Synopsys said on September 28, 2026, that AgentEngineer and the Autopilot Platform had more than 50 engagements, and that it planned general availability by the end of 2026. The count is company-reported; the year-end date is a plan.
Electronic design automation (EDA) software supports the work of turning chip designs into verified layouts. The new agentic tools aim to coordinate multiple steps in that process, from writing register-transfer-level (RTL) code to running verification and helping engineers debug or implement designs.
What the tools do across chip design
Agentic EDA systems coordinate specialized agents and established design tools to carry out multi-step workflows. Depending on the product, those workflows can include generating RTL, creating tests, running verification, sorting through failures, and supporting implementation, analog design or packaging.
Synopsys describes AgentEngineer as a portfolio of long-horizon agents for areas including verification, system validation, implementation, analog, manufacturing, simulation and analysis. Its Autopilot Platform is designed to provide orchestration, skills, memory, telemetry and governance. Synopsys says customers can choose among its own, partner and third-party models, infrastructure and tools.
How the named offerings differ
| Company and product | Described workflow focus | Reported 2026 status |
| Cadence ChipStack | RTL generation, testbench creation, regression orchestration and debug | Early access began in February 2026. |
| Synopsys AgentEngineer and Autopilot | Long-horizon agents spanning verification, system validation, implementation, analog, manufacturing, simulation and analysis | Synopsys reported more than 50 engagements on September 28 and planned general availability by the end of 2026. |
| Siemens EDA: Questa One Agentic Toolkit and Fuse EDA AI Agent | Questa One covers RTL code, lint, clock-domain crossing, verification planning and debug. Fuse coordinates Siemens and third-party EDA tools. | Siemens announced the toolkit in February and launched Fuse in March 2026. In July, it announced self-verifying loops for a forthcoming Fuse update. |
| Empyrean agentic platform | Simulation and layout | Empyrean was developing the platform. |
The comparison is about the work each offering is designed to support—not a ranking. For example, a workflow for RTL verification is a different engineering task from analog layout or packaging.
Cadence’s autonomy levels describe its own framework
Cadence’s five-level framework runs from optimization AI through conversational language models, reasoning and agentic workflows to full autonomy. Those levels are Cadence’s terminology, not an industry-wide measurement standard.
At a June 2026 demonstration, Cadence described its Level 5 scope as bounded to one domain. Cadence executive Rob Knoth characterized most of the company’s super-agent technology as advanced Level 4. The demonstration’s label therefore applies to that bounded domain, not to every Cadence tool or the entire chip-design process.
Cadence vice president Ziyad Hanna discusses the company’s five-level autonomy framework for chip design.
Performance claims use different tasks and baselines
Synopsys claimed that its agentic flow delivered RTL validation up to 50× faster and 20% higher coverage than its own non-agentic flow. Cadence separately reported that one RTL-validation loop ran more than 40 times faster, describing a typical five-week loop reduced to less than a day.
Fujitsu executive Toshio Yoshida reported a 10%–30% productivity boost in RTL code generation using Synopsys Verification AgentEngineer capabilities. The figure applies to code-generation work, not to the same validation task described in the other claims.
These results cover different workflows and baselines, so their multipliers cannot be lined up as a like-for-like speed test. They do not establish which vendor is fastest.
Human approval remains part of described workflows
The agents operate alongside established EDA engines that check design outputs. Synopsys executive Anand Thiruvengadam said crucial approval checkpoints in the company’s workflows remain human-driven. Agents can coordinate and iterate through engineering tasks, while people retain those approval decisions.
