IN Brief:
- ADS 2027 uses Model Context Protocol servers to connect compatible AI assistants with RF design tools.
- Engineers can record design workflows and generate Python scripts for reuse by colleagues or AI agents.
- Keysight simulation remains inside the workflow so generated circuit changes can be evaluated against engineering models.
Keysight Technologies has added agentic AI functions to Advanced Design System 2027 and RF Circuit Simulation Professional, allowing external AI agents to interact directly with RF design software, perform defined engineering tasks and invoke circuit simulation as part of the same workflow.
The connection uses Model Context Protocol servers that expose documented software functions to compatible AI assistants and large language models. An engineer can issue a request in natural language, after which the agent can select available tools, act inside ADS and use simulation results to decide whether another design iteration is required.
That arrangement separates language model reasoning from the electrical calculations used to judge the circuit. The model can interpret the request and choose a sequence of actions, but Keysight’s simulation engines still calculate behaviour from device models, network parameters and operating conditions rather than relying on the model to infer RF performance from text or graphics.
ADS 2027 also records engineering workflows and can convert graphical activity into Python scripts. Repetitive setup, simulation or verification procedures can therefore be captured once and reused, while code representations give an agent a structured description of the design steps instead of forcing it to interpret a schematic image each time.
RF and microwave circuits are particularly sensitive to small changes in component values, physical geometry, impedance and coupling. A plausible change generated from natural language can still alter gain, stability, matching or electromagnetic behaviour in ways that are difficult to predict without a suitable circuit or field model, so simulation remains the point at which an automated proposal is tested against the design constraints.
The MCP interface limits an agent to functions that Keysight has exposed and documented for the software environment. That is different from giving a model unrestricted control of a desktop application because the available actions can be defined, named and called consistently, making the resulting workflow easier to reproduce and inspect.
With the agent able to alter a design parameter, run the relevant analysis and inspect the result through the same interface, simulation can form a closed iteration loop that explores a larger design space without a manual command between every step. The quality of that search still depends on the specifications, models, constraints and optimisation targets given to the system.
RF Circuit Simulation Professional 2027 extends the same approach to programmatic simulation and optimisation. Its Python interface can connect native RF analysis with external optimisation routines and machine learning frameworks, while the 2027 version also adds reinforcement learning based optimisation and a Python Simulation Model Interface for surrogate models.
Recorded workflows can also preserve how an experienced engineer configures an analysis. Company specific model locations, test conditions, parameter sweeps or verification steps can be encoded into reusable procedures, reducing repeated setup while keeping the underlying engineering choices visible in code rather than hiding them inside a free form conversation with an AI model.
Synopsys has introduced AgentEngineer for autonomous engineering workflows, while OpenAI and Synopsys are developing a specialist semiconductor design model, placing Keysight’s work within a wider move towards agents that can act inside established engineering software. Keysight’s implementation concentrates on RF, microwave and related workflows where established simulation engines can provide a formal check on the actions chosen by an agent.
Poorly defined objectives remain a direct failure mode because an agent can run the wrong analysis repeatedly, optimise towards an incomplete target or select a result that meets the simulated goal while violating a constraint that was never included. Model quality, boundary conditions, convergence settings and engineering review remain part of the workflow even when more of the execution is automated.
The MCP servers, workflow recording and Python script generation are available in ADS 2027 and RF Circuit Simulation Professional. Their role is to let AI agents operate through defined engineering functions while keeping circuit simulation between an automated design change and any claim that the resulting hardware will perform better.
