Keysight Technologies has added agentic AI capability to its radio-frequency design software, allowing compatible external assistants and large language models to invoke engineering tools while Keysight simulation checks the designs they produce. The functions are being introduced through Advanced Design System 2027 and RF Circuit Simulation Professional 2027, where Model Context Protocol interfaces, Python integration and workflow automation connect conversational requests to defined RF design operations.
The architecture separates orchestration from calculation. A language model can interpret an engineer’s instruction, decide which documented function to call and move through a design sequence, but the electrical behaviour of the circuit is still calculated by the specialist simulation software. That division is important because plausible generated code or a convincing conversational explanation provides no evidence that a design actually meets requirements for gain, impedance, stability, linearity, noise or other RF performance measures.
Model Context Protocol gives the agent a structured route into the application rather than asking it to infer how specialist design software works from an unrestricted interface. Tools can be exposed for defined tasks, allowing the model to request a simulation, change parameters or retrieve results through documented functions. Keysight is also supporting Python-based automation, giving engineering teams the option to combine their own scripts, models and optimisation methods with the commercial simulation environment.
Repeated setup and iteration are obvious targets for that automation. RF development can involve sweeping component values, comparing topologies and rerunning analyses under different operating conditions long after the engineer has decided which calculations are required. An agent can execute more of those deterministic steps without removing the need to specify the objective, making it possible to explore a larger set of alternatives during the same development period.
Recorded workflows can also preserve parts of the method used by experienced engineers. Capturing a sequence does not reproduce the judgement behind every design decision, but it can standardise recurring tasks and reduce the amount of time colleagues spend rebuilding a known setup. Keysight’s argument is therefore less about replacing RF expertise than about allowing specialist knowledge to be expressed through reusable operations that humans and agents can invoke consistently.
The limitation appears when the design objective is incomplete. An automated optimisation can improve the variables it has been told to measure while creating problems in areas omitted from the model, including thermal behaviour, manufacturing tolerance, packaging or system integration. Running more iterations merely accelerates the wrong search if the constraints have been defined badly, which keeps engineering review central even when much of the workflow around simulation becomes automated.
RF Circuit Simulation Professional 2027 adds native MCP support alongside reinforcement-learning optimisation and Python interfaces for external surrogate models. ADS 2027 extends AI and automation across a broader environment covering RF and microwave circuits, high-speed digital systems, power electronics, quantum electronics and photonics, giving Keysight several points at which an agent can coordinate established numerical tools.
The open interface also gives customers some choice over the AI model attached to the engineering software. Organisations with internal assistants or approved enterprise models can potentially connect those systems rather than sending designs through one vendor-selected service, an important distinction where schematics, simulation models and process information form part of commercially sensitive intellectual property.
Security and governance will therefore sit alongside productivity in any real deployment. Connecting an agent to design tools creates a path through which proprietary engineering data may be interpreted or transferred, so companies have to decide where the model runs, what information it receives and which actions it is permitted to perform. The software can expose deterministic engineering functions; it cannot decide the organisation’s acceptable risk boundary for AI access.
Keysight has made the new capabilities part of commercially available 2027 releases, moving agentic RF design beyond a demonstration. The useful benchmark will be whether engineering teams genuinely complete more viable design iterations while preserving review discipline, because faster interaction with a simulator is valuable only when the model, constraints and engineer remain aligned closely enough for the additional output to improve hardware rather than simply multiply possible answers.
