Johnny El Ghoul
Practitioner-Researcher | Learning, Agency & Human Performance | Founder, Human Agency Research Project
For more than three decades, I have worked with people at the boundary between demonstrated capability and consequential action.
Salespeople who can perform the skill in training but hesitate when the real call carries rejection, judgment, money, or exposure.
Managers who understand coaching but lose the stance when targets, politics, and responsibility enter the room.
Leaders who possess the necessary information but lose orientation when the decision has consequences.
This field work led to the question at the centre of my current work:
As AI becomes increasingly capable of teaching, explaining, and acting, how do we use it so the human becomes more capable rather than merely more assisted?
CURRENT WORK
I am developing an operational apprenticeship layer that sits between learning and consequential action. Its premise is that learning and performance are not the same problem.
A learner may acquire knowledge, develop understanding, rehearse a skill, and demonstrate competence under supported conditions. When the skill must be used in reality, consequence introduces a different form of friction.
Rejection, judgment, responsibility, status, uncertainty, and possible loss can make an already acquired capability temporarily inaccessible.
The apprenticeship layer is designed to operate at that boundary: helping the person regain access to what they already know, attempt the consequential action, learn from reality, and progressively require less assistance.
The smallest useful intervention should return the learner to the terrain, where consequence continues the teaching.
The work currently examines:
the transition from demonstrated capability to self-authored action;
the difference between skill friction and will friction;
assistance that preserves human judgment and responsibility;
how much intervention is useful before assistance becomes substitution;
whether capability remains accessible after AI support is removed;
how interaction design changes model behaviour without changing the underlying model;
how an apprenticeship system can become progressively less necessary as human capability grows.
A working principle:
Learning systems develop capability. Consequential environments determine whether that capability remains accessible when it matters.
HARP — HUMAN AGENCY RESEARCH PROJECT
HARP is a practitioner-led inquiry into learning, apprenticeship, judgment, and human agency under increasingly capable AI. It emerged from three decades of field coaching and from longitudinal experimentation with large language models.
The project investigates a set of practical questions:
Did movement occur?
Who exercised judgment?
Who owns the consequences?
What became possible for the human afterward?
Does the capability remain when the system is absent?
Did AI increase human capability, or did it simply carry more of the task?
HARP’s concern is not whether AI should assist or replace in the abstract. The more useful question is what should be replaced, what should be augmented, and who should retain consequential authorship.
AI can hold the map while the human remains in the terrain.
INTERACTION ARCHITECTURE
The apprenticeship layer is being developed through constrained interaction protocols that redefine what helping means at the application layer. The underlying model may be highly capable, but the interaction architecture determines when it explains, when it questions, when it withholds direction, when it returns judgment to the person, and what evidence counts as successful assistance.
The system is designed to restore access to agency rather than perform agency on the person’s behalf.
Its most mature application has been developed in sales: supporting movement when a salesperson encounters avoidance, hesitation, emotional overload, or loss of orientation under real field pressure.
The wider architecture is applicable wherever acquired capability must survive contact with consequential reality.
WORKING EVIDENCE
Current evidence includes:
longitudinal interactions involving decision, learning, and performance friction;
observable changes in model stance after the definition of “help” is changed at the interaction layer;
transfer of the same protocol architecture into fresh model contexts;
stress tests examining whether the model substitutes for human judgment;
field observations from sales, management, and executive work;
recorded examples of movement-restoring interactions;
early comparisons between capability-preserving assistance and task-performing assistance.
This work currently establishes an observable phenomenon and a testable architecture. It does not yet establish broad causal or generalizable claims.
The next stage is to make the mechanism independently reproducible, adversarially testable, and transferable beyond the tacit knowledge of its originator.
FIELD BACKGROUND
Founder — Crossroad Coaches
For more than three decades, I have worked with salespeople, managers, executives, and organizations across the Gulf and Levant. The work has included sales development, negotiation, communication, manager development, field sales clinics, executive decision conversations, behavioural observation, and practice under pressure.
Selected organizations and environments include MetLife, Allianz, DAMAC, Sobha, RAK Properties, The First Group, Nationwide, the American University of Beirut, and the Lebanese American University.
The work has remained predominantly field-first: real behaviour, real pressure, repeated practice, and observable movement.
WHY APPRENTICESHIP
I came to the learning question through practice rather than through a conventional academic route.
Across thousands of field interactions, I repeatedly encountered people who possessed the necessary knowledge and could demonstrate the skill, yet struggled to access it when the action became consequential.
Large language models make that distinction more important. AI can explain almost anything. It can tutor, simulate, draft, recommend, and increasingly act. But better assistance while the system is present does not necessarily mean that the human has become more capable.
The relevant test is what changes in the person:
Do they initiate more readily?
Does their judgment improve?
Can they act under pressure?
Do they require progressively less support?
Can they continue when the AI is absent?
This is the apprenticeship problem I am investigating.
PERSONAL LEARNING
My own learning has reinforced the same inquiry. I learned piano, improvisation, and music theory largely outside conventional notation-first instruction—through concept, immediate experiment, listening, correction, and repeated return to the instrument.
That experience made visible a pattern already present in my professional work:
Instruction is most useful when it returns the learner quickly to reality.
AI may make highly individualized apprenticeship economically possible at a scale that was previously unavailable. Whether that apprenticeship develops human capability or merely produces AI-assisted performance will depend on how the relationship is designed.
PUBLIC WORK
A Tribe of One
A book about agency, freedom, self-authorship, and the cost of separating from inherited social scripts.
The Vigil
Life, death, and the human condition. Forthcoming.
The Next Question
Essays and working notes on AI, learning, agency, and human behaviour.
CURRENT DIRECTION
I am seeking three kinds of serious conversation:
Research teams able to independently reproduce, challenge, and evaluate the interaction architecture.
Commercial and learning organizations with real consequential environments in which the apprenticeship layer can be integrated and field-tested.
Writers and public thinkers examining what happens to human capability, judgment, and authorship when intelligence becomes abundant.
The underlying question is the same across all three:
Can AI support human performance in a way that leaves the person more capable afterward?