Designing printed circuit boards (PCBs) is a task requiring precision, attention to detail, and a deep understanding of electronics. Traditionally, this process was performed by engineers manually or with the help of specialized software, where autorouting often required significant rework. Now, it seems a new player has emerged on the horizon – artificial intelligence. OpenAI has introduced its GPT-6 Astra model, claiming it can fully design a PCB in KiCad, starting from a finished schematic. The AI's output is a fully routed board that, in theory, can be sent directly to production. This raises a natural question: how serious is this breakthrough, or is it just another demonstration of capabilities working under ideal conditions? To test these claims, EEBench conducted an evaluation. It's worth recalling that in 2024, attempts to use large language models for board design showed that manual work remained more effective. EEBench is a benchmark for electronics engineers designed to assess how well AI agents handle real-world tasks in hardware engineering. Creating and routing a PCB is just one stage of this complex process. According to the test results, GPT-6 Astra scored 69.3% (+/- 10%) on the EEBench benchmark. This result is comparable to the performance of the Claude Opus 5 model, with GPT-6 Astra being cheaper and faster. It should be noted that the EEBench developers also created Atopile, a PCB design system for KiCad focused on working with AI agents. Despite the achieved results, it's still a long way before hardware engineers can comfortably retire. Rapid creation of PCB prototypes using AI can be a convenient acceleration, but for final projects, especially one-off designs, maximum reliability is required. Such boards must account for all edge cases, undergo multiple checks by independent engineers, and meet strict testing standards before production launch. For now, GPT-6 Astra is more of an assistant for accelerating routine tasks than a complete replacement for human expertise.