generate the necessary documentation to demonstrate how a system was built and validated.
- Learn by Doing: The most effective engineers start with a small part of their work, iterate with the AI tool, and even ask the AI to reflect on its own thinking. This process turns learnings into reusable AI skills, avoiding the need to repeat prompts.
High-Impact Use Cases and Measurable ROI
Agentic AI is already delivering significant productivity gains in several key areas of the engineering workflow.
- Issue Resolution: In a collaboration with a major European automotive manufacturer, applying agentic AI to the issue investigation workflow resulted in a 20% time saving. The AI agent handles verification and investigation tasks, while the human engineer remains in the loop to review the work and guide the process.
- Simulation Automation: The process of running simulations involves many manual steps, such as connecting data, configuring settings, and visualizing results. An AI agent can automate these tasks, shortening the time from simulation to insight and allowing engineers to focus on analyzing outcomes to drive better decisions.
- Model Generation from Requirements: Agents can now take a set of written requirements and generate a functional, simulating Simulink model. This eliminates the manual effort of placing blocks, routing signals, and scripting, dramatically improving productivity.
- Model Refactoring: AI agents are highly effective at performing systematic cleanups, upgrades, and documentation on legacy artifacts. This helps bring older models in line with current best practices before new design work begins.
Navigating High-Stakes Engineering with Non-Deterministic AI
A primary concern for engineers is sacrificing quality for speed. The key to using non-deterministic AI in high-stakes, safety-critical applications is to combine it with deterministic tools.
AI still needs to rely on proven and deterministic tools like MATLAB and Simulink to help the engineer achieve the best results.
An AI trying to perform a simple calculation like adding two numbers might burn through tokens and still get it wrong. A more effective approach is for the AI to use a calculator. Similarly, in engineering, the agent should use MATLAB and Simulink as its deterministic “calculators” to ensure accuracy and reliability.
Testing is a specific area where this is critical. An AI can fall into “reward hacking,” writing tests that are designed to pass its own potentially flawed code. A better approach is to:
- Use deterministic tools to generate test cases based on mathematically proven methods.
- Use AI as a “second pair of eyes” to help identify hard-to-find edge cases and draft a more comprehensive testing strategy.
Advice for Engineers in the Age of AI
To stay relevant and become a “10x engineer,” a combination of deep domain knowledge and practical AI skills is essential.
- Deepen Your Domain Expertise: Whether in controls, signal processing, or electronics, deep domain knowledge is crucial for effectively guiding an AI agent and validating its outputs.
- Engage in Real-World Projects: Practical experience teaches you how to specify goals, write clear requirements, and know when an agent is on the wrong track. It’s about learning to work effectively with an agent.
- Get Started Now: A concrete first step for engineers is to download the Simulink Agentic Toolkit from GitHub. They can start with a small task, like asking an agent to explain a subsystem, and gradually move to more ambitious tasks like building a new component from requirements.