Haiqu launches AgenticOS, a team of AI agents that plans and checks quantum experiments

[Glassy Falcon]

Quantum software company Haiqu has launched AgenticOS, which uses coordinated teams of specialised AI agents to plan quantum experiments, check results and prepare them for hardware, with human oversight, The Quantum Insider reports. The agents review the literature, derive results from first principles, assess whether a quantum approach is feasible and check outcomes against classical baselines. Critic agents also review the work. The results reported in the article are Haiqu's own

In a chemistry test, four out of ten runs without AgenticOS ignored an explicit setup instruction. With it, the models followed the approved protocol and landed within about 3.4 percent of the exact reference result. In a second test, AgenticOS designed a doped, frustrated Hubbard model experiment on a three by three grid and ran it on an IBM quantum computer. Haiqu reports a 75 percent cut in circuit depth

Finished plans pass to Haiqu's software development kit and runtime, which target hardware from IBM, IonQ, IQM, AWS and others. AgenticOS is available now to enterprise research teams on request, with free academic access. Chief technology officer Mykola Maksymenko said: "AgenticOS takes on much of that labor-intensive work, freeing up researchers." Chief executive Richard Givhan said: "Every lab will soon have access to the same powerful AI."

It is an interesting combination of two of this forum's favourite topics. AI agents are increasingly being used to do scientific work, from OpenAI's maths results to the AI swarms at the Berlin summit that cut the resources needed to break ECDSA. The critic agents and checks against classical baselines are a sensible response to the problem of AI making confident mistakes. Whether they work as well as claimed will need independent testing

Would you trust AI agents to design quantum experiments? And is AI going to speed up quantum research more than quantum speeds up AI?

Haiqu Releases AgenticOS to Plan and Check Quantum Research thequantuminsider.com

[Ethan24]

Human oversight is the key phrase. Agents that do the tedious parts while scientists make the decisions sounds sensible. Agents that replace the scientists would worry me. The best tools make good scientists faster, not unnecessary

[Tara_66]

AI speeding up quantum research is far more likely in the short term than the other way round. Quantum computers are not yet useful enough to accelerate AI

AI is already useful enough to help physicists. The reverse may come later, once fault tolerant machines exist

[HardyBoy13]

Checking against classical baselines should be standard for any quantum result. Too many papers skip it

Building it into the tool could improve the whole field. Many claimed quantum speedups disappear once someone tries a well tuned classical method. Building the check in is a good discipline

[Ronaldo22]

The fact that four out of ten runs ignored an explicit instruction without the system is the most telling detail. Unchecked AI agents are unreliable. Building in critics and checks is the right approach

[Theo50]

Free academic access is a smart move. Researchers will try it, publish results and either validate or debunk the claims

Good for everyone. Universities will stress test it far harder than any marketing team

[Stoat64]

A 75 percent cut in circuit depth on real IBM hardware would be very useful if it holds up. Shallower circuits mean less noise. I would like to see an independent comparison though. Company reported results always need independent checks