Keysight puts repeatable RF workflows at the centre of agentic design
Keysight has added Model Context Protocol interfaces, macro recording and Python script generation to ADS 2027 and RF Circuit Simulation Professional. Announced on 6 October, the features allow an engineer's chosen AI agent to work directly with the design tools.
The workflow connects natural-language instructions to defined operations. An agent can create or change a circuit, configure an analysis and run the simulator, then use the result to guide another iteration. Recorded macros give it an existing sequence of engineering actions to follow.
That changes the unit of automation. Instead of asking an assistant to describe how to run a simulation, the engineer can give it access to the operations required to execute the work. A circuit represented in code also supplies something more explicit than a screenshot of a schematic.
A protocol connects tools, not engineering judgement
MCP defines how an AI application exchanges context and invokes capabilities exposed by a server. The protocol's own documentation is clear that it does not determine how the application uses its language model or manages that context.
The engineering value therefore comes from the operations behind the interface. Documented tools can constrain the action to a recognised task, such as configuring an analysis or changing a parameter. They also produce inputs and outputs that can be retained with the project.
A repeatable simulation needs more than a prompt history. It needs the circuit revision, models, analysis settings and results. Preserving those together allows another engineer to distinguish a different design from a different simulation setup. This is an engineering inference from the workflow, rather than a claim that MCP itself supplies design traceability.
A concrete RF example
Keysight's published tutorial demonstrates a DC current-voltage sweep using an existing transistor design. The agent accesses documentation, configures the simulation and creates plots through Python. The exercise is then repeated in Nexus to compare the two environments.
That is a useful example because success is observable. The generated configuration can be inspected, the sweep can be rerun and the resulting curves can be compared. The task has a defined input and a recognisable engineering output.
Keysight also cites Sphere Semi's separate ADS 2026 case study. The company used code-defined circuits, Python APIs and RF circuit/electromagnetic co-simulation to evaluate hundreds of RF switch and tunable-filter layouts. Keysight reports first-pass RFIC success, approximately 6dB better switch isolation and more than 30% filter-area reduction.
Those are results reported for that particular workflow. They are not a general performance promise for every ADS 2027 agent or circuit.
Capture a method that survives the conversation
Macro recording adds another layer: the experienced engineer's sequence becomes a reusable project asset. It can encode how a team sets up a recurring task, rather than leaving the whole procedure to be reconstructed from natural language each time.
The simulator still evaluates the supplied models and conditions. An apparently successful run can answer the wrong question if the agent selects an unsuitable model or omits a required operating corner. A useful deployment consequently makes the specification and permitted design changes as explicit as the simulation actions.
The practical opportunity is a faster loop through well-defined engineering work, with a record that another person can follow. The design team can then assess the resulting circuit and evidence rather than treating a convincing conversational explanation as the deliverable.
Sources
Keysight announcement; MCP architecture; Keysight workflow tutorial; Sphere Semi case study.
