During the final days of WebMCP-QCG, I stopped asking what else I could add.
I started asking a harder question:
Could someone understand and verify this project without me standing beside them?
That question changed the work.
WebMCP-QCG is a browser-native preflight workbench for quantum-development decisions. It can inspect a bounded artefact, evaluate a proposed action, present a recommendation, stop for human authorization, run a supported local classical simulation, and export an evidence receipt.
It does not reach a QPU. It does not promise quantum advantage. Most importantly, it does not allow an agent to make the final decision for the human.
Three surfaces, one shared truth
The project now operates across three connected surfaces:
- the Web application, where the artefact and human decision remain;
- a QCG panel inside browser DevTools;
- a persistent Companion side panel for inspection, collaboration, accessibility and evidence.
These surfaces share the same sanitized state, but they do not share every capability.
Raw quantum source remains inside the page. Consent tokens, credentials and private paths never cross into the extension. The Companion can inspect, explain and coordinate, but it cannot create consent or quietly transform a recommendation into authorization.
This separation became my most important lesson from the hackathon.
Listen to the short Companion podcast
What A2A should protect
Before this project, agent-to-agent communication felt like an architectural possibility. After building the Companion, it became tangible.
One agent can inspect. Another can challenge the first analysis. A human can compare their observations. The application can enforce its local boundary. A receipt can preserve what happened afterward.
That is the A2A-shaped collaboration I want to continue building: several systems contributing without merging their responsibilities.
Gemini participates through a deliberately manual and sanitized handoff. I am not claiming a native Gemini DevTools API or a complete authenticated A2A server. The useful mechanism is simpler and already real: recommendations can travel while human authority remains visible.
Clarity helps agents and humans
Direct use exposed another important problem. The first Light version of the Companion was too bright and visually tiring. Controls that looked acceptable on a large design canvas became harder to read inside a narrow persistent panel.
I refined the palette, contrast, focus indicators and accessibility controls through actual browser use. That reinforced a lesson I want to carry forward: clarity created for agents is also useful to humans.
Named states, explicit controls, readable history and predictable navigation reduce ambiguity for everyone. Accessibility is not decoration added after the engine. It is part of the interaction contract.
I invited cold judges into the process
By Day 7, I was too close to the project to trust familiarity alone.
I used external AI reviewers as cold judges. Their role was not to praise the architecture. Their role was to approach the repository like unfamiliar evaluators, identify fragile assumptions and force every concern back against the real code.
Some findings became corrections. Some became tests. Some required documentation. Others were rejected because the claimed defect did not match the implementation.
The human decision remained mine, but the counter-analysis made that decision stronger.
Why I continued
During these days, I helped an expert collaborator transform specialized knowledge into a professional publication. His passion was real, the purpose was visible, and I could strengthen the delivery without taking ownership of his work.
At the same time, an unclear professional request reminded me how quickly hope and intellectual attention can be consumed when identity, scope and commitment remain undefined.
Those situations were two views of the same avenue.
I want to help people build meaningful tools, even when they arrive without perfect technical language. But generosity needs trust, purpose and boundaries.
I did not finish this project only for a prize. I finished because I want to be ready when someone writes with a real dream and honestly asks for help.
Read the full Summer field report
The complete English and French editions belong to one canonical publication family:
The final video and submission remain outside this journal. If you want to see how the story ends, the evidence is public:
I built the gate so decisions would follow evidence.
I intend to close the hackathon the same way.
Protect Your Heritage. Secure Our Legacy.
Top comments (0)