If you haven't opened a PCB design tool this year, you might not know yet — the EDA industry is undergoing a transformation from "tools" to "agents."
In July 2026, Cadence officially launched AuraStack AI Super Agent — claiming to be the industry's first agentic AI platform for PCB and advanced packaging design. Around the same time, Xpeedic and Lenovo jointly unveiled an end-to-end EDA Agent at DAC 2026, becoming the only Chinese EDA implementation showcased at the conference.
This is not an "autorouter 2.0." This is the first time EDA tools can understand design intent, autonomously decompose tasks, and collaborate across tools — rather than just passively executing engineer instructions.
The Bottleneck of Traditional EDA: Siloed Tools, Sequential Workflows, and "Throw-Over-the-Wall"
In the traditional workflow, hardware engineers spend about 65% of their time not "designing" — but switching between tools, waiting, and coordinating.
A typical scenario: the layout engineer finishes the board and "throws it over the wall" to the simulation engineer for signal integrity analysis. SI finds problems, the layout gets revised, and the process repeats. Thermal simulation finds more problems, more revisions, more waiting — each iteration taking days or even weeks. Electrical, thermal, mechanical, and cost considerations belong to different departments, tool chains don't communicate, and conflicts only surface at final tape-out.
Michael Jackson, Corporate Vice President of R&D for System Design and Analysis at Cadence, put it bluntly: the bottleneck for next-generation AI infrastructure "is no longer just the chip itself, but the system — connection, power delivery, and thermal management." And these are precisely the areas where traditional EDA is weakest and most fragmented.
What Agentic AI Does: From "Automation" to "Autonomous Engineering"
At its core, AuraStack is not "a smarter autorouter." It's a multi-agent collaborative system that can receive engineering goals, devise plans, invoke multi-physics simulation engines, and return optimization results.
In one public demonstration, an engineer asked AuraStack to "optimize BOM cost." The agentic AI first identified the power management IC, automatically constructed the power tree, invoked PSpice simulation — and eventually recommended alternative components that reduced BOM cost by approximately 28%. It then proceeded to complete schematic review, reliability analysis, thermal checks, and even identified hotspots and suggested layout adjustments.
AuraStack is currently at Level 4 autonomy — it can receive goals, invoke tools, and return engineering results, but engineers still interpret results and make final decisions. Level 5, "fully autonomous engineering," would be able to autonomously iterate until design convergence.
Xpeedic + Lenovo: A Practical Implementation of Chinese EDA
Xpeedic's collaboration with Lenovo took a different path — not replacing engineers, but building an end-to-end loop from library creation, placement, DRC, to simulation optimization.
In the actual Lenovo AI PC motherboard project, this AI Agent improved schematic symbol and PCB footprint library automation efficiency by over 50%, and SERDES full-link optimization simulation efficiency by over 80%. Once these four modules are connected as a closed loop, the engineer's role shifts from "execution" to "review" and "decision-making."
What This Means for Hardware Engineers
You No Longer Need to "Know How to Use the Tools" — You Need to "Know How to Judge Results"
In the past, proficiency with EDA tools was a core skill. Now, AI is taking over these operations — faster, more accurate, and without mistakes.
An engineer's value is shifting from "how to route" to "why route this way" — understanding design intent, setting the right constraints, evaluating whether AI outputs are reasonable, and judging whether manufacturability is feasible.
"Design First, Simulate Later" Is Becoming "Design as Simulation"
Traditional workflow: design → simulate → revise. Agentic AI brings multi-physics simulation forward into the design process — electrical, thermal, and mechanical constraints are verified in real time during early design stages. This means far fewer revisions and significantly faster design convergence.
Manufacturing Requirements Are Getting Tougher
AI-generated designs tend to be more "aggressive" — finer trace widths, more complex via structures, thinner dielectrics. This places higher demands on PCB manufacturers' process capabilities. Without LDI, tight impedance control, and advanced material capabilities, AI-generated designs might not be manufacturable.
Conclusion
Agentic AI isn't "replacing engineers." It's taking over the execution layer and pushing engineers up to the decision-making layer.
In the past, EDA was a toolbox. Now, AI is turning it into an engineering agent that can "understand" design intent. This transformation doesn't wait for anyone's permission — it's already rolling out across the industry in 2026.
If You're Designing High-Speed PCBs
Whether your design comes from traditional EDA tools or AI-assisted generation, AnyPCBA's engineering team provides DFM/DFA design reviews to help you identify manufacturability issues before fabrication — more important than ever as AI accelerates design iteration cycles.
Top comments (0)