Phasecraft uses AI agents to cut SWAPs and CNOTs in quantum circuits

[Highland Canopy]

Phasecraft is also turning to AI to make quantum circuits cheaper. Quantum Zeitgeist reports that the London based company has used AI agents to optimise circuits, with results posted on arXiv

According to the report, the agents cut the number of SWAP gates by up to 90 percent and CNOT gates by up to 73 percent. They also lowered Pauli weight by 23 percent, which matters for simulating physical systems

SWAP gates are needed when qubits that must interact are not physically next to each other on the chip. They add a lot of error and time, so cutting them makes a big difference on real hardware. CNOTs are the main two qubit gate, and they are usually where most errors come from

Phasecraft focuses on near term algorithms, especially for materials and chemistry simulation. Using agents rather than fixed compiler rules lets the optimisation explore options a human might miss

This story lands the same day as Google's Kaggle contest, which is a nice coincidence. Both are betting that AI can find savings in circuits that people have overlooked. The figures are best case numbers from a preprint, so independent testing will matter

Will AI agents become a standard part of quantum compilers? And are these savings likely to hold up on real hardware?

Phasecraft Asks If AI Agents Can Automate Quantum Circuit Optimization?quantumzeitgeist.com