HARP

HUMAN AGENCY RESEARCH PROJECT

THE THESIS AND THE FIELD-LAB ASK

​

HUMAN CAPABILITY UNDER INCREASINGLY CAPABLE AI ​

​

AI is becoming better at carrying human work. HARP asks whether it can also leave the human more capable.

​

AI can now explain, teach, synthesize, recommend, decide and increasingly act. Much of the market is rightly focused on how much work these systems can remove from people. HARP studies the other side of the equation: how AI can increase human capability without quietly taking over judgment, initiative, responsibility or authorship.

​

The question is not whether to replace or augment. It is what to replace and what to augment.

​

THE THESIS

When intelligence becomes abundant, better answers stop being the whole problem. The harder question becomes what happens to the human after the answer arrives.

​

AI can remove repetitive burden, search, synthesis and administrative friction. It can also strengthen perception, rehearsal, judgment and decision clarity. But if the system repeatedly crosses the consequential boundary for the person, performance may improve while human capability moves into the machine.

​

The test is therefore not simply: Did the AI help?

The harder test is: Where did the capability end up?

​

If support is reduced, can the person still see, judge and act? Who made the consequential choice? Who owns the outcome? Did the system help the human cross, or did it cross for them? These are the questions HARP is designed to make observable.

​

THE GAP AFTER EXPLANATION

For more than three decades, my work has been at the point where knowing fails to become doing. Salespeople know whom to call and do not call. Managers understand coaching and become directive when the numbers turn bad. Leaders have the information and lose orientation inside noise, politics and consequence.

​

Learning systems can move a person from not knowing, to understanding, to practice, and eventually to demonstrated skill. That matters. But a second boundary appears when the training wheels come off and reality enters: rejection, money, status, judgment, exposure, error and responsibility.

At that boundary, the problem may no longer be knowledge or skill. It may be overload, avoidance, distorted responsibility, a conflict of expectations, or pressure entering from the surrounding system. Another lesson may be the wrong intervention.

​

Training asks: Can you do it? The field asks: Will you? Apprenticeship asks: Can we help you cross without crossing for you?

​

Here, will is not a character judgment. It is the crossing from demonstrated capability to self-authored action under consequential conditions. Learning systems develop capability. Consequential environments determine whether that capability remains accessible when it matters.

ARI: THE BRIDGE BETWEEN INTELLIGENCE AND AGENCY

ARI - Agency Restoring Intelligence - is the developing human-facing interaction architecture within HARP. It is intended to sit between increasingly capable AI and the human who must still judge, choose and live with the consequences.

​

The model brings intelligence, language, memory and reasoning. The interaction architecture determines how that capacity enters the human system: when to teach, when to rehearse, when to orient, when to hold the friction, when to refuse substitution, and when to get out of the way.

ARI is a bridge, not a carrier. It should not simply transport the person from one bank to the other. The bridge remains available while the person crosses, returns, pauses, or eventually decides that it is no longer needed. The consequential movement remains theirs.

​

Its operating discipline is simple: observe the episode, identify the actual friction, offer the minimum necessary assistance, return the person to reality, and test what remains when support recedes. Minimum necessary assistance. Maximum retained capability.

​

This is not a trait score and not a claim to read the person perfectly. Any interpretation must remain provisional, contestable and connected to observable behavior. The architecture works within a model's governing values; it is not designed to override them.

ECHO: THE FIRST FIELD INSTRUMENT

​

HARP is inquiry first. Products are instruments.

​

Echo is the first purchasable expression of the architecture: a coach in the pocket for salespeople working under real pressure. It is also the first serious field instrument for testing the larger HARP thesis.

​

A CRM can show that a call was not made. Training can tell the salesperson what to say. Echo works in the space between those two facts. It helps determine whether the blockage is missing knowledge, weak skill, overload, avoidance, a specific status trigger, or friction arriving from the manager or organization. The response changes; the stance does not.

​

Echo is not another library of lessons, a gamified compliance layer, or a surveillance channel that sends private coaching transcripts to management. The objective is movement with retained authorship. Like a hand on the bicycle seat, the support should progressively recede as the rider becomes steadier.

​

Why the sales pit

Sales is more than a commercial use case. It is a strong field laboratory for human augmentation. Decisions happen quickly, repeatedly and with consequences. The salesperson approaches the difficult prospect or stays with the safe list. Asks for commitment or retreats. Returns after rejection or becomes busy with administrative work. The cycle repeats fast enough to study, and the surrounding systems can provide measurable signals.

​

Better results while Echo is present would not, by themselves, prove augmentation. The stronger standard is whether difficult action becomes more available to the salesperson, whether less support is required over time, and whether capability remains when the system is absent.

FROM A COACH IN THE POCKET TO AN ORGANIZATIONAL ARI

Echo is the wedge, not the whole architecture. The larger opportunity is a functional layer running across the organization, with constrained protocols delivered where work actually happens and an organizational learning layer above them.

​

At the interviewer's desk, the layer can surface possible conflicts in temperament, expectations and working conditions, give both parties a reality map, and leave the hiring decision with them. During induction, it can connect the company syllabus to the recruit's first real encounters. In sales, it can work at the moment of hesitation or pressure. In management, it can expose when support has turned into control. In customer service, it can help an agent handle anger, prevent avoidable cancellation, and navigate the tension between service, retention, cross-selling and incentive design.

​

At the leadership level, the opportunity is not an all-seeing machine issuing objective verdicts. It is a cleaner, more contestable organizational reality map. Properly bounded signals from the functional layers can reveal repeated blockages, avoidances, expectation conflicts, responsibility distortions and workflow friction that are normally filtered as they move upward.

​

The organizational layer should aggregate validated meta-signals, not expose private conversations by default. Local context remains local where it should. Access, consent, provenance and the right to challenge an interpretation must be designed into the architecture. The CEO receives a better map; the CEO still makes the decision and owns the consequences.

​

AI can hold the map while the human stays in the terrain.

​

This is where ARI becomes more than mentorship or tutoring. Each functional protocol helps at the desk. The organizational layer learns across desks. The value is not only a better individual interaction; it is an organization that can see where capability, responsibility and movement are breaking without silently transferring authority to the system.

​

This is the beginning of what we currently call an organizational brain: not one central intelligence deciding for everyone, but a governed system of local protocols, bounded organizational memory and aggregated evidence that helps each level see more clearly while leaving responsibility where it belongs.

ONE ARCHITECTURE, SEVERAL TERRAINS

The architecture is moving from roles toward functions. A salesperson may need orientation before movement. A manager may need emotional regulation before a clean decision. A buyer may need claims, evidence, uncertainty and risk tolerance separated before choosing. A learner may need less explanation and a faster return to practice.

​

Early applications include a decision layer used in a real-estate buying conversation and an interview layer used during actual hiring processes. Both were designed to improve the reality map while leaving the consequential judgment with the people involved.

​

Education is an important future terrain. AI may make genuinely individualized apprenticeship possible for children learning to read, write and reason across languages, but that proposition still has to be tested with appropriate safety, teacher and parent roles, and clear measures of retained capability. Riff, a music-apprenticeship experiment, is exploring the same principle in piano and musicianship: give the smallest useful instruction, return the learner quickly to sound and practice, and let consequence continue the teaching. Relationships and parenting are further possible configurations, not validated products.

​

WHAT HAS BEEN OBSERVED SO FAR

The work began in decades of field coaching and developed through longitudinal experimentation with large language models. Existing evidence includes before-and-after interaction traces, longer use across different forms of performance and decision friction, transfer of a constrained protocol into a fresh model context, and later evidence that the protocol's stance remained bound to that context over time.

​

There are also completed interaction records from live hiring interviews, a buyer decision experiment, music apprenticeship work, and repeated sales and leadership episodes. These are real observations, but they are not yet generalized proof.

​

The evidence is treated as episode-based and bounded: human x task x context x support level x time. The relevant record includes the interaction, the resulting action or decision, the level of assistance, what happened on the independent follow-up, and who made the consequential choice.

​

Full transcripts are not being published openly at this stage. A consented, privacy-aware Evidence Vault is being assembled for serious learning-platform, enterprise, research, frontier-lab and investment inquiries. The public thesis comes first. If the question is serious, the evidence becomes the next conversation.

​

THE NEXT FIELD EXPERIMENT

The next step is not to prove a finished theory. It is to make the phenomenon explicit, falsifiable, reproducible and transferable in a real operating environment.

​

A sponsored sales field lab would follow two longitudinal cohorts in parallel. The appropriate comparison conditions and evaluation design—including controls where feasible—would be agreed with the host and research partners.

​

Cohort one: the new recruit journey

Fresh recruits would enter from the interview process. The human-AI layer would support the interviewer and candidate, follow the recruit through induction and the company syllabus, accompany the first encounters with the field, and then provide progressively lighter support through the first months of real work. This lets us study how capability forms before habits and organizational distortions harden.

​

Cohort two: existing medium and lower performers

A sample of existing salespeople would move through a reset of training language, field diagnosis and longitudinal support. The purpose would be to distinguish missing skill from will friction, overload and organizational blockage, then observe whether consequential activity increases and whether assistance can recede.

​

The organizational lane

L&D and frontline management would run alongside both cohorts. They are often the translation point between upper management and the field, and therefore part of the system being studied. The lab would examine where individual friction ends and organizational friction begins: language, targets, coaching behavior, responsibility, incentives, information flow and unnecessary pressure.

​

Customer service and retention are a natural expansion lane once the first design is stable. Those teams sit at the point where angry customers, cancellation risk, service recovery, cross-selling and incentive contradictions become visible at scale.

​

Duration and measurement

A short independent pilot can test usability and marketability. A sponsored longitudinal field lab should run for at least three months; six months is more credible; ten to twelve months allows stronger testing of retention, withdrawal, cohort development and organizational change.

​

Movement: calls initiated, difficult prospects approached, follow-ups completed, presentations made, offers advanced, time to action, and recovery after rejection.

​

Capability: performance tasks, transfer to situations unlike the original exercise, independent attempts, and whether consequential skill remains accessible under pressure.

​

Support: what intervention was required, when the system held back, whether the required support decreases, and what happens when it is reduced or removed.

​

Agency: who framed the problem, who made the judgment, who acted, who owned the consequences, and whether the interpretation remained contestable.

​

Organization: recurring friction patterns, manager and L&D effects, responsibility distortion, workflow blockage, and downstream commercial outcomes where market and timing allow attribution.

​

The exact evaluation design should be agreed with the host. CRM and operational data are valuable, but they are not enough. Quantitative signals must be joined to consented qualitative evidence from the moment performance actually breaks.

THE ASK

The experiment is moving forward. There are two useful ways to participate.

1. Host or sponsor the field lab

Bring HARP into your own operating environment, or help place the experiment in a suitable partner field lab. Together we would define the cohorts, functions, duration, metrics, governance, consent, privacy boundaries, CRM access, technical integration, operating roles, compensation or sponsorship, and ownership or publication of the resulting evidence.

​

For a learning platform, the question is whether acquired skill survives the jump into consequential work. For a sales or service organization, it is whether movement, retention and capability improve. For a frontier lab, it is whether the interaction architecture is portable, stable and distinguishable from ordinary model performance. For an investor, Echo is the purchasable wedge; ARI is the larger organizational category it can open.

2. Help define the evaluation target

If sponsorship is not the right fit, tell us what evidence would make the independent experiment useful to your work. Which evaluations, outcomes and failure conditions would matter? How would you distinguish augmentation from temporarily assisted performance? What withdrawal test would be credible? What would demonstrate portability beyond one model, one founder or one unusually successful conversation? What result would disconfirm the thesis?

This is not a request for permission to continue. It is an invitation to make the experiment more legible, more rigorous and more useful before the evidence is produced.

​

Come inspect the evidence. Then let us put the architecture somewhere it can fail properly.

​

If it breaks, we learn where. If it holds, we may have found a way to scale part of something technology has never been particularly good at scaling: apprenticeship that develops the person, and organizational intelligence that improves the map without taking the decision away from the human.

​

Teach broadly. Measure accurately. Intervene precisely. Return agency. Then get out of the way.

ABOUT JOHNNY EL GHOUL

Johnny El Ghoul is a practitioner-researcher, coach and author who has spent more than three decades working with salespeople, managers, executives and organizations across the Gulf and Levant at the point where knowing fails to become doing.

​

That field work now sits inside HARP - Human Agency Research Project - an inquiry into learning, consequential action, decision-making and human agency under increasingly capable AI. His public work includes A Tribe of One, The Vigil (forthcoming), and The Next Question(Substack).

If these questions touch work you are already doing, come inspect the evidence or discuss a field lab.