AiCybr

Synopsys AgentEngineer and Autopilot Bring Long-Horizon AI Agents to Chip Design

By AiCybr Editorial · Published September 28, 2026

Synopsys announced AgentEngineer on September 28, a portfolio of long-horizon AI agents designed to execute multi-step engineering workflows across chip and systems development. The agents run on the new Synopsys Autopilot Platform, which provides orchestration, engineering context, reusable skills, persistent memory, telemetry and governance.

The initial scope spans verification, system validation, implementation, analog and mixed-signal design, manufacturing, and simulation and analysis. Synopsys says more than 50 customer engagements are underway, with general availability planned for the end of 2026. That availability date matters: the September announcement describes a platform in customer evaluation and deployment work, with broad commercial availability still planned for later this year.

Synopsys also published performance results from current engagements. The company reports up to 50× faster verification closure, up to 20% higher coverage, a 10–30% RTL code-generation productivity improvement reported by Fujitsu, and up to 2× better token efficiency in selected workflows. These are vendor and customer-reported results from specific engineering workloads; they are useful evidence of current deployments, not a universal benchmark across EDA projects.

What AgentEngineer covers

AgentEngineer combines long-horizon agents that plan complete workflows with narrower task agents that perform individual engineering operations. Synopsys lists use cases including autonomous coverage closure, software bring-up and validation, multi-die 3DIC assembly, power-performance-area closure, analog layout synthesis and migration, mask synthesis, combustion analysis, resonance analysis and signal-integrity work.

The distinction is operationally important for EDA. A verification workflow can require repeated generation, simulation, debug, coverage analysis and modification before a design converges. Synopsys is positioning the long-horizon agent as the coordinator for that loop while existing EDA and simulation tools remain the execution and validation engines.

Earlier demonstrations provide more detail on the verification architecture. At DAC 2026, Synopsys showed an AgentEngineer workflow that decomposes verification goals from specifications, design files, test repositories and user input, then coordinates specialized agents and tools through test-plan generation, coverage closure and debug. Synopsys reported up to 50× faster time to validated RTL and an additional 20% coverage improvement in that demonstrated workflow compared with its stated traditional baseline.

Autopilot is the orchestration layer

The Autopilot Platform supplies the shared infrastructure behind the agents. Synopsys describes four central capabilities: model and infrastructure optionality, engineering context intelligence, security and data controls, and efficiency optimizations through privileged APIs and proprietary engineering context.

Customers can use Synopsys, partner and third-party models, infrastructure, tools and agents. The platform's context layer combines engineering knowledge, reusable skills and persistent memory so an agent can retain workflow state and design intent across longer tasks. Access controls, encryption and runtime guardrails provide the governance layer for customer and partner intellectual property.

Synopsys is also integrating NVIDIA's agent stack. Its earlier autonomous-engineering demonstrations used NVIDIA Agent Toolkit, Nemotron models and the OpenShell runtime. NVIDIA describes OpenShell as a policy-controlled runtime for AI agents; Synopsys is using it as part of the security architecture for long-running engineering agents.

Six engineering domains, with real customer work already underway

The September launch names six broad domains, expanding the scope beyond Synopsys' earlier individual agentic-EDA demonstrations focused on specific design and verification workflows.

Fujitsu says it observed a 10–30% productivity improvement in RTL code generation for SystemVerilog assertions, wrapper modules, parameterized modules and refactoring. Intel is evaluating verification and debug workflows. MediaTek is working with Synopsys on autonomous analog and mixed-signal flows. Samsung is applying the technology to memory-development workflows, while TSMC describes agentic AI work around power-integrity analysis for advanced packaging.

Synopsys and TSMC had separately announced agentic design collaboration on September 23. That work includes automated analog, digital and multi-die workflows, plus an agentic chiplet floorplan co-optimization flow using 3DIC Compiler for TSMC 3DFabric designs. The September 28 Autopilot announcement supplies a common orchestration layer for this wider class of engineering agents.

How to read the performance numbers

The strongest published numbers come from different workloads and baselines. The 50× verification figure is associated with Synopsys' autonomous verification flow; Fujitsu's 10–30% result concerns specific RTL code-generation tasks. They should therefore be evaluated independently instead of being combined into a single platform-wide productivity claim.

The more consequential architectural change is the move from task-level AI assistance toward agents that can maintain context, invoke engineering tools repeatedly and work toward a closure target over a longer execution window. EDA is a particularly relevant test case because design correctness can be checked by established compilers, simulators, formal tools, signoff engines and physical-design systems instead of relying only on an LLM's generated answer.

For engineering organizations evaluating the platform, the practical questions are workload coverage, reproducibility, compute and token consumption, access to proprietary design data, auditability of agent actions, and the degree of human review required before signoff. Synopsys' announced architecture addresses orchestration and governance, while workload-specific evaluation remains necessary for productivity and quality claims.

Bottom line

AgentEngineer and Autopilot consolidate Synopsys' existing agentic EDA work into a platform spanning six engineering domains. The strongest evidence today is the combination of active customer engagements, named deployment partners and measured results from specific verification and RTL workflows. General availability is planned for the end of 2026, so the next useful evidence will be production availability, product-level packaging and broader customer measurements across the announced domains.

Sources

Continue reading

AI Lattice Prompt Brings MCP-Based AI Agents Into the FPGA Design Flow

Lattice Prompt connects AI coding agents to Lattice Radiant through MCP, covering simulation, synthesis, place and route, timing analysis and bitstream generation. Here are the requirements, supported tools and current evidence.

AI Microsoft Copilot Adds Home, Code and Autopilot: Rollout, Runtime and Billing

Microsoft's new Copilot adds Home, Code and Autopilot, with sandboxed app building, persistent cloud agents, Managed Runtime and usage-based billing.

AI NVIDIA Isaac ROS 5.0 Adds Agent Skills, ROS 2 Lyrical and a New GPU Buffer Path

Isaac ROS 5.0 moves NVIDIA’s accelerated robotics stack to ROS 2 Lyrical, adds agent skills, replaces direct NITROS APIs with rosidl::Buffer integration and expands Jetson deployment support.