We’re Building Zorgax as a Consent-Driven AI Operating Layer
We’re currently experimenting with a different approach to AI agents inside MyZubster.
The goal is not to build an agent that “does everything automatically.”
The goal is to build an agent that can operate across projects, identity, evidence, GitHub workflows, and eventually economic actions — while keeping consent, auditability, and human approval as first-class constraints.
That agent is Zorgax.
The current pilot
We’re testing Zorgax with a real participant in a digital entrepreneur workflow.
The process is intentionally structured:
Idea → Validation → Evidence → Human Choice → Blueprint → MVP → Approval → Possible Launch → Measurement → Learning
Instead of letting AI jump directly from “idea” to “build,” Zorgax first collects and evaluates evidence.
For the current pilot, two possible product ideas are being validated with real users.
Before evaluating the next responses, we created a fixed scoring model.
Each response can be scored against six criteria:
real problem;
urgency/intensity;
weakness of current solution;
willingness to test;
clarity of expected result;
clarity of required feature.
The model is defined before seeing the full dataset.
This avoids one of the easiest failure modes in AI-assisted decision making: changing the evaluation logic after seeing the answers.
If the evidence is insufficient, Zorgax should not invent confidence.
It returns:
MORE EVIDENCE REQUIRED
If the evidence is equivalent:
TIE / HUMAN CHOICE
That distinction matters.
Consent-driven automation
We are now adding another layer:
Zorgax Participant Profile Automation v0.1
The idea is to let Zorgax automatically maintain a participant’s project profile, but only inside an explicit consent boundary.
The intended flow is:
Consent → Authorized Scope → Data Processing → Proposed Update → Review → Human Approval
The participant can authorize Zorgax to use things like:
confirmed questionnaire responses;
project goals;
public GitHub activity;
documented progress;
validated evidence;
roadmap status.
Sensitive data stays outside the automation:
passwords;
tokens;
2FA codes;
wallet secrets;
private keys;
banking data;
unnecessary personal information.
If a field is unknown, the system should preserve that uncertainty instead of filling it with generated content.
GitHub becomes the audit layer
One design choice we are testing is using GitHub as the operational audit trail.
The preferred workflow is:
ZORGAX → BRANCH → COMMIT → PULL REQUEST → HUMAN REVIEW → MERGE
This makes AI-generated work inspectable.
You can see:
what Zorgax changed,
why it changed,
what evidence supported the change,
and whether a human approved it.
That is much safer than giving an agent unrestricted write access to production state.
Identity and the Metaverse
We’re also applying the same principle to MyZubster Metaverse identities.
Instead of treating avatars as disconnected aliases, we’re working toward account-linked characters.
A participant can choose:
character name;
archetype;
starting world;
while the identity link is based on a verified account flow.
The interesting part is not the 3D character.
It is the identity graph behind it.
A character could eventually represent:
verified participation;
contributions;
project history;
achievements;
evidence;
reputation.
Without exposing credentials to the AI.
Human-in-the-loop is not a limitation
A lot of agent systems treat human approval as friction.
We’re treating it as architecture.
For example, Zorgax can prepare recommendations, profile changes, project updates, and code changes automatically.
But sensitive actions remain gated:
merge to protected branches;
public commercial publication;
real pricing;
spending;
payments;
wallet actions;
destructive database changes;
external commercial communications.
The larger model is:
Knowledge → Reasoning → Decision → Human Approval → Action → Measurement → Learning
That is the direction we are building Zorgax toward.
Why we think this matters
AI agents are becoming more capable very quickly.
But capability without boundaries is not enough.
When an agent starts interacting with real identity, money, public reputation, infrastructure, or business decisions, we need stronger operational guarantees.
We need:
consent, provenance, auditability, reversibility, and human authority.
Our current hypothesis is simple:
the best AI operating layer is not the one that removes the human.
It is the one that removes repetitive work while making important decisions easier to understand and safer to execute.
That is what we are trying to build with MyZubster + Zorgax.