Agentic AI Is Hitting a Wall. Quantum Computing Might Be the Way Through

By Bill Genovese, Chief Quantum Officer | Technology Fellow | Head of Quantum Innovation | Global Emerging Technology Focused CIO & CTO | Delphi member 

Every agentic AI system eventually runs into the same wall: too many agents, too many possible interactions, not enough classical compute to handle it. The math gets ugly fast. Add ten agents to a coordination problem and the number of possible interactions doesn't grow by ten, it multiplies. Somewhere past a certain scale, brute-force classical computing simply can't keep up.

That's the operational reality behind a lot of the current agentic AI enthusiasm. The demos look great at small scale. The friction shows up when organizations try to run these systems at the size they actually need.

Reality vs. Hype

The hype cycle around agentic AI tends to treat scale as a solved problem, something that just needs more cloud compute or a bigger model. Leading firms building these systems in production are finding it isn't that simple. The bottleneck isn't just processing power, it's the nature of the optimization problems themselves: scheduling, resource allocation, multi-agent planning. These are the kinds of problems classical computers handle inefficiently no matter how much hardware you throw at them.

Quantum computing doesn't fix agentic AI outright. But it targets exactly this kind of problem. Because qubits can represent multiple states at once, quantum systems can explore many possible solutions in parallel rather than working through them one at a time. For search and optimization tasks, that's not a marginal improvement, it's a structural one.

There's a coordination angle too. Quantum entanglement raises the possibility of tighter, faster coordination between agents, potentially reducing how much explicit communication a multi-agent system needs to stay in sync. That work is still early. It's a promising direction, not a deployed capability.

Where the Convergence Actually Helps

A few places this shows up in practice, or is close to it:

Quantum algorithms built for optimization, the kind used in scheduling and resource allocation, map naturally onto the problems agentic systems already struggle with. Firms working in this space are watching this closely, less for immediate deployment and more for what it signals about where the ceiling on classical approaches actually sits.

Data-heavy training workloads are another area worth watching. Fields like drug discovery and climate modeling already lean on datasets large enough that classical processing is a real constraint. Faster analysis there could translate into faster, more capable training for the AI models built on top of it.

And the relationship runs both directions. AI isn't only a beneficiary of quantum computing, it's increasingly a tool for managing it. Calibrating quantum hardware requires exact control over a large number of interacting parameters, which is exactly the kind of precision task AI agents are well suited to handle. Some of the more interesting near-term applications may end up being AI making quantum systems more reliable, not the other way around.

What This Isn't

Worth being direct about the limits here. Quantum hardware capable of running these algorithms at meaningful scale is still maturing. Error correction remains a hard, unsolved problem across the industry, not a rounding error. Nothing here suggests agentic AI's scaling problems get solved this year, or that quantum computing is a drop-in fix. What it does suggest is that the two fields are converging in ways that matter for anyone planning multi-year AI infrastructure, and that's worth tracking now rather than waiting for the technology to fully mature.

Executive Q&A

Is this relevant to my organization if we're not running large-scale multi-agent systems yet? Even single-digit agent deployments run into scheduling and optimization bottlenecks sooner than most teams expect. Understanding where the classical ceiling sits is useful before you're the one hitting it.

Should we be evaluating quantum vendors now? Yes, at least for quantum preparation and quantum-safe migration. Post-quantum encryption for agentic AI systems isn't optional homework for later, it's a near-term planning item. More broadly, tracking the space and understanding which of your AI infrastructure's pain points and vulnerabilities are actually quantum-addressable, versus just under-optimized classical systems, is a reasonable near-term step.

How far out is this, realistically? For quantum-vulnerable AI infrastructure broadly, timelines vary widely by application and there's no single consensus figure worth anchoring to. Optimization and calibration use cases are further along than large-scale training acceleration, and some leading global banks are already moving these from R&D toward production.

Post-quantum encryption migration is a different story: the planning needs to start now. NIST's principal target for completing the migration is 2035, but organizations should treat 2030 as the real operational deadline. The practical timeline looks something like this:

• Now: begin migration using NIST's finalized PQC standards (ML-KEM/FIPS 203, ML-DSA/FIPS 204, and SLH-DSA/FIPS 205).
• By 2030: quantum-vulnerable public-key algorithms such as RSA, ECDSA, EdDSA, and classical Diffie-Hellman are expected to be deprecated for most uses.
• By 2035: those algorithms are expected to be disallowed and removed from NIST standards, with PQC migration substantially complete.
• Sooner than 2030: high-risk systems and data with long confidentiality requirements should move earlier, given "harvest now, decrypt later" exposure.

The executive takeaway: inventory and plan now, implement materially before 2030, and complete the transition no later than 2035. NIST's transition schedule is currently laid out in draft NIST IR 8547, so algorithm-specific dates may still shift as that finalizes.

What should we be doing differently today? Mostly this comes down to architecture decisions: designing agentic systems in ways that don't lock you into scaling assumptions that quantum-classical hybrid approaches could later change.

About the author: Bill Genovese is Chief Quantum Officer and Technology Fellow, with a background spanning CIO and CTO roles across financial services, cloud, and high-performance computing. Based in St. Augustine, Florida, he has worked across Asia, Australia, Europe, and the Americas, with a focus on emerging technology adoption in quantum computing, AI, and next-generation infrastructure.

‍