AI models can both speed up and help advance science, but can they be trusted? An AI agent can produce convincing math and working code but might alter the underlying science, leaving researchers with results that look right but aren’t.
That’s where quantum software startup Haiqu, a member of the World Economic Forum’s Innovators community, comes in. The company, which splits its operations between the U.S. and Europe and does R&D in Ukraine, has developed a team of AI agents built specifically for scientific exploration and designed to catch mistakes.
Haiqu’s AgenticOS, which was launched on October 7, aims to help researchers design an entire project—from the initial question to an experiment that’s ready for quantum hardware—and assign multiple teams of AI agents to validate and confirm the research findings during each step of the process, says the company. The human team overseeing the experiment can inspect each assumption and conclusion to determine how to proceed.
AgenticOS builds on the Haiqu platform announced in May. A researcher starts with a question, and the system assigns teams of specialist AI agents to review the literature, work through first-principles mathematical derivations, assess quantum feasibility, and check the results. Haiqu’s AgenticOS shows its investigative process as a map of connected research tasks. Results are checked against classical baselines, tests, critic agents, and human review, with formal verification. Scientists decide what gets locked, which tasks run on their own, and where a human signs off, Mykola Maksymenko, co-founder and CTO, said in an interview with The Innovator.
In a recent chemistry study accepted at the NeurIPS 2026 workshop on AI for Scientific Discovery in Paris, Haiqu demonstrated the need for its platform by giving leading AI agents the same research brief, with and without AgenticOS. On their own, four out of 10 agent runs ignored an explicit setup instruction, resulting in widely different answers, says Maksymenko. Using Haiqu’s AgenticOS, the same models held to the approved protocol from start to finish and produced a simulated quantum result within approximately 3.4% of the exact diagonalization reference.
Making Quantum Computing Work for Science and Corporate R&D Teams
Haiqu builds software that sits between quantum applications and quantum hardware. Its premise is that organizations need not wait for fully fault-tolerant machines. Today’s noisy, error-prone quantum processors can already do useful work if the software layer is smart enough. Investors backing Haiqu frame it the same way: quantum computing needs to prove a commercial edge over classical computing, and software is what will get it there.
Quantum software—the space Haiqu competes in—is a crowded and well-funded sector. What makes Haiqu different is that it has created an operating system for quantum applications, says Maksymenko.
Haiqu’s platform has three layers. AgenticOS uses teams of specialized AI agents to turn a research question or business problem into an experiment ready to run on quantum hardware. The Haiqu SDK then optimizes that experiment for noisy hardware through data loading, circuit compression, and error mitigation. Haiqu Runtime manages execution on simulators and on quantum processors from IBM, AWS, IonQ, IQM, and others. The startup’s software is hardware-agnostic but also hardware-aware, meaning it can tailor circuits to the quirks of each type of machine rather than treating all processors alike.
AgenticOS, the SDK, and Runtime work in tandem and together form an operating system similar in nature to early operating systems used in classical computing, says Maksymenko, who previously worked as a quantum researcher at the Max Planck Society and the Weizmann Institute of Science. “Our OS makes it easier for engineers,” he says. “They don’t need to know how the hardware works, so they can concentrate on algorithmic innovations tied to deep research applications.”
Haiqu says its hardware-agnostic software can run applications with up to 1,000x savings in cost and time. “What would take hours on the quantum computer with some of our competitors’ technology, we can do in 30 seconds,” says Maksymenko, who co-founded the company in 2022 with CEO Richard Givhan, a Stanford-trained engineer. “This will democratize access to quantum computers, speeding up the rate organizations can innovate new algorithms and apply quantum computing to new verticals,” he says.
“This is the dramatic moment for all the computational sciences,” he says. “A few weeks ago, we saw how AI impacted mathematicians. Now with this system we are going to see a similar moment in physics.” Turning a research idea into a real experiment can take years of specialized expertise, he says, and AgenticOS will absorb more of that labor, so researchers reach results faster without losing accuracy.
Haiqu’s business model is subscription-based. It charges an additional fee for co-designing algorithms for its customers. It is targeting the automotive, financial services, energy, and pharmaceutical sectors but has only recently released its commercial offerings.
AgenticOS is included with the Haiqu operating system and is now available to enterprise R&D teams, says the company. Academic researchers can apply for free access to the Haiqu platform, including AgenticOS, through the startup’s Academic Program. Participating labs get the Haiqu SDK and can also apply for AWS credits to cover classical and quantum computing workloads.
Competitive Landscape
Competitors fall into four categories: full-stack development platforms; error suppression and mitigation specialists; application and algorithm companies; and the hardware companies’ own software. Tel Aviv-based Classiq is the closest platform rival. Sydney-based Q-CTRL directly overlaps with Haiqu’s SDK and Runtime performance layer and is natively integrated with IBM and IonQ. Tel Aviv-based Qedma competes on squeezing results out of noisy hardware and is distributed via IBM’s Qiskit Functions catalog. Germany’s Kipu Quantum overlaps on circuit compression and the combination of AI and quantum. Haiqu is small and modestly funded compared with Classiq, which has raised more than $200 million, and Q-CTRL, which has raised around $130 million.
Maksymenko says Haiqu’s main differentiators from competitors revolve around three things. It is hardware-aware across multiple machine types rather than tied to one vendor or type of quantum computer. Haiqu brings performance-enhancement tools into a modular stack rather than a single black box and combines subroutines like data loading or quantum machine learning with execution orchestration. With AgenticOS, it has staked an early claim on verified AI agents for quantum research, emphasizing trust and accountability rather than simple chat-like assistance.
The company, which has raised a total of around $16 million in preseed and seed funding, counts Toyota Ventures among its backers. It plans to raise a Series A round of venture capital in the near future.
