Can AI support
quantum computing
work?
Quantum vibecoding is using an AI coding agent to design, submit, and analyze quantum circuits — without writing SDK code yourself. You describe experiments in natural language. The agent does the rest.
This is how one person with no quantum programming experience produced 100+ experiments, replicated 6 papers, and built 20+ interactive tools in 48 hours of wall-clock time.
How it works
"Run a Bell state on Tuna-9 and measure the fidelity"
Natural language, not code. You describe the experiment, the hypothesis, or the question.
H q[0]; CNOT q[0], q[1]; measure
Claude Code generates cQASM 3.0 or OpenQASM 2.0, handles qubit mapping, basis rotation, parameter optimization.
qi_submit_circuit → job_id: 12847
MCP servers wrap quantum hardware as tools. The agent calls them directly — no SDK code on your end.
Bell fidelity: 96.6% on q[2,4]
Statistical analysis, error mitigation, comparison to reference values. All automated.
What this method produced
The human role is not to write code. It is to direct attention, ask skeptical questions, and frame what matters. The AI handles implementation. The human decides what to investigate, when to doubt results, and how to communicate findings. This is a new mode of doing science.
The 5 Prompt Patterns
From 349 prompts across 445 sessions, five patterns emerged.
“how might I demonstrate the capacity of claude code on quantum computing?”
Open-ended questions that orient the agent to the domain.
“Queue them up and start gathering data. After every experiment, reflect and adjust.”
Directing experiments with adaptive feedback loops.
“wait, it was that easy? Are you sure that is real?”
Every major discovery came from doubting results.
“it's like AI is the interface between humans and quantum...”
Half-formed thoughts that crystallize the thesis.
“you are stuck.”
Sometimes the best prompt is two words.
The Stack
AI Agent
Quantum Hardware
Dive In
Step-by-step setup: Claude Code + MCP + 3 quantum backends. All free.
Set up your environment →349 prompts, 5 phases, the workflow that emerged from 445 sessions.
See the prompts →The academic context: QCT framework, metaphor design, HCI frontier.
Read the research context →What this method produced: 6 papers, 100+ experiments, 3 chips.
See the results →