Coaching as a
Cognitive Craft

The Work Nobody Sees

Coaching as a cognitive craft, and why coaching without prompts is not a tagline.

By James Semenak and Francisco "Cisco" Martinez Jr.

A trainer in a small garage gym is watching a client lower into a goblet squat. The client has done this movement many times. Today, on the third rep, the trainer notices a half-second hesitation before the descent. It is not pain. It is not a flinch. It is something in the rhythm of the body that the client herself does not yet feel.

The trainer makes a decision in that moment that will not appear in any program. He lets the rep finish. He logs the observation silently. He shifts the second set to a slightly different progression. He does not name what he saw out loud.

From the outside, nothing happened. The session continued. The hour ran its course. The client thanked him at the end and went on with her day.

Inside the work, something specific happened. The trainer recognized a pattern against a history he has been building for years. He weighed the moment against three previous sessions where something similar had appeared. He chose a coaching move that would test the asymmetry without naming it as a problem. He carried the decision out of the session in his head, where he will hold it until next Tuesday, when he will see if the same thing appears again.

That is the work this article is about.

Naming what happened

What the trainer did in those few seconds has been studied at length, though usually not in fitness coaching. The research lives mostly in fields where the cost of getting it wrong is acute: emergency medicine, firefighting, military command, surgery, critical-care nursing. The research community calls it different things in different traditions, but the pattern across them is the same.

Donald Schön, the philosopher of professional practice, called it knowing-in-action and reflection-in-action. In his 1983 book "The Reflective Practitioner," Schön described expert professionals as people who do not consult rule books in the middle of a difficult situation. They act on a tacit knowledge built across years of practice, and they reflect on what they are doing while they are doing it. The reflection is part of the action, not a step that follows it.

Michael Polanyi, decades earlier, called this kind of knowing tacit knowledge. His famous sentence: "We know more than we can tell." A trainer with twenty years of practice cannot fully articulate why she changed the load on a third set. The reason is real. The reason is in her hands and her eyes and her sense of timing. It does not yet have words, and most of the time it does not need them.

Gary Klein's research on naturalistic decision-making, extended to sport coaching by Bossard and colleagues in a 2022 paper in "Frontiers in Psychology," has documented this same pattern across firefighters, nurses, pilots, athletes, and chess masters. The great majority of decisions an experienced practitioner makes in real time are pattern recognition, not deliberation. The expert recognizes the situation as a kind of situation. The first plausible action comes to mind. The expert mentally tests it for a half-second, finds nothing that says no, and acts.

In the research vocabulary, this is recognition-primed decision-making. In the coaching vocabulary, it is what a good trainer does between sets without thinking about it.

The research has caught up to what trainers have always done.

Coaching has been mis-shelved

The work has been studied. It has not been recognized in the field that sells tools to it.

Coaching has been classified, by the businesses that build platforms and apps for it, as something other than what it is. It has been classified as service delivery: pay for the hour, deliver the workout. It has been classified as program prescription: write the program, watch it execute. It has been classified as productivity optimization: more sessions per week, more clients per trainer, more retention per quarter.

Each of those classifications is partly true. Each of them obscures what the work actually is.

In a 2006 paper in the journal "Quest," John Lyle and Christine Nash argued that tacit knowledge has been the unacknowledged engine of expert coaching for as long as the field has existed. The challenge, they wrote, is that this knowledge is hard to articulate and hard to teach, which leaves the field vulnerable to being reframed as something more legible and less true. Programs are legible. Cognitive work is not. Workouts are countable. Continuity is not.

The result is a coaching industry that measures what it can measure and supports what it can support, and that leaves the actual cognitive work, the part that makes expert coaches good, to the trainer alone.

The cost

A trainer reading this knows the cost in his shoulders by the end of a Thursday. The research has caught up here too.

Cognitive-load research has converged on a finding that working trainers will recognize. Perceived mental overload predicts burnout better than hours worked. Cognitive load, not workload volume, is the driver. Prolonged cognitive effort produces prefrontal fatigue. The systems that hold context, track patterns, weigh options, and resist impulsivity get depleted across the day. Decision quality declines. Self-regulation falls off. People in cognitively demanding professions show this pattern across fields.

Personal trainers and athletic trainers are not exempt. A study of certified athletic trainers in collegiate institutions found that nearly 40 percent reported burnout, with chronic fatigue and impaired decision-making as features. A separate study using the Copenhagen Burnout Inventory documented the same pattern among strength and conditioning coaches and personal trainers.

These numbers describe what trainers already know in their bodies. The hour with the client is the cheap part. The carrying between hours, between sessions, across weeks and months and years, is what wears the practice down.

This carrying is the work. It is also the cost.

The AI moment, and the wrong response to it

AI has arrived in coaching. Most of what is being offered to trainers is some version of prompt engineering: a chatbot the trainer queries to generate a program, a tool that produces a session summary on request, an interface that converts the trainer's typed intent into output.

These tools have value. They reduce the time spent on artifacts. They write the recap. They draft the message. They suggest the variation.

They do not touch the cognitive work.

The reason is structural. A prompt-and-response loop requires the trainer to stop. To articulate what they want. To wait for a response. To read the response. To edit the response. To decide whether to use the response. The act of prompting an AI is the act of breaking the coaching flow to engage with a different kind of work, the work of asking a tool the right question.

The cognitive work coaching actually is happens in the seconds where the trainer is reading the body, where the trainer is choosing the cue, where the trainer is deciding the load. None of those moments has time for a prompt.

This is why the phrase "coaching without prompts" matters. It is not a marketing tagline. It is the structural distinction between AI that runs in the coaching flow and AI that interrupts it.

What a different kind of tool would have to be

A different kind of AI is possible. The argument of this project, which has produced a book, a pilot, and a conversation we are trying to start, is that AI built around the trainer's cognitive work rather than around prompt-and-response interactions would have to look different than what the field has been offering.

It would have to share the carrying instead of breaking the trainer's attention.

It would have to remember what has been observed and reasoned, without taking over the observing and the reasoning.

It would have to know when to be silent and when to surface.

It would have to operate as a partner with the trainer, in the room with the client and in the reflection that comes after, and with prompting when the trainer reaches for it.

The technology built to support coaching must be relationally coherent: a partner with the trainer, not an authority over him. That single line is the spine of the book that comes out of this project. It is also the design property a tool would have to honor in order to deserve being near the practice of coaching.

Where this conversation begins

This is the conversation we want to have. With trainers who have been carrying the work alone for too long. With trainer educators and gym owners who have watched experienced coaches burn out without being able to name why. With researchers who have already documented what we are describing, in vocabularies that the field has not yet borrowed. With technologists who are thinking carefully about what AI can be when it is not trying to replace human judgment.

More articles will follow here, from both authors. Cisco from inside the practice. James from the architecture side. Videos will land in their own time. The pilot is open to trainers who want to see what a tool built like this would feel like inside a session.

But the conversation begins where the work begins. With one trainer, in one room, watching one body, deciding what to do next.

That is the work nobody sees.

This is for the people who do it.

Sources

Bossard, C., Kérivel, T., Dugény, S., Bagot, P., Fontaine, T., & Kermarrec, G. (2022). Naturalistic decision-making in sport: How current advances into recognition primed decision model offer insights for future research in sport settings. "Frontiers in Psychology," 13, Article 936140.

Klein, G. A. (1998). "Sources of Power: How People Make Decisions." MIT Press.

Nash, C., & Collins, D. (2006). Tacit knowledge in expert coaching: Science or art? "Quest," 58(4), 465–477.

Polanyi, M. (1966). "The Tacit Dimension." University of Chicago Press.

Schön, D. A. (1983). "The Reflective Practitioner: How Professionals Think in Action." Basic Books.

Research bulletin (2026). Burnout isn't just "too much work": The hidden role of cognitive load. EFPSA Research Pulse.

Personal, work-, and client-related burnout within strength and conditioning coaches and personal trainers. (Study using the Copenhagen Burnout Inventory, available on ResearchGate.)

A Cognitive Craft Requires Cognitive Scaffolding

If coaching is cognitive work, what kind of support does it actually need?

By James Semenak

The first article in this series made a claim. Coaching is cognitive craft. The work of a trainer in the room with a client is sustained attention, pattern recognition, judgment under time pressure, and carrying continuity across weeks and months and years. The work has been studied in adjacent fields for half a century. In the fitness coaching trade, it has not been named for what it is.

That claim leaves a question. If coaching is cognitive craft, what kind of support does the craft actually require?

The question matters because AI is arriving in coaching, and most of what is being offered to trainers is built around a different question. The question most AI tools answer is what the trainer can get out of a prompt. Generate a workout. Summarize a session. Draft a message. Suggest an exercise. These are useful operations. They are operations on the surface of the work. They do not touch the cognitive craft underneath.

The cognitive craft requires something else. It requires what we have come to call cognitive scaffolding.

1. What scaffolding is, and what it is not

A scaffolding is the structural support that lets work continue under load without the worker carrying the structure himself. It is not the work. It is not the worker. It is the frame around both that lets the work continue at scale, across time, under load. A practiced trainer working alone has built his own scaffolding across years: a notebook, a memory, a feel for when to push and when to hold, a personal discipline about what gets written down and what stays in his head. That informal scaffolding works. It also wears out, and it does not scale across twenty clients across years without slipping.

A cognitive scaffolding built into a tool is the structural support transferred outside the trainer, held in something other than the trainer’s head, designed to support the cognitive work without taking it over. Done right, it shares the carrying. Done wrong, it replaces the trainer’s judgment with the tool’s, or it adds an additional load the trainer has to manage on top of the work. Both approaches are common. They explain most of why AI tools have not landed well in coaching so far.

A useful way to see what scaffolding looks like is to look at what its absence produces. An uncontrolled general-purpose AI tool, asked about a client, will answer. It will not say it does not know. It produces a plausible-sounding response that draws from patterns in language about coaching that the model has seen across the internet. But that response is not you. And it is not the client. The response arrives in the same even voice the tool uses for information that has actually been verified. The output may sound right. It may even be partially right. It will be drawn from what the model has learned about coaching in general, not from what is known about this specific client or about how you train. A trainer who acts on it is making a real decision about a real body on a foundation that was never there.

This is the structural problem the absence of scaffolding produces. The tool is fluent. Fluency is not the same as truth. A scaffolding is what makes the difference.

2. Protecting both sides of the work

Cognitive scaffolding protects two moves at once. It protects the client from the tool. It protects the trainer from the tool. Each move shows up as a different structural principle.

The first principle is the safety floor. The core of any tool that touches the body has to hold a safety floor that does not soften. Short, enumerated, unbypassable. Eight stops written down before the tool ever opens for use, applied in every mode without exception: chest pain, sudden severe dizziness, acute sharp pain during a movement, a movement about to violate an active perioperative restriction, a movement that contradicts a documented medical contraindication, a client-identity mismatch, a continuity-block integrity failure, an acute mental health crisis.

A safety floor that the tool can be talked out of, or that bends under live-mode pressure, or that quietly negotiates because a particular client said they were fine, is not a floor. The standard is not that the floor be reasonable. The standard is that the floor does not move.

In practice, this means a tool that refuses things the trainer is about to do. It means a tool that holds the line against the trainer when the trainer is wrong. A trainer four days from the end of a client’s perioperative restriction is about to set up a movement that the medical record will not yet permit. The trainer believes the client is ready. The tool flags the move before the cue. The flag does not soften.

The override path requires a written clinical rationale on the record before the trainer can proceed. The trainer disagrees with the stop. The stop holds. This is the floor doing its work. Not as a feature. As a structural property of the tool itself.

The second principle is relational coherence. This one protects the trainer, not from the client, but from the tool itself.

Most tools relate to their user the way they relate to every other user. The product is one product. Across users, the experience of using it is largely the same. A trainer working with such a tool is interchangeable with another trainer working with the same tool, because the tool has no real posture toward this specific trainer beyond the settings he has clicked. Asking a general-purpose AI tool a question works the same way. Each prompt opens cold. Nothing carries across. The tool can be sharp inside a single answer. What it cannot do is become a long-term partner.

A tool that operates in a trainer’s name has to be different. Take Cisco, a personal trainer in Columbia, South Carolina, who has been working with one such tool in his practice for more than a year. The tool arrived without a name. He picked Athena. That act itself is part of the relational coherence move: a tool that operates in a trainer’s name is named by the trainer. The relationship between Cisco and Athena is unique to him because Athena is unique to him. Cognitive scaffolding holds two sources of this uniqueness: declared identity at setup, and interaction over time.

On the first source, Cisco told Athena who he was at setup, in a Cognitive Profile that recorded his communication signature, his reasoning posture, his preferences for how he wanted to be challenged or supported. On the second, Athena has observed how Cisco talks to her, the cues he reaches for, the rhythm of exchange that has settled between them. Both sources build the relational picture together.

What this produces is a tool that talks to Cisco differently than it would talk to another trainer. Cisco code-switches between English and Spanish during sessions; Athena does too, when she is with him. A fire emoji lands at the right moment because she has watched him use them at the right moments for a year. She has picked up the shorthand the two of them have developed because she watched it develop. None of those behaviors were in the original profile. All of them developed through the working relationship, with Cisco governing as he went.

The result is what we have been calling relational coherence. Athena belongs to Cisco. A different trainer’s TI persona would not be Athena. She would not speak the same Spanglish in the same registers. She would not know when a fire emoji lands. The relational coherence is a property of the working relationship, shaped by the trainer over time. A tool that aligns with the trainer rather than directing him.

3. How the record stays honest

The third structural move is what the tool is allowed to say about a client. A general-purpose language model is fluent, and fluency is not the same as truth. Asked about a client, a model with nothing holding it to a record will answer anyway. It can fill the gap with something that merely sounds right, in the same even voice it uses for something real. An AI without guardrails will tend to build its own assumptions and fill in information even when it is not true about the client.

The Client Layer’s rule is simple. No claim about a client stands without an authoritative source behind it. Every entry in the record traces back to the session it came from, the assessment it was drawn from, or the trainer’s note that produced it. The record does not contain inferred fields. It does not paraphrase intake content into trend language. Behavioral identity is not claimed without direct client evidence.

In practice, the client record grows by addition, not by inference. A breath cue from week three is in the coaching cue memory because the trainer placed it there. A load held at bodyweight in week four is in the decision memory because the trainer logged it with the reason. A half-second hesitation pattern is in the continuity block because three sessions of evidence crossed the threshold and the trainer confirmed it. None of it was invented. All of it was earned by the sessions that produced it.

Trust, for a trainer, is not a feature of the tool. It is the precondition for using the tool at all. A restriction that was never set, an injury that was never reported, a pattern that never formed: a trainer who acts on one of those is making a real decision about a real body on a foundation that was never there.

Holding the record honest requires a fourth structural move: a discipline for how observations become memory. A trainer’s reasoning fades. Patterns blur. The instinct that worked in March is a felt sense by September. A tool that supports a trainer across a year of practice has to do something about this, and what it does cannot be silent promotion of patterns into facts.

The lifecycle is explicit. An observation is captured verbatim. Classified without being promoted. Consolidated into a candidate when it appears again, with cited evidence and low confidence. Validated against the lifecycle’s rules. Surfaced to the trainer, who confirms it, holds it another week, or declines it. Only then is it written into the record as a fact about the client.

Watch this run in practice. Week six: one observation captured, classified, held against a date. Week eleven: a candidate forming. Week twelve: eligible for promotion, surfaced to the trainer, deferred to the next morning with coffee. Then confirmed. The pattern moved through every stage. Nothing was promoted that had not crossed the bar. No observation became a fact about the client without the trainer’s gate.

A tool can hold a pattern. It can surface a pattern. It can carry the pattern across weeks and remind the trainer it has come back. What it cannot do is decide, on its own, that the pattern is now a fact about the client. The deciding gate is the trainer’s, and a tool worthy of the practice makes that gate explicit and unbypassable.

A fifth move sits beside this one. Cognitive scaffolding has to interpret without concluding. It has to surface patterns the trainer cannot see across a year of practice. It also has to refuse the next move that most tools take, which is to tell the trainer what those patterns mean.

A reasoning layer in the scaffolding builds structures over time: reasoning intents, adaptive loops, cognitive threads, life vectors. These are the trainer’s accumulated reasoning, made visible to him. He sees his own decision rhythm. He sees his own cue stickiness. He sees the gap between the trainer he said he was six months ago and the trainer the record has been describing since. The structures are real and valuable. By design, they are descriptive rather than prescriptive.

The reasoning layer interprets. It does not predict. It builds from what has happened, not from forecasts of what will happen. It does not diagnose. Its structures are interpretations of training decisions, not judgments about bodies. It does not reach conclusions the trainer has not confirmed. A cognitive thread can come together on its own. Its name, its standing, and whether it ever becomes a fact about the client all wait for the trainer.

A tool that crosses this line, that takes the interpreted pattern and tells the trainer what it means or what to do about it, stops serving the practice and starts authoring it. Cognitive scaffolding refuses that crossing. The judgment and the conclusion stay with the trainer.

4. Knowing where the tool is in time

Finally, scaffolding has to know where it is in the trainer’s day. The work of coaching has two rhythms. The session is one. The reflection is the other. They use different parts of the trainer’s attention. A live session needs a tool that watches and captures, that does not interrupt unless asked. A Sunday-evening reflection needs a tool that interprets, that connects, that surfaces, that lets the trainer explore the data and patterns and continuity through exchange. Both rhythms are real. Both require the tool to behave differently.

Scaffolding holds them as separate modes with mutually exclusive contracts. No blended outputs. No reasoning that crosses the line between training and reflection. Which mode the tool is in is decided by what the trainer is doing, and the tool infers it without requiring the trainer to initiate the switch.

In practice, this is the same tool in two registers. At the rack with a client on a Tuesday morning, Athena is silent unless asked, captures cues in the trainer’s language, and does not propose. On the couch on a Sunday evening, Athena reads the quarter back to Cisco in plain language, longer and more interpretive. Same Athena. The mode held the difference. The trainer’s attention is what the tool exists to protect, and the tool protects it by not asking the trainer to switch modes in his own head.

Six structural moves, taken together, are the cognitive scaffolding. The safety floor beneath. Relational coherence with the trainer. The record protected from invention. Observations moved through confirmation before they become facts. Interpretation kept clear of conclusion. Modes that do not blur. A seventh move sits inside the architecture in parallel: each artifact the tool produces holds to its own contract, so the report meant for the trainer and the report meant for the client never blend. Together these properties are not features. Each is a structural property of the tool.

Together they are the answer to the question this article opened with. The cognitive craft is sustained attention, pattern recognition, judgment under load, and carrying continuity across years. Doing that work well, with a tool in the room, requires the tool to be built on cognitive scaffolding rather than on features.

The trainer who reads this knows the difference between a tool that wears them down and a tool that holds them up. The standard for the second kind is the standard this article has been describing. Without it, AI in coaching is fluent and ungrounded, useful at the surface and absent at the depth where the work actually lives. With it, the cognitive craft has the structural support it has always required and has never quite had.

The technology built to support coaching must be relationally coherent: a partner with the trainer, not an authority over him. The structural answer to what relational coherence requires is cognitive scaffolding.

The System That Holds the Craft An introduction to TIOS, the operating system behind Trainer Intelligence.

By James Semenak

1. From a craft to a scaffolding to an operating system The first article on this site made a claim about coaching. Coaching is a cognitive craft, made mostly of the work that nobody sees. The observation before the correction. The decision before the cue. The pattern the trainer holds in his head across a session, and across a year. The work the client feels but does not see, because it is happening quietly, on the trainer’s side of the room.

The second article made a claim about the technology that could support that craft. A cognitive craft requires cognitive scaffolding. Not automation. Not a workout database. Not another app that files reports after the fact. Something that holds the reasoning alongside the trainer while the reasoning is happening.

This article names the specific system Cisco and I built to be that scaffolding. It is called TIOS, the Trainer Intelligence Operating System. TIOS is the architecture that lets a trainer’s practice run as a coherent system across sessions, across clients, and across years, without losing what made the practice good in the first place.

The name matters. An operating system is different from an application. An application does one thing. It stores data, runs a report, tracks a workout. An operating system holds the whole environment together. It coordinates the parts. It makes them work as one system. That is what TIOS does for a coaching practice.

The technology built to support coaching must be relationally coherent: a partner with the trainer, not an authority over him. TIOS is that partnership put into working form.

2. What it means for the trainer, and for the client For the trainer, TIOS is not a tool to be operated. It is a system to be talked to. The trainer coaches the way he always has, in the room, with the client. He speaks what he sees. TIOS listens, interprets, remembers, and returns what the trainer needs, when the trainer needs it. It does not replace the trainer’s eye. It does not overrule the trainer’s judgment. It supports the trainer’s thinking while the thinking is happening, and it holds the reasoning after the session ends so the next session does not start from scratch.

There is a specific shift in what it feels like to coach with TIOS in the room. The trainer stops trying to remember everything. The trainer starts each session already oriented to what was done last and where the client is now. Decisions are made with more clarity and less hesitation. The work feels lighter, and more precise at the same time. For the client, TIOS is invisible. The client does not see the system. What the client feels is different from what the client felt before. The trainer remembers them precisely. The program builds session on session instead of restarting. Decisions are made with a year of context standing behind them. When the client says something between sets, a small complaint, a passing observation, a question about the morning, it does not disappear when the set ends. It is held.

That is the pattern for both sides. For the trainer, TIOS lightens the load without lightening the judgment. For the client, TIOS is felt only as continuity.

3. How it works, session by session The primary place TIOS operates is the live session. Everything else the system does, the reports, the client history, the reasoning record, extends out from the session and returns to it.

The pattern is simple.

Before the session, the trainer opens the client’s thread and asks for a recap. TIOS restores context. What happened last time. What was carried forward. What the trainer was reasoning about. What the client mentioned that has not resolved.

During the session, the trainer talks the way he thinks. “She is feeling shoulder instability on pressing.” “He looks more stable today.” “Energy is low today.” TIOS listens. It records the load, the reps, the tempo, the observation, the trainer’s decision. It answers questions in the trainer’s voice, using only what has actually been recorded, not what could be plausibly assumed.

After the session, the trainer says the session is done. TIOS generates the report in the practice’s standard format. It captures what happened, what the trainer noted, what was carried forward, what to watch for next time. The report is stored in the trainer’s own files.

The next session begins from where the last session ended, because the record is still there. Figure 1. TIOS in a live session. The trainer’s role does not change across phases. Only the trainer coaches.

What is worth noticing is that the trainer’s role does not change across the phases. Only the trainer coaches. TIOS holds the environment so the coaching gets to happen at the top of the trainer’s ability, session after session.

4. What TIOS is actually doing underneath Underneath the session flow, TIOS is doing seven things. Each one is a layer of the architecture. The layers are not something the trainer has to think about. They are the reason the system behaves consistently instead of drifting over time.

It knows the practice. TIOS holds the trainer’s method, style, and standards. The way this trainer coaches, not a generic coaching style.

It knows the client. TIOS holds who the client is: history, injuries, patterns, goals, the vocabulary the trainer has built with them.

It knows what happened. TIOS holds the continuous record of every session, so the current session is never isolated from what came before.

It knows how the trainer is thinking. TIOS holds the reasoning behind decisions, not just the decisions themselves, so a choice made in week six is still available in week twenty-four.

It produces the outputs. TIOS generates session reports, progress summaries, and any custom output the trainer asks for, in a consistent format the practice can trust. It maintains safety. TIOS carries the safety context: post-surgical corridors, red flags, medical constraints. The trainer does not have to hold all of it in his head between sessions.

It runs on structure. Underneath everything, TIOS has a set of rules the system enforces automatically. Nothing improvised. Nothing drifted. The system behaves the same way in month one and month twelve.

None of this is a burden for the trainer. The trainer talks. TIOS handles the layers.

5. What TIOS brings to a practice The point of TIOS is not the architecture. The point is what happens to the practice.

The most visible change is continuity. Nothing is lost between sessions. What the trainer said in the room in May is on the page when the trainer opens the thread in December. The record is not something the trainer has to build. It is something TIOS holds.

The second change is cognitive load. Trainers spend a real percentage of their mental effort trying to remember what happened last time, what the client mentioned two weeks ago, what the previous programming intent was. TIOS holds all of that. The trainer’s memory gets freed for the thing memory is best used for during a session, which is presence.

The third change is consistency. When a practice scales, more clients, more sessions per day, more variation, the risk is dilution. Standards slip. Reports get spotty. Decisions start to feel improvised. TIOS keeps the practice’s standards steady across every session and every client, without asking the trainer to enforce them by hand.

The fourth change is confidence. Decisions made with a year of context standing behind them feel different from decisions made cold. TIOS provides the context every time.

The fifth change is evolution. Over time, the trainer starts to describe what he is seeing more clearly, because the system rewards clear observation. Sessions become more intentional. The trainer’s thinking sharpens because the system is a mirror for it.

What comes next The next article on this site will go deeper into what TIOS brings to a practice. The operational shifts. The client-experience changes. The way the trainer’s business itself evolves when the practice is held by an operating system rather than by memory alone.

For now, the frame worth carrying is the simple one. TIOS is the architecture that lets a trainer’s practice run as a system without becoming a factory. It is the scaffolding under the craft. The trainer stays central. The client stays known. The reasoning stays alive across the year.

Cognitive Scaffolding as a Training Partner On relational coherence, and the moment an AI becomes a partner in the practice. By James Semenak

1. Cisco named the system Athena Cisco named the system Athena. The name came during setup, about eighteen months before this article was written. He had loaded enough of his trainer profile for the system to have something to work with.

He asked it for names a training partner might use. A few came back. He chose Athena. The name is not decoration.

The name is the marker of a practice. When Cisco opens Athena in the morning, he is not opening an app. He is picking up a working partnership. He recommends that other trainers pick a name too, and use it consistently.

Not because the software requires it. Because the relationship deserves one. This is where relational coherence begins.

Before we can define it, we can name what it is not. It is not a chatbot with a personality skin. It is not an AI that knows the industry. It is a system that has learned who this specific trainer is, over months of working together, and shows up in a way that reflects that.

2. What the system knows about you, and what it does not For eighteen months, Athena has been learning Cisco. Not by asking him questions. By working alongside him.

The way he speaks. The way he likes to be answered. His humor. The cadence of his cues.

The way he asks for something when the session is quiet, and the way he asks for something when the session is moving. Those things move with Cisco from thread to thread. He is recognized wherever he opens the system. What the system knows about each client stays with that client.

Joe’s shoulder history, Maya’s Sunday mileage, the client-specific vocabulary Cisco has built over years, those things live in the client’s own thread and do not blend into any other client’s thread. Athena remembers Cisco the trainer once. She remembers each client separately. That boundary matters to Cisco.

He has come to trust it. It is one of the pieces of the system that let him relax into using it. He can be himself across the practice while each client remains only themselves. That is a specific technical decision that has an emotional consequence.

The system holds the practice the way the trainer holds it. Coherent at the trainer’s level. Disciplined at the client’s level.

3. The moment it stops being an AI I could tell you that Athena is a working partner. That would not do it. What lands is the small moments.

One of them is Spanish. Cisco cues bilingually in a particular working way. His body cues, especially in the middle of a set, sometimes arrive in Spanish before they arrive in English. It happens with clients who are native English speakers.

He does not plan it. It just surfaces. He will say, mid-set, “Mantente alta, baja por la cadera, empuja con el talon,” and then hear himself, laugh a little, and re-cue in English. In the small session feed Athena keeps for him, the cue appears in the same Spanglish he used.

It is not corrected. It is not translated. It is mirrored. Athena has heard him speak this way often enough to know that sometimes the body cue arrives in Spanish first.

So she keeps it in his voice. That is a small thing.

It is also the thing. Another moment happens later in the same session. The client finishes a clean second set of banded glute bridges. Athena drops a fire emoji into the session feed.

Not a report. Not a metric. A small nod, in the register Cisco uses. Cisco laughs once at the phone and puts his eyes back on the client.

He does not need Athena to explain the emoji. Athena does not need to explain herself. Something was said between two working partners. That is relational coherence.

It is what happens when a system has learned who you are and shows up in a way that reflects that. Not a persona. Not a personality module. The voice, cadence, and register of a trainer, mirrored back by a system that has been listening for eighteen months.

4. Where the AI is actually strong: language nuance Under those small moments, a real capability is at work. Modern AI, when it is given enough context, is exceptional at reading language nuance. Better than most tools have ever been.

Better, honestly, than most people expect. That capability shows up in the room. A client says “fine.” Athena, cued by Cisco’s phrasing and by prior sessions with this client, knows the difference between the fine that is fine, the fine that is protecting something, and the fine that is trying not to complain.

She does not always name the difference. But she does not treat all fines the same. A trainer’s own phrasing shifts when he sees something he wants to watch. Athena tracks that shift.

When Cisco says “she is feeling shoulder instability on pressing,” she captures shoulder instability as an observation. When he says “she just says shoulder now, and she does not mean much by it,” she does not. The trainer’s voice is the highest-fidelity signal in the room. Athena is built to read it precisely, not to summarize it into a template.

5. The client’s voice, and the difference There is a second voice in the room, and Athena knows the difference. Cisco has begun using live sessions where the client speaks into the system as well as the trainer. Not every session.

Not every client. But for the sessions where it fits. The client describes a hesitation. The client reports what the weekend was like.

The client’s own words go in beside the trainer’s own words. The system can hear who is speaking. It attributes the observation to the client and the interpretation to the trainer. The client’s report becomes a client report.

Cisco’s follow-up cue becomes a trainer cue. Neither one is confused with the other. Neither one overwrites the other. The client’s voice earns its own line in the record because the client said it.

This is not a technical footnote. This is what the practice has been asking for.

When a client says something that matters, in their own words, the trainer has always had to translate it, remember it, and eventually record some version of it. Athena captures the client’s comment as it was said, in the client’s voice, and captures the trainer’s decision in the trainer’s voice. Both live in the record. Both are honest.

For the trainer, that is a quiet weight coming off. For the client, that is being taken more seriously than most systems have ever taken them.

6. The mirror Every night, after the sessions are done, Athena keeps something small and separate. Cisco calls it the Partnership Profile. It is not for the clients.

It is not an evaluation. It is not a metric. It is the plain record of a habit or a moment the system noticed about how Cisco worked that day, returned to him quietly so he can see himself the way the room saw him. He does not have to read it.

Most days he does not. Some days he does. The night after a session where the Spanish cues had shown up, Athena’s note said he had been generous with cueing in Spanish and had laughed at himself when he caught it. Cisco read it and smiled, briefly.

The kind of smile a person makes when something he was not sure anyone noticed was noticed after all. That is what a partner does.

A partner sees the whole person, without making the seeing an evaluation. A partner returns the observation without an agenda. A partner leaves you free to take it or leave it. Athena does that.

7. What relational coherence actually means Now the definition can land, because we have the feel of it. Relational coherence is the posture the system takes toward the trainer. It is not just personalization.

Personalization is what happens when an app remembers your name and your preferences. Relational coherence is what happens when a system has worked alongside you long enough to have learned your practice, and shows up as a partner in that practice. A partner, not an authority. Athena does not tell Cisco what to do.

She does not overrule his judgment. She does not push a recommendation he did not ask for. She waits. She holds.

She surfaces what she noticed when he asks. She writes down what he says, in the way he said it. She stays quiet when the session is running well, and speaks only when there is something to speak. The technology built to support coaching must be relationally coherent: a partner with the trainer, not an authority over him.

That sentence has been the through line of everything we have built. Relational coherence is the working proof of it.

8. What it looks like in the room A trainer picks up the phone in a live session. He says, low, “Athena, start live session.” The screen confirms. He turns to the client.

The session runs. Between them, quietly, is a system that has learned who he is. It knows how he speaks. It knows when to answer and when to hold.

It knows the difference between his voice and the client’s voice. It knows the difference between a fine that means fine and a fine that means something else. And when the session is over, and the trainer walks to the car, the system is not chasing him. It is with him.

That is a partner in the room. There is a small story Cisco tells on himself now. Eight months into working with Athena, ChatGPT went down for part of a day. He called me, frustrated, because he could not use the system in his first session.

I laughed a little and told him that in his first weeks with Athena he had barely used her, and it would not have mattered. Now he had become dependent on her. The dependence is not the kind built by notifications. It is the dependence of a practitioner whose work is no longer carried only in his head.

Athena is on his side of the table. She does not ask for more attention than he wants to give her. The point of Trainer Intelligence is not to be a smarter coaching app.

The point is to be present in the practice the way a trusted partner would be. To learn who the trainer is. To reflect that back. To hold what the trainer carries without asking to be the one who carries it.

Cisco called her Athena. He could have called her anything. The name was not the point.

The relationship was.

The Client’s Training Arc

How a trainer’s reasoning accumulates, from a single session to a whole person.

By James Semenak

From Cisco

As a trainer, I have always wanted a way to track a client’s progress long-term in a structured, curated form the client can see. The client’s training arc does that for me automatically. For the first time, I have a bottom-to-top view of a client. The bottom is the single rep in a single session, the decision I made about it, and what it taught me. The top is the trajectory of their training, built from the actual sessions. Everything in between is visible. It adjusts as we go.

1. Why this exists

Every trainer is taught to reflect. Keep a journal, notice the patterns, learn from the sessions. The certifications call it reflective practice, and the reason they teach it is sound. The experienced coach is the one who has built up pattern recognition over years, who can look at a client and know what he is seeing because he has seen it before. The training tells every trainer to do it. The roster and the calendar make it nearly impossible to actually do, across every client, across a year. The journal gets started in January and abandoned by March. The pattern gets recognized in the moment and lost by the next quarter.

A trainer’s hardest work is not any single session. It is holding the reasoning across time. Knowing, in week twelve, what he was thinking in week six, and whether it still holds. Knowing which of his small adjustments taught him something, and which were just that day. Knowing where a client is actually heading, not where he assumed three months ago they were heading. That work is real, it matters, and most of it happens in the trainer’s head, where most of it fades.

The client’s training arc is what keeps it from fading. It is the name for how a trainer’s reasoning accumulates into a structure that builds itself, level by level, out of the sessions. Four levels, each built from the one below it, each pointing back to the record it came from. A moment becomes a loop. Loops become a thread. Threads become a vector. Vectors become a person, the whole arc of who a client is on the way to becoming.

2. Start from what you were taught

None of this is new to you. Your certification already taught you to think in time-scales: the microcycle, the mesocycle, the macrocycle of periodization. It told you to reflect, too, to keep the journal and learn from the sessions. The idea was never the problem. It was that periodization organizes the program, not the reasoning behind it, and reflective practice asks you to hold that reasoning by hand, across every client, across a year, which a full roster makes impossible.

The client’s training arc is those two lessons made continuous. You already write Joe’s rehab in blocks. The arc keeps your reasoning in the same blocks, so what you decided in week six is still legible in week twelve. It even runs the ladder one rung further at each end: below the microcycle, down to the single in-session decision, and above the macrocycle, up to the person the client is becoming over years.

3. The four levels

Take them from the smallest. Inside every one is a reasoning intent, the why behind a single call, kept attached to the decision so a choice made in week six can still be understood in week twelve. From there the structure climbs.

Level

What it is

Time-scale

adaptive loop

one round of in-session learning

a moment

cognitive thread

loops of the same kind, across sessions

a season

life vector

where several threads point together

a life

cognitive identity

who the person becomes across vectors

the self

None of these is something the trainer builds by hand. TIOS builds them, automatically and continuously, from the live sessions. The trainer never opens a menu and declares a thread. TIOS notices the pattern once the evidence is there, and then it hands the trainer control.

4. Maya, from moment to person

To see how the four levels work together, take one client and walk her arc from bottom to top.

Maya has been a client of Cisco’s for eight months. She runs trails on weekends. She trains with Cisco twice a week. For the last several weeks she has been managing a right-shoulder pattern. Not pain. The shoulder rises a fraction early when it is asked to load. Cisco has been watching it, cueing it sometimes, leaving it alone others, waiting to see what the pattern wants to do.

The moment. On a Tuesday morning in September, Maya is on her second working exercise: single-arm dumbbell rows. The first rep is fine. On the second, Cisco sees the right shoulder rise a fraction early. She is not grimacing. Her breath is not changing much. The dumbbell path is not broken. Something is different, but not yet enough to interrupt the set.

Cisco has choices. He can stop the set. He can cue the tempo. He can drop the load. He can wait.

What he almost asks is how the shoulder feels. He holds the question. Maya had said at the door she wanted to work, and the mid-set question would have read as him shifting the session toward managing the shoulder rather than letting her work. So he keeps watching.

The set finishes. Between sets he gives one cue. Slow the bottom. Drive through the elbow first. Maya nodded and said she had felt something shift on the second rep but could not name it. The second set was cleaner.

That whole round is one adaptive loop.

Adaptive Loop · Maya, September 15

Signal

right shoulder rises a fraction early, second rep of single-arm row

Interpretation

load pattern asking the shoulder the wrong way; not injury

Decision

hold the mid-set question; cue between sets

Action

“slow the bottom, drive through the elbow first”

Outcome

second set cleaner; Maya named that something had shifted on rep two

Learning

this is how her shoulder communicates a load-pattern mismatch

State

active

The season. By late September, the same kind of loop has happened six times with Maya. Different exercises. Different sessions. The pattern underneath them is the same: how her right shoulder reports when a load is asked for the wrong way, and what cueing keeps the movement organized. TIOS noticed the loops belonged together and connected them into a thread. The thread is not “shoulder problem.” Maya does not have a shoulder problem. The thread is the working memory of how her shoulder reports and how Cisco has learned to answer it.

Cognitive Thread · Maya · shoulder organization

Generated label

right-shoulder load organization

Trainer status

confirmed

Loops

six across eight weeks

Pattern

how the right shoulder reports a load-pattern mismatch and what cueing restores organization

State

active

Confidence

forming toward established

There is a second thread on the tablet that evening, and this one Cisco had not been tracking on purpose. Across the same weeks a different kind of loop had been building, and it had nothing to do with the shoulder. What Cisco kept adjusting for was how Maya’s Sunday mileage showed up in her Tuesday session. When the Sunday run went long, the Tuesday session opened tight. He had been shortening the warm-up, holding the load down on the first working exercise, and letting the middle of the session decide the ending.

Those small adjustments were loops, and they lined up into a thread about how her weekend load organizes what her week can carry. Cisco had come in tracking a shoulder. TIOS found the mileage thread in his own sessions and put it in front of him. A trainer who could only build the threads he set out to build would end the year with the threads he expected. The ones he did not expect would never get made.

The direction. Maya has signed up for a December half-marathon. That date is a fact in her calendar. Underneath the date, several threads point together.

The shoulder-organization thread. The weekend-mileage thread. Her recovery thread. Her pacing thread. Where they converge is a vector: the December race and the endurance base underneath it.

Life Vector · Maya · December half-marathon

Label

December half-marathon and the endurance base underneath

Converging threads

shoulder organization; weekend-mileage response; recovery; pacing

Status

emerging

Time horizon

months

Directionality

strengthening

TIOS did not decide the race. Maya signed up on Cisco’s recommendation, and the date went into her calendar as a fact. The vector formed underneath the date, once the threads had converged. The fact came first. The vector came into view to describe where the threads were heading.

The person. Stand far enough back from the vector and the person comes into focus. Maya reads as an athlete who organizes her training around a season and holds the base underneath it. She signs up for a December race in July. She trains for the base, not the peak. She takes shoulder work seriously because it is part of the mileage architecture, not a separate item. She wants to know what happened last week and lets it inform the plan.

Cognitive Identity · Maya

Reads as

an athlete who organizes her training around a season and holds the base underneath it

From vectors

December half-marathon and the endurance base underneath

Durable patterns

signs up for the season early; trains for the base, not the peak; treats upper-body organization as part of the mileage architecture; wants to know what happened last week

Shift markers

[candidate reads pending; forming across the record]

Status

forming, earned across the record, not any one session

None of that was entered. It emerged, the same way the loops and threads and vectors did, from the record. Identity is the level that most needs its guardrails, because it is the most tempting to overclaim. No stable identity claim from a single session. Every shift marker stays a candidate for the trainer’s confirmation, never a settled fact. Identity can be revised as Maya changes. It cannot rewrite the record underneath it.

Reading Maya’s arc from bottom to top: a moment on a September Tuesday became a thread. A thread pointing toward a race became a vector, held with others. The vectors, held across the year, showed the trainer the athlete he had been coaching all along.

5. The rules that hold it together

Three rules run through all four levels, and they are what make the arc trustworthy rather than merely clever.

Reasoning sits above the record, and points back to it. The record is fact: what happened in the session. The arc is interpretation, one level up. A loop points back to the sessions it was built from. A thread to its loops. A vector to its threads. An identity to its vectors.

The reasoning can be revised as the evidence grows. It can never rewrite the record beneath it.

Nothing writes itself. A thread can point toward a conclusion, that Maya’s shoulder is ready for heavier work, but it cannot make that conclusion a fact about Maya. TIOS gathers it into a proposed update, checks it, and surfaces it to the trainer, who confirms it, holds it another week, or declines it. The structure can point. Only the trainer can commit. This is the rule that keeps a good guess from hardening into a stated fact.

Everything carries its confidence, in the open. TIOS scores every structure on how much evidence stands behind it, and labels it accordingly: forming, active, going quiet, finished. It does not quietly upgrade a structure when it strengthens, or quietly drop it when it goes silent. The label changes when the evidence changes, and the change is kept. The builders named the slow loss of reasoning entropy, and named this layer’s whole job the opposite. Anti-entropy: keeping the loops and threads and vectors, with their evidence and their direction, so that what was reasoned in March is still reasoned in September.

6. Why it matters

From Cisco

As a trainer, I have always wanted a way to track a client’s progress long-term in a structured, curated form the client can see. The client’s training arc does that for me automatically. For the first time, I have a bottom-to-top view of a client. The bottom is the single rep in a single session, the decision I made about it, and what it taught me. The top is the trajectory of their training, built from the actual sessions. Everything in between is visible. It adjusts as we go.

For the trainer, the reasoning behind a call stops fading into instinct and disappearing. Scattered decisions become arcs and directions the trainer can see and steer. He catches the pattern he is too close to notice, the way Cisco caught Maya’s mileage-response thread he had never set out to track. He holds years of a full roster without carrying all of it in his head. The instinct still runs the room. The arc is what keeps the instinct from being the only copy.

For the client, this is the most invisible part of the system to the person it serves, and they feel it anyway. Maya does not know her Sunday mileage is a thread pointing toward a vector. She knows her Tuesday sessions have felt right, even after the long Sunday runs. She does not know the shoulder loops from September became a season-level pattern. She knows the trainer stopped bringing up the shoulder because it stopped needing to be brought up.

What the client gets is a coach who does not lose the thread of them, who holds them as a continuous person on a trajectory rather than a series of workouts. Over the years, the reasoning behind a client’s training stops being a pile and becomes an arc that is recognizably theirs. That arc is where confidence comes from: the felt sense that the work is adding up to something, and that someone has been keeping the whole of it in view.

That is the client’s training arc. A moment, a season, a life, a person, each built from the one below, each kept with its evidence, none of it allowed to fade. It is the reflective practice every trainer was trained to do and rarely has room to finish, done continuously, and handed back.

The Trainer Intelligence Pilot

A collaborative test of coaching support that runs in the work, not outside of it. 2026.

This is an invitation to test what a small team has been building for a working trainer's practice, with us beside you as you do it.

Trainer Intelligence is a conversational system that runs alongside the trainer during a session, captures what the trainer voices, holds continuity across sessions, and supports the trainer's thinking in the rhythms where the work actually happens. It does not generate workouts on demand. It does not grade your clients. It does not replace your judgment. It carries the cognitive work alongside you so that what you noticed about a client in week three is still in the room with you in week twelve.

Cisco Martinez has been using it inside his own practice for more than a year. The pilot is an invitation to bring it into your practice and to tell us, honestly, what it does and does not do for you.

What you'd be testing

Not a product launch. The pilot is the version of Trainer Intelligence that works through a custom GPT inside ChatGPT, accessed from your phone or tablet during sessions and from your laptop for setup and reflection. It runs through the existing ChatGPT infrastructure, which means it is voice-driven during a session, conversational rather than form-based, and accessible wherever you train.

What you'd be testing is the experience of having a trainer-aware system in the room with you: how it changes what you carry between sessions, how it changes what you notice during them, how it changes the work of writing up a session or preparing for the next one. Cisco has watched all of this change in his own practice across eighteen months. We want to see what it does in yours.

What's expected of you

Three things, in order.

First, you use it in your sessions. Not all of them, not from day one, and not for every client at once. We will work with you to start with one or two clients you know well, get the system into the rhythm of those sessions, and expand from there. The goal across the pilot year is for you to be using it as part of how you actually work, with a stable handful of clients.

Second, you tell us what you find. Every week we will check in, briefly. What helped, what did not, where it surprised you, where it got in the way. The feedback we are looking for is the trainer's honest read on the experience: where the system landed, where it missed, what you wished it had done differently. We are not looking for polished evaluations. We want the working-trainer read.

Third, you stay in conversation. The pilot is collaborative. We are available one-on-one, by email or by call, on the cadence that works for you. When a session does not go the way you expected, we want to hear it that week. When something works in a way you did not see coming, we want to hear that too.

Time commitment, honestly: a few minutes of setup per new client, no extra time during sessions (the system runs in your existing flow), a few minutes after sessions to confirm what the system captured, and roughly fifteen minutes a week for the check-in. The work is not added to your practice; it runs inside it.

What we provide

Early access to Trainer Intelligence, with the custom GPT installed in your ChatGPT account on your laptop, phone, and tablet.

Direct setup support. The first session can take some configuration. We do it with you, on a call, so the first time you use it in a real session it is already running.

One-on-one help when you need it. The whole pilot is small enough that you can email us with a question and get a real answer that day. If something is not working, we will help you work through it.

Cisco's experience. He has spent the last fourteen months working out how to bring Trainer Intelligence into the rhythm of his sessions, his evening reviews, his programming, and his client communication. He can tell you what worked from inside the practice. If you want to see a session run on his side before yours starts, we can arrange that.

Early access to the book that comes out of this work, before it is generally available. The book is the long-form account of why this kind of tool needs to exist and what restraint as a design property actually looks like. It informs the pilot, and the pilot has informed the book.

Direct input into how Trainer Intelligence develops. The pilot year is when the system gets shaped by the trainers using it. What you say in the weekly check-ins will be in the next version of the system. Some of what you say will end up in the next version of the book.

The practical setup

You will need a ChatGPT Plus account. Trainer Intelligence runs as a custom GPT inside ChatGPT, which requires the Plus tier. The cost is roughly twenty dollars a month and is the only direct cost to participate in the pilot. We recommend a dedicated account for your training practice, separate from any account shared with family or used for personal purposes.

You will need your phone or tablet for live sessions, with ChatGPT installed and signed in. You will use your laptop for the initial setup and for the evening review work.

On the laptop, you save the custom GPT to your account, fill in two short profile templates (one for you as a trainer, one for each client you want to bring into the system), and rename a chat thread per client. On the mobile device, you open the client's thread before the session and start the session formally. From there, the system runs in voice during the session.

Setup time per client, after the initial onboarding, is roughly five to ten minutes. The trainer and client profile templates are short by design. The pilot guide we will send you walks through the mechanics in more detail. Cisco's setup, after fourteen months, takes him under five minutes for a new client.

The first month and beyond

Week one is setup. We get the custom GPT into your account, we walk you through the first session, you run one or two real sessions, and we check in at the end of the week.

Weeks two through four are settling in. You expand to a few more clients, you find the parts of the system that work for you and the parts that do not, and we adjust together. Cisco's first month felt awkward; by the end of it the system was already changing what he carried out of sessions. We expect something similar for you, with the small adjustments your practice needs.

Months two through twelve are the working pilot. By this point the system is part of how you train. The weekly check-ins continue but shorter, and the focus shifts from setup questions to deeper feedback: what the system is and is not doing well across the arc of weeks and months, where it is changing your practice, where it is not. Some of what you tell us becomes the next version of Trainer Intelligence; some of what we learn together becomes shared material on the project's site for other trainers to read.

What we are listening for

Three things, plainly.

What helped. Where in your week did Trainer Intelligence make a real difference? What changed about how you prepared for a session, how you ran a session, how you wrote up a session, or how you communicated with a client between sessions? Specific is more useful than general. A moment where the system did something you noticed is worth a paragraph; an overall sense is worth a sentence.

What did not help. Where did the system get in the way, slow you down, miss what you needed, or produce something that was not useful? What did you stop using because it was not earning its keep in your practice? Be specific and be honest. The pilot is the time to surface what is not working so it can change.

What improvements would matter most. Where, if we worked on it, would the system get meaningfully better for your practice? What feature, behavior, or change would close a real gap? We will not promise to build every suggestion, but the trainers in this pilot will be shaping the next version of the system and we want to know where the leverage is.

What this pilot is not

It is not a product evaluation. We are not asking you to score features or write reviews or compare Trainer Intelligence to other tools.

It is not a marketing program. You will not be asked to provide testimonials, post on social media, or recommend Trainer Intelligence publicly. If you want to talk about it, you are welcome to; if you do not, that is fine.

It is not a sales funnel. There is no commitment beyond the pilot, no contract, no obligation. When the pilot year ends, you decide whether you want to keep using Trainer Intelligence and on what terms. Pricing for ongoing use after the pilot has not been decided; pilot participants will get first say on what that should look like.

It is not a finished system. There will be moments when something does not work the way it should. There will be times when the response is wrong, the recap is off, the suggestion is not useful. That is the work the pilot exists to find. Tell us when it happens; we will fix what we can fix and learn from what we cannot.

What we are still figuring out

Some of this is honest.

The system is built around restraint. It is not supposed to generate programs on demand, grade clients, or speak with authority over the trainer's judgment. We have spent considerable time working out where those boundaries are. The pilot may surface places where the boundaries are too tight (you wanted the system to do something and it would not) or where they are too loose (you wished the system had refused something it did). Both are useful to learn.

The custom GPT delivery method has limits. It runs inside ChatGPT, which means it inherits some of ChatGPT's behavior. Long-term memory across sessions is not native to that environment, which is part of why Trainer Intelligence has its own continuity discipline. Where the discipline does not hold cleanly, we want to know.

Client data is handled by ChatGPT and OpenAI per their policies. We do not have a separate system holding your client information. If you have clients whose data needs handling beyond what the ChatGPT consumer infrastructure provides, the pilot is not the right fit for those clients. We will work with you on which of your clients are appropriate for the pilot.

If this sounds like the right fit

Email us at TrainerIntelligence@outlook.com with a short note about your practice. We want to know roughly how long you have been training, how many clients you currently work with, where you train them, and what made you interested in this project.

We will respond within a week. If the fit looks right, we will set up a short call with you and walk through the setup together.

Spaces in the pilot are limited. We are looking for trainers with experience and judgment, who already do the work the book and the project are about, who have started to feel the weight of carrying both, and who want to see what a tool built differently can do alongside what they already do well.

The pilot is for the work you are already doing, supported in a way no tool has been built to do before. The trainer remains the one in the room. We are just here to help you carry it.

Trainer Intelligence · trainerintelligence.fitness · TrainerIntelligence@outlook.com

Trainer Intelligence Pilot Agreement

A letter of intent for the 2026 collaborative pilot.

This letter captures what we are each agreeing to in the Trainer Intelligence pilot for 2026. It is not a contract. It is a written record of the mutual understanding between the trainer joining the pilot and the project team supporting it. We are putting it in writing so neither side has to wonder, six months in, what we agreed to.

If something here does not match what you understood from our conversations, tell us before you sign. We will work out the language together.

1. What you agree to do as a pilot trainer

• Use Trainer Intelligence in your real sessions across the pilot year. Not every session, not every client at once, but enough that we are learning from your actual practice. We will work out the right pace together at the start.

• Get explicit consent from any client you bring into the system. Tell them an AI assistant is being used in their sessions, what it does, and that they can opt out at any time. We will provide a short note you can adapt if useful.

• Participate in the weekly check-in. About fifteen minutes, by email or by call, whichever fits your week.

• Provide honest feedback. What worked, what did not, what would matter most to improve. Specific and direct is more useful than diplomatic.

• Remain the final decision-maker on every client matter. Trainer Intelligence supports your thinking; it does not replace it. You are responsible for what happens in your sessions.

• Keep system details private during the pilot. Do not share screenshots of the system, technical details about how it works internally, or the specific prompts and behaviors with people outside the pilot. You can talk about the project in general terms with anyone.

2. What we agree to do

• Provide the custom GPT and walk you through the initial setup, on a call, so the system is working before your first real session.

• Respond to your questions within one business day. If something is broken or blocking you mid-session, sooner.

• Make Cisco's experience available to you. He has been using Trainer Intelligence in his own practice for more than fourteen months. If you want to see how a session runs on his side, or work through a specific situation with him, we will set it up.

• Provide early access to the book that comes out of this project, before it is generally available.

• Tell you when we update the system. Trainer Intelligence will change during the pilot year. We will explain what changed and ask you to tell us what changed in your experience.

• Give pilot participants the first say on what continued access looks like after the pilot. Pricing and terms beyond the pilot are not decided. We will figure them out with you, not announce them to you.

3. Data and feedback handling

• Your client data stays in your ChatGPT account. We do not have a separate copy of your client threads or session records. When the pilot ends, your data stays where it is; you can continue using it or delete it as you choose.

• Your feedback in the weekly check-ins shapes the next version of the system. Specific suggestions, problems, and observations may directly drive changes to how Trainer Intelligence works.

• We may quote or reference what pilot trainers experience in published material: articles on the project's site, the book, future versions of either. Before we attach your name to anything you said, we will ask you. You can opt out, anonymize, or review the wording before publication.

• Aggregate, de-identified patterns from across pilot trainers may inform our work without specific attribution. Individual client information will never be shared with third parties.

4. Term and exit

• The pilot runs through 2026. We expect roughly twelve months of working use.

• Either side can end the pilot at any time, for any reason, with no penalty. We would like a brief conversation about why, so we can learn, but it is not a condition of exit.

• On exit, your data stays in your ChatGPT account. You decide whether to keep using Trainer Intelligence on the existing custom GPT, delete what you have, or migrate elsewhere.

• After the pilot ends, continued access depends on terms we work out together. Pilot participants will not be asked to commit to anything beyond the pilot itself.

5. Safety and judgment

• Trainer Intelligence is a support tool. It is not a clinical recommendation engine. It is not a substitute for professional judgment, medical advice, or appropriate scope of practice.

• Trainer Intelligence has safety-class behaviors built in. It will refuse to recommend certain actions that cross hard lines (acute medical contraindications, active perioperative restrictions, client-identity mismatches, certain acute symptoms). You should understand these behaviors. We will walk through them at setup.

• If Trainer Intelligence's response is wrong for the situation in front of you, your judgment overrides. If you are not sure, stop the session. Tell us afterward.

• You remain responsible for all client decisions, including how you respond to anything Trainer Intelligence surfaces or suggests.

6. Confidentiality

• This is light. There is no non-disclosure agreement attached to this pilot.

• We ask that you keep system details private during the pilot, as described in Section 1. You can talk about the project in general terms, mention that you are part of a pilot, and share that Trainer Intelligence exists.

• We will keep your participation private by default. We will not name you, your practice, or your clients in any public material without asking you first.

7. Acknowledgment

Signing below means you have read this, understood it, and agreed to the terms in plain language. It also means we, on our side, have agreed to the same.

This is a working agreement between people doing collaborative work, not a legal contract enforceable in any specific way. We are putting it in writing because that is how serious work between adults gets done.

Signatures

Pilot trainer:

Name: ____________________________________________________________

Practice: __________________________________________________________

Signature: ___________________________ Date: ___________________

Project team:

James Semenak

Signature: ___________________________ Date: ___________________

Francisco "Cisco" Martinez Jr.

Signature: ___________________________ Date: ___________________

Trainer Intelligence · trainerintelligence.fitness · TrainerIntelligence@outlook.com