Strengthening agentic AI with quantum computing

Across industries, organizations are increasingly moving from isolated AI tools toward agentic AI systems. These autonomous or semi autonomous entities can reason, decide, and act across complex processes. This shift is central to the Atos strategy and marks a fundamental transition — from prompt based automation to goal driven, decision centric systems that operate continuously, at scale, and under real world constraints (learn more about the Atos agentic AI approach).

However, as autonomy increases, so does the difficulty of decision making. Many of the toughest problems faced by agentic AI are not deterministic. They are characterized by partial information, competing stakeholders, strategic interaction, and uncertainty about outcomes. These are precisely the settings where traditional AI approaches — based on exhaustive search, simulation, or heuristics — begin to struggle.

This is where recent advances in quantum computing, combined with agentic AI, present a new perspective.

A new quantum algorithm: Not a replacement but an enabler for agentic AI

In the recent paper Imperfect Information Games on Quantum Computers: A Case Study in Skat, which I have co-authored with Gabriel Maresch (Technical University Vienna), Stefan Edelkamp (Charles University Prague), and Erik Schulze (Bull), we introduce a new quantum algorithmic approach that addresses exactly this class of problems.

The card game of Skat serves as a compact and well understood proxy for decision making under the uncertainty of multiple agents, hidden information, strategic interaction, and high combinatorial complexity. Concretely, the paper proposes quantum computing algorithms for the following steps of a typical Skat game:

State preparation: Build an initial quantum state representing a uniform superposition over all valid deals (with given cards on the hand of one player).
Game evolution operators: Define gates for playing a card and trick-taking according to the game’s rules.
Scoring: Based on quantum counting algorithm, this scoring yields a quadratic improvement over classical counting.

This paper has been accepted for peer-reviewed publication at the International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026). Now, AAMAS is not a quantum computing conference; it is the leading forum for autonomous agents and multi-agent systems.

While this is in line with Atos’s focus on agentic AI, publishing this work at AAMAS reflects the approach of quantum computing to be treated, not as a standalone technology, but as a capability inside hybrid or agentic systems. And I am honored to present this paper at AAMAS and illustrate how it leverages the integration of quantum computing into agentic AI.

While the underlying method is based on quantum computing algorithms, the conceptual takeaway is simple. Instead of evaluating one assumed world at a time, quantum computation allows us to evaluate many possible worlds simultaneously, and to assess how well a given decision performs across that space of uncertain yet possible options.

The contribution of the paper is about demonstrating that decisions under imperfect information can, in principle, be evaluated more efficiently when quantum methods are used as part of the decision loop.

Quantum computing is particularly well suited for search, optimization and sampling problems—the very problems that lie at the heart of multi agent coordination and game theoretic reasoning. Rather than replacing agentic AI, quantum algorithms can augment it, acting as a computational accelerator inside agent decision loops.

The newly accepted research explores this idea by introducing a quantum algorithm tailored for multi agent settings, enabling agents to reason more efficiently about joint strategies, equilibria and optimal actions. One of the key insights is that quantum speed ups are most effective when embedded in well designed hybrid or agent architectures, rather than used as isolated solvers.

AI, in particular Deep Learning, can handle huge amounts of data as training and analysis input, but is generally limited in solution space. Quantum computing, however, cannot handle large data input unless directly connected to quantum sensors, but can provide for a large solution space. This also shows the value of combining quantum computing and AI, benefitting from the complementary capabilities. A more general overview of the state of the art of how quantum computing fits into AI can be found in the article: Quantum computing and artificial intelligence: status and perspectives.

Fitting into the Atos agentic AI vision

Atos describes agentic AI as systems that reason, decide, and act on their own, embedded in mission critical environments and governed by enterprise grade controls. In this vision, autonomy is not about unrestricted freedom — it is about trusted, goal oriented decisioning.

From games to markets, infrastructure, and enterprises

Although the case study is Skat, the implications are much broader. Many real world decision problems share the same structural characteristics:

Award processes and auctions, where agents reason about unknown or partial knowledge and competitor behavior
Negotiation and contracting, where optimal behavior depends on beliefs about other’s information and strategies
Asymmetric information settings, where decisions must remain robust across uncertain futures and adversarial behavior
Risk analysis and security planning

In all these domains, agentic AI systems are increasingly being deployed — not to compute a single optimal answer, but to continuously adapt, learn, and act under uncertainty. The quantum approach explored in the paper I have co-authored shows that there may be new ways to strengthen the decision core of such agents as problem sizes grow. It offers an early but concrete illustration of how quantum computation can support this need.

Atos reinforces the same clear strategic message: Agentic AI is the system layer, and emerging technologies such as quantum computing can be integrated selectively to enhance its most critical capabilities.

As autonomy increases, so does the need for better ways to reason under uncertainty.

>> Connect with me to understand how quantum computing and agentic AI can accelerate your business strategy and growth. I will be at AAMAS, presenting my paper along with my colleagues and co-authors: https://cyprusconferences.org/aamas2026/.

>> Find out more about the Atos commitment to agentic AI and how we are uniquely poised to help organizations scale agentic AI safely, securely, and with full control over cost, value, and sovereignty: Agentic AI White Paper.