How Agentic AI Is Reshaping Chip Design

Artificial intelligence is already changing how engineers work. The next shift is even more significant: AI is moving from a helpful assistant for isolated tasks to a set of connected, agentic workflows that can help solve complex engineering problems across the design process.

That is where the real transformation begins.

For years, the semiconductor industry has managed rising complexity by building deeper expertise and better tools, but the scale of that complexity has continued to grow. Modern chip design spans architecture, verification, physical design, packaging, thermal considerations, software optimization, and more. No single engineer can hold all those variables in mind at once. That is why AI is no longer just interesting for engineering work. It is becoming essential.

From Prompts to Workflows

A lot of early excitement around generative AI centered on one-shot interactions: ask a question, get an answer. That can be useful, but it only goes so far.

What is changing now is the rise of agentic frameworks that can chain tasks together, critique outputs, iterate, and improve results over time. Instead of generating a single response, these systems can participate in a workflow. They can refine, evaluate, and move work toward a better outcome.

That matters because engineering is rarely a one-step process. High-value work comes from repeated cycles of generation, analysis, correction, and optimization. Agentic AI is a much better fit for that reality than a simple chatbot model.

Why This Matters in Semiconductor Design

Chip design is one of the clearest examples of where this shift can matter. Engineers write design code, verify behavior, translate designs into physical implementation, and prepare products for manufacturing. Each stage has its own tools, data, and quality requirements. Each also produces huge volumes of outputs that must be understood and acted on quickly.

Take debug and triage as one example. In a typical design flow, regressions can involve millions of simulations, with thousands of issues that need to be bucketed, prioritized, and resolved. Historically, much of that work was manual. With agentic workflows, AI can help triage those issues, group similar problems, and accelerate resolution so teams can address more of what is found in a given cycle.

That is not only a productivity gain. It can also improve quality.

The same principle applies in other parts of the flow. AI techniques can support tasks such as timing closure and other forms of optimization that once required large amounts of manual effort. The goal is not to replace engineering rigor. The goal is to extend it.

Productivity Alone is Not Enough

In semiconductors, speed only matters if quality remains uncompromised. A chip must be right. Any AI-enabled workflow still must operate within strict validation and verification processes. Established tooling, quality checks, and deep domain expertise remain critical.

The best outcomes come from combining agentic AI approaches with proven engineering methods and the specialized capabilities of the broader ecosystem.

That is why I see Applied AI in engineering as an architectural challenge, not just a tooling decision. You need to think carefully about how workflows are designed, how agents are connected, how outputs are evaluated, and where human oversight is applied. These are not plug-and-play systems. They require real engineering.

Breaking Through Silos

One of the most promising aspects of agentic AI is its ability to connect work across traditional domain boundaries. Engineering teams often operate in specialized areas for good reason, but when data, workflows, and insights can move more effectively across those boundaries, optimization can happen at a broader level.

That creates opportunities for better co-optimization across design domains and, over time, across software and hardware as well. This is where things become especially interesting. The industry is moving toward a future where optimization happens not only within one phase of design, but across the full stack and eventually at the system level too.

The Hard Part is Not Just Technical

There is a tendency to talk about AI adoption as if it is mainly a tooling problem. It is not. It is also a people and process challenge.

The pace of change is extraordinary. When engineers begin adopting AI-native techniques, progress can accelerate unevenly across teams. Some groups move faster than others. New ideas emerge quickly. Established ways of working are challenged. Even when the momentum is positive, that can create real friction.

That is why enablement matters. Teams need the language, skills, and confidence to understand what these systems are, how to use them, and where they fit into real engineering workflows. In many cases, the bottleneck is no longer compute or tool access. It is whether people know how to rethink the work itself.

Speed is the New Advantage

One of the clearest lessons from this moment is that speed increasingly matters as a competitive advantage. However, speed does not simply mean doing the same work faster. It means learning faster, iterating faster, finding more issues earlier, and gaining confidence sooner. It means bringing more of the right work into the schedule without sacrificing quality.

That is why companies that learn how to apply agentic AI effectively will have an advantage. Not because AI replaces deep expertise, but because it amplifies what expert teams can do when they build the right workflows around it.

What Comes Next

We are still early. The tooling is improving rapidly, and the engineering patterns for Applied AI are still being invented in real time, but the direction is clear. Agentic AI is moving from experimentation into practical engineering use, and its impact will extend well beyond point solutions.

Over time, we are likely to see more autonomous workflows, better cross-domain optimization, and faster progress from concept to high-quality product.

This is not just about chip design. The broader lesson is that when work is highly complex, iterative, and distributed across specialized teams, agentic AI can unlock a new level of coordination, productivity, and quality.

For a deeper conversation on these ideas, Mark Papermaster and I go into more depth in the latest episode of Advanced Insights.

About Alex

With AMD for more than two decades, Alex Starr is a Corporate Fellow leading Applied AI at AMD. He is helping shape how agentic AI transforms chip design and engineering workflows. Drawing on his experience in hardware emulation, verification, and pre-silicon validation, Alex works at the intersection of AI, engineering productivity, and high-quality silicon design for next-generation compute.

Promotion to AMD Corporate Fellow is an honor bestowed only to the most accomplished innovators; there are currently less than 15 at AMD.