Agentic artificial intelligence is moving from concept to hands-on experimentation in electronic design automation, with both global providers and Chinese EDA vendors stepping into the arena. According to reports from Pandaily and 36Kr Europe, the field is seeing a noticeable acceleration as researchers and companies explore how multi-agent systems could streamline parts of the semiconductor design flow.
This shift is playing out across conference stages, product announcements, and proof-of-concept experiments. 36Kr reports that Xpeedic and Lenovo presented EDA-agent work at DAC 2026, pointing to growing interest in agentic methods across both established toolchains and emerging platforms. Within that, Xpeedic described a staged L0-to-L5 AI-EDA development framework, according to 36Kr, offering a structured way to think about the progression of agent capabilities over time.
Chinese players spotlight agentic EDA while global providers push forward
Both Pandaily and 36Kr describe a recent uptick in activity around agentic AI for EDA, involving global providers and Chinese vendors. Although the contours and maturity of these efforts vary, the shared emphasis is on agents that can coordinate discrete tasks within the design and verification process, backed by iterative workflows that incorporate feedback and domain constraints.
Within China’s cohort, EpicSys and UniVista have emerged with new offerings and positioning statements that speak directly to agentic EDA ambitions. Both sources report that EpicSys released its XEPIC Intelligent System in July. According to the company, the system includes an intelligent verification system and a design system, framing a top-to-bottom approach that aims to combine design intent, tooling, and automated assessment under one umbrella. Product capabilities and how they perform in practice remain as described by EpicSys.
36Kr reports that UniVista released UDA 2.0 in March, and that UniVista described it as an agentic digital-design platform. That positioning underscores how vendors are now using the agentic label to define their roadmaps and to align internal R&D with multi-agent orchestration. As with other claims of product capability in this space, UniVista’s description should be read as the company’s characterization of its system.
Research debuts at DAC 2026 hint at how agentic EDA might evolve
The concerted push to bring agentic concepts into chip design is increasingly visible at academic and industry gatherings. According to 36Kr, Xpeedic and Lenovo presented EDA-agent work at DAC 2026, signaling a desire to position agentic approaches within mainstream EDA discourse. For Xpeedic, the L0-to-L5 AI-EDA development framework reported by 36Kr outlines a staged pathway. While the framework details are proprietary to the company, the label suggests a stepwise approach to increasing autonomy and coordination among agents, tied to how they handle design tasks and interact with traditional tools.
These research debuts do not represent production validation, and neither source claims that any Chinese EDA agent is fully autonomous or proven in real-world deployment. Instead, they reflect an exploratory phase where frameworks, prototypes, and targeted pilots act as scaffolding for future work. That measured pace is consistent with how both reports portray the landscape: an acceleration of interest and activity, alongside caveats about current capability and the distance to mature adoption.
Beyond specific company announcements, the broader context is a hybrid ecosystem where agentic methods are evaluated within established flows. The interplay between tried-and-true toolchains and emerging agents is a recurring theme in both reports, and it anchors the conversation in incremental integration rather than sweeping replacement. That approach resonates with how EDA has historically incorporated new forms of automation, with careful attention to verification, repeatability, and tool interoperability.
Past experiments provide context for today’s ambitions
Both Pandaily and 36Kr note that Kimi K3 completed an earlier autonomous chip-design experiment. Importantly, the reports emphasize that the resulting chip was far from current advanced technology. The episode is best read as context for what today’s agentic EDA momentum may or may not deliver in the near term, not as a product launch or as proof of parity with cutting-edge nodes or flows.
That distinction matters. It underscores how agentic AI in EDA is still building its track record, even as more vendors engage. Each new release or framework introduces hypotheses about how agents might help with partitioned tasks such as synthesis, floorplanning, or verification loops. Yet the performance of these systems, and how they will coexist with established commercial tools, remains an open question in the reporting. Both sources refrain from declaring which implementations are ahead or suggesting that the latest entrants are on par with the current global leaders.
For readers interested in how compute access and lab strategies affect the pace and priorities of AI development in China, EastFrontier has covered the broader landscape in our recent analysis of training choices and availability. That coverage can be explored here and here. These pieces do not alter the facts reported by Pandaily and 36Kr about agentic EDA but provide context on adjacent dynamics that shape the environment in which these tools are developed and evaluated.
Looking ahead, the most tangible signals to watch, based on the reporting, are continued conference appearances, additional staged frameworks, and product updates that attribute capabilities to the vendors themselves. The two sources describe a field in motion, one where Chinese EDA vendors are participating alongside global providers but without claims of production-proven autonomy. EpicSys’s XEPIC Intelligent System and UniVista’s UDA 2.0, each described by the companies, exemplify how the market is beginning to define categories and benchmarks that future announcements will be judged against.
The acceleration highlighted by Pandaily and 36Kr sets expectations for more experiments, demonstrations, and incremental tooling partnerships. Whether agentic approaches ultimately reshape how design and verification are organized will depend on sustained evidence from vendors and users, along with careful attribution of what has been demonstrated versus what remains aspirational. For now, the race is on, and the reporting places Chinese vendors firmly in the pack, contributing frameworks, platforms, and proof-of-concept work that keep the conversation moving.