Most AI tools answer the question you ask. CoTrainer helps you ask a better question first. This is the story of how I built it, what it’s grounded in, and what happened when I used it on a real client with a real problem.

I’ve been an instructional designer long enough to recognize the pattern. A manager calls. Someone on the team isn’t performing. The instinct — the organizational reflex — is to find a training solution. Build a course. Send them to a workshop. Create an eLearning module. Check the box.

The problem is that most performance gaps aren’t training problems. Gilbert’s Behavior Engineering Model has been telling us this since 1978. Environmental causes — unclear standards, broken tools, misaligned incentives — account for the majority of performance failures. Training delivered into a broken environment doesn’t fix the environment. It just produces an expensive artifact that doesn’t transfer.

I built CoTrainer to change that instinct. Not to replace the consultant — to sharpen them.

What CoTrainer Actually Is

CoTrainer is an Android app that functions as a portable performance consulting system. It lives on your phone. It’s grounded in human performance technology — Gilbert, Kirkpatrick, Stolovitch & Keeps, Schunk (2020) — and it produces the kind of deliverables that used to take an engagement team days to generate.

The core workflow is a five-question diagnosis grounded in Gilbert’s BEM, followed by a suite of AI-generated consulting artifacts tailored to the specific situation. Not generic templates. Not boilerplate outputs. Specific, evidenced, theory-grounded analysis of the situation you actually described.

But before I explain what it does, let me explain what it doesn’t do. CoTrainer does not generate training content by default. It diagnoses before it prescribes. The first screen the user sees is a choice: Quick Diagnosis or Guided Discovery. Not “what course do you want to build?”

The Theoretical Foundation

I hold an MS in Learning Design and Technology from Purdue University. That credential matters here not as a line on a resume but as a commitment — CoTrainer’s AI system prompts are grounded in the actual research, not a popular interpretation of it.

The diagnosis engine applies Gilbert’s six BEM cells — E1 through I3 — before concluding anything about training. E1 asks whether performers have clear standards and timely feedback. E2 asks whether they have the tools and resources to perform correctly. E3 asks whether correct behavior is being rewarded or incorrect behavior is going uncorrected. Only after the environmental cells are examined does the model consider whether a genuine knowledge or skill gap (I1) exists.

training_warranted is a field in the diagnosis output. It’s true only when there is clear evidence of an I1 gap independent of environmental causes. Most diagnoses return a BEM flag alongside a training recommendation — because most real performance problems are multi-causal.

The Seven Advisor Voices

Once a diagnosis is complete, the user selects an AI advisor voice grounded in a specific learning science tradition. The same situation produces different artifacts depending on which advisor analyzes it. This is a feature, not a bug — different theoretical lenses reveal different aspects of the same problem.

Every voice prompt is fully operationalized — not just named after a theorist, but grounded in the specific frameworks that theorist contributed. The Behaviorist applies Mager’s three-component objective structure (behavior, condition, criterion) and explicitly designs reinforcement schedules. The Cognitivist applies Cowan’s (2001) revised working memory estimate of four chunks, not the outdated Miller 7±2, and maps every output section to Gagné’s Nine Events. The Constructivist designs for productive failure (Kapur, 2016) — problem before instruction — and requires all four stages of Kolb’s experiential learning cycle.

A reviewer who knows learning theory should be able to identify the voice from the output alone, without seeing the name. That’s the differentiating requirement built into every system prompt.

Guided Discovery — Morgan

Not every performance problem arrives with a clear description. Often the manager knows something is wrong but can’t name the root cause. Guided Discovery was built for that situation.

Morgan is an AI performance consultant embedded in the app. She doesn’t ask five structured questions. She has a conversation. She summarizes what she hears before asking anything. She pulls on contradictions. She notices what the user isn’t saying as much as what they are. She takes as long as the problem needs.

Morgan’s diagnostic conversation ends the same way the five-question flow does: a DiagnosisResult object with all the fields the rest of the system needs — root cause, BEM flag, recommended sequence, confidence score. The conversational path and the structured path produce the same downstream workflow. The difference is how much context the AI has to work with when it generates the Leadership Brief.

The Leadership Brief Package

This is the feature that changes what CoTrainer is. Once a diagnosis is complete, the user can generate a Leadership Brief — four documents produced in sequence from a single session, each grounded in the selected advisor voice.

Every executive summary includes a confidence score from 0–100%, calculated based on the specificity and completeness of the diagnostic evidence. A single intake conversation might produce a 52% score. That’s not a failure — it’s intellectual honesty. It tells the consultant and client exactly where the evidence is solid and where stakeholder validation is needed before committing to an intervention.

Then It Worked on a Real Client

Stephen owns Island Sailing Club. He’s been dealing with the same pattern for years — new hires interview well, underperform on the job. Under time pressure, his structured hiring process collapses into gut feel. He had access to an HR consultant, peer input, AI tools. None of it was sticking.

We ran a diagnosis through CoTrainer. One session.

The diagnosis identified two causes operating together. First, no structured hiring workflow Stephen could follow under time pressure — when urgency hit, shortcuts replaced process. Second, a skill gap in behavioral interviewing — evaluating candidates on interview performance rather than job performance predictors.

From that session, CoTrainer generated five deliverables: executive summary, stakeholder interview guide, intervention scorecard, consulting role analysis, and a 6-week engagement cadence.

Within 48 hours, two strong candidates were in the pipeline — one for member services, one for a fleet maintenance lead role that had also opened. Stephen took the CoTrainer output and started running applicants through it independently, without additional consulting intervention.

The 6-week consulting cadence is now in progress. The 30-day post-hire follow-up will provide Level 3 Kirkpatrick data — whether the new hiring system produced different on-the-job performance outcomes than the previous approach. That’s the measurement plan built into the system from the start.

What I Learned Building It

The hardest thing wasn’t the AI — it was the epistemics

Making CoTrainer say “52% confidence” instead of generating a confident recommendation regardless of evidence quality was a deliberate design decision. Most AI tools optimize for appearing certain. CoTrainer is designed to surface uncertainty — because that’s what a real consultant does. The confidence score calculation took more iteration than any other feature in the app.

Plain language is a feature, not a concession

The first version of the diagnosis output used BEM cell codes. E1, E2, E3. It looked rigorous. It was unusable for the non-practitioner. The current version uses plain language equivalents: “unclear standards or missing feedback,” “missing tools or access,” “the wrong behaviors being rewarded.” Same diagnosis. Completely different relationship with the user.

The system prompt is the product

Every advisor voice is a fully operationalized system prompt — not a named persona that happens to mention Vygotsky, but a structured reasoning framework that applies specific theoretical mechanisms to every output it generates. The Behaviorist doesn’t just sound behavioral. It applies Mager’s three-component objective structure, specifies reinforcement schedules, and distinguishes between fluency training and accuracy training. That depth is what produces outputs a real practitioner can use.

Honesty about what training can and can’t do is the differentiator

CoTrainer’s diagnosis system prompt has a critical rule built in: training_warranted is true only when there is clear evidence of an I1 gap independent of environmental barriers. Most diagnoses return a BEM flag. Most real performance problems have at least one environmental cause. Building an app that tells people not to build training when training isn’t the answer is, somehow, a competitive advantage.

The Consulting Cadence — What Was Delivered

The 6-week plan generated for Island Sailing Club shows what CoTrainer produces as a complete engagement scaffold, not just a point-in-time artifact:

What Comes Next

The Island Sailing Club case study is the first. The 30-day follow-up will close the measurement loop. The goal is a portfolio of case studies across industry contexts — maritime hospitality, B2B sales, healthcare, corporate L&D — demonstrating that the diagnostic-first approach produces different outcomes than the training-first reflex.

CoTrainer is available on Google Play. The first session is free. If you work in L&D, HR, or organizational consulting and have a performance problem you’re trying to get to the root of — that’s what it’s built for.

And if you’re an L&D professional who’s been told to build training for a problem that smells like an E1 or E3 issue — you already know what I mean. CoTrainer gives you the language and the evidence to push back.