Keysight connects AI agents to RF design tools

Keysight Technologies is allowing AI agents to work directly inside parts of its Advanced Design System, giving them controlled access to RF and microwave design functions through Model Context Protocol interfaces and Python automation.

The ADS MCP server connects a supported AI client to an active ADS session. Through that connection, the client can execute Python code and retrieve product documentation, allowing an engineer to assign a defined task without manually translating every instruction into separate commands inside the design application.

RF circuit development depends on numerical models rather than plausible text output. A change to component values, bias conditions, transmission line geometry or matching networks alters electrical behaviour that has to be calculated against frequency, stability, gain, efficiency, impedance and other design requirements. An AI agent can propose or execute a change, but ADS simulation still produces the engineering result used to judge it.

ADS 2027 can also turn schematic and layout activity into Python code. Engineers can record a sequence of operations, preserve it as a macro and reuse the resulting script for repetitive work such as changing parameters, launching simulations and extracting results. The sequence originates in work performed inside the engineering environment, so it can preserve an established design method rather than relying on an agent to construct one from an open prompt.

A typical optimisation cycle may require hundreds of parameter changes before a circuit reaches the required performance. Automating the mechanical part of that process allows the software to alter a variable, run the relevant simulation, collect the result and compare it with a defined target before the next iteration. Engineers still set the constraints and decide which results are acceptable.

Errors in device models, boundary conditions, material properties, operating temperatures or simulation settings will carry through the automated process. An incorrect assumption can therefore be repeated across many iterations more quickly than it would be in a manual workflow. Agent access changes how the analysis is executed, but the models and constraints still determine whether the result has engineering value.

Python scripts and recorded macros become part of the design record once they influence a circuit or layout. Version control has to preserve the script, model set, simulation configuration and design state that produced a given result, otherwise another engineer may be unable to reproduce the analysis or establish why a change was accepted.

Conventional electronic design automation already requires that level of traceability, but agent access increases the number of operations that can be performed without direct manual input. Teams need a record of which actions the agent executed, which data it used and which design revision was active at the time. Without those records, a sequence of automated changes can become difficult to reconstruct during technical review.

Radar, electronic warfare, secure communications, navigation and satellite electronics all rely on RF and microwave design methods of this kind. Keysight has not tied the capability to a named military programme or customer, but the workflow applies wherever ADS is used to design and verify circuits operating at radio and microwave frequencies.

AI client access to a live design environment also introduces additional controls around project data and script execution. Permissions determine which files and functions the client can reach, while the organisation using the software has to decide where prompts, generated code and design information may be processed. Sensitive defence programmes may impose restrictions on models, networks and data movement that sit outside ADS itself.

ADS 2027 extends programmable access beyond the MCP connection. Keysight has expanded Python support across schematic and layout workflows and added related automation functions to 3D interconnect design. EDA Chat and EDA Copilots provide further AI interfaces, including deployment options intended to keep more processing inside controlled engineering environments.

Parameter changes, established simulations and result extraction are necessary parts of RF development, but they do not always require an engineer to perform every command manually. Moving those operations into Python workflows can increase the number of design variants evaluated within the same development period while keeping the underlying simulation method unchanged.

Higher iteration rates can create additional work if results are not filtered against explicit requirements. A large set of automatically generated candidates still has to be reduced to designs that satisfy electrical, physical and programme constraints. Clearly defined acceptance criteria allow unsuitable configurations to be rejected during the automated sequence instead of being passed forward for manual assessment.

Physical implementation adds effects that may not appear in a simplified circuit model. Layout, packaging, connectors, neighbouring structures, material behaviour and manufacturing tolerances can alter RF performance after the initial circuit analysis. Electromagnetic simulation, prototype measurement and hardware testing therefore remain necessary as the design moves towards production.

Keysight’s agent interfaces operate within the earlier design and simulation stages of that process. They can automate parameter changes, software operations and repeated analysis, but they do not remove the engineering evidence required before an RF circuit is released. Their practical value will depend on whether teams can examine more viable designs while retaining reproducible models, controlled scripts and traceable simulation results.