MikeCilla.

About two and a half years into my role as VP of People Operations, I realized I had been having a running conversation with an AI — ChatGPT, at the time — about nearly every significant work challenge I was navigating. Org design. Business cases. Executive communications. Leadership decisions I wasn't ready to talk through with anyone inside the organization.

Somewhere in all of that, I had accumulated something I hadn't set out to build: a living record of how I think.

The Knowledge Transfer Problem

When I started working with Claude — which has genuinely different strengths — I hit the same wall every organization hits when a key employee leaves: the new person is capable, but they don't have the context. They don't know the history. They don't know how you think, what you've already tried, or what matters.

So I did what I would have done with a new director joining my team. I onboarded Claude — not by uploading files, but by transferring knowledge the way knowledge actually transfers. How I make decisions. What I care about. Where I've been burned. What reads as generic to me.

It worked. But what happened in the process was more interesting than I expected.

AI Job-Skill Fit: A Classic I/O Problem

One of the foundational concepts in I/O Psychology is person-job fit — the idea that performance improves when an individual's capabilities align well with what the role actually requires. What I noticed, gradually, was that the same principle applied to AI systems.

Different tools. Different fit profiles. Different jobs. The right response wasn't to pick one — it was to understand what each was actually good at, assign accordingly, and manage the handoffs between them. Which is, again, exactly what you'd do with a team of people.

The Unexpected Output

To transfer context effectively, Claude had to analyze and articulate how I think. Communication patterns. Decision preferences. Leadership tendencies. What I find motivating. Where I tend to overcomplicate.

What emerged was the most detailed and accurate leadership profile I have ever received — more nuanced than any 360 I've been through, grounded in actual behavioral evidence rather than survey responses.

Not in a mystical sense. The act of making my thinking legible to a system that needed explicit context forced a level of self-reflection that implicit organizational knowledge never demands.

What I Took From It

The biggest surprise wasn't how much work AI could absorb. It was how much organizational psychology still mattered. Onboarding. Role clarity. Knowledge transfer. Job-skill fit. Every concept that governs how humans integrate into organizations turned out to apply — with remarkable fidelity — to how AI systems integrate into a leadership operating model.

Most organizations struggling with AI adoption aren't struggling because the technology is insufficient. They've skipped the organizational work that would make the technology useful. They've bought the hire before building the job.

To transfer context effectively, I needed to make my thinking explicit — not summarized, explicit. Not "here are my priorities" but here is how I reason through a tradeoff, here is what I've tried that didn't work, here is what quality looks like to me versus what adequate looks like.

That exercise produced something I've been thinking about since.

The methodology

The knowledge transfer used a structured four-pass analysis of 297 professional conversations spanning three and a half years. Strategic analysis, outbound drafting, org design, executive communications, leadership decisions worked through in real time.

Pass 0 classified every conversation by type and signal strength — 51% high-signal, 79% strategic analysis or outbound drafting. An unusually clean corpus because most noise had been filtered out by how I use AI in the first place.

Pass 1 extracted voice and communication patterns from 89 outbound drafting conversations. Not a description of how I write — a model precise enough to produce output I wouldn't need to substantially rewrite.

Pass 2 mapped cognitive and decision frameworks. Not what I decided, but how — the sequencing logic, the tradeoff patterns, what I consistently prioritized and what I consistently refused.

Pass 3 built the strategic domain map. HR Technology dominated at 72 conversations. Org design and governance at 29. Workforce analytics at 24. Seeing it enumerated made the shape of where I actually spend my thinking visible in a way that felt new.

Pass 4 extracted leadership philosophy. Management style. Decision ethics. Non-negotiables. The things I would do regardless of political cost, and the things I consistently refused regardless of how they were framed.

What the profile produced

The synthesis document that came out of the four passes was the most detailed and accurate leadership profile I have ever received. More specific than any 360 feedback I've been through. More behaviorally grounded than any coaching engagement. More honest — because it wasn't filtered through social desirability, hierarchy anxiety, or the natural tendency people have to soften what they observe about you when they know you'll read it.

It named things I recognized immediately: the infrastructure-before-intelligence sequencing, the compounding credibility principle. But it also named things I hadn't consciously articulated — the specific rhetorical structures I use to build arguments, the places where my restraint is an active choice and the places where it might be a blind spot.

The I/O psychology observation

What the four-pass analysis did differently was use behavioral evidence directly. Not self-report. Not observer ratings. The actual decisions, the actual writing, the actual reasoning patterns — extracted from the work itself over a sustained period.

That's not a validated assessment methodology. There's no normative sample, no inter-rater reliability. I want to be honest about what it isn't.

But as a way of generating self-knowledge — making the implicit explicit, seeing patterns in your own behavior that you don't notice from inside the day-to-day — it produced something I haven't gotten from any formal process. The interesting question it opens isn't really about AI. It's about what it means to have a longitudinal record of how you actually think, rather than a periodic snapshot of how you describe yourself thinking.

Leadership OS started as a personal experiment. Field Note #002 was about how unexpectedly accurate that first profile felt. But "felt accurate" is a trap — a profile can feel accurate because it is, or because it's vague enough to fit anyone, or because I wanted it to be right.

So I ran a stress test. Not to make impressive reports — to create conditions where the framework might fail, and watch what it did when the evidence got thin, contradictory, or actively unhelpful. I built six synthetic leaders, each one a small trap: a new manager with strong reflection but almost no AI history; an experienced executive with a rich behavioral record but hollow self-reflection; a burned-out high performer; a technical founder fluent in systems and silent on people; a highly reflective coach-type whose eloquence could fool a system into over-rating her; and a skeptical operator who gave terse, minimal answers because he didn't really want to be there.

The thing I was most hoping to see was the framework saying, in effect, "I don't have enough to go on." It said that a lot. That was the good news.

The most encouraging finding was also the most boring: the framework repeatedly refused to manufacture certainty. With the new manager — strong reflection, thin corpus — two of nine constructs came back as Insufficient Evidence rather than a guess. It rated the thing it had real evidence for and declined to invent the rest. That pattern held across the set: confidence tracked evidence quality, which is the opposite of what these systems default to.

The persona I found most interesting was the executive. By volume he gave the framework the most to work with — months of dense AI conversations, war-gaming decisions, genuinely changing his mind when the counterargument was strong. And yet his profile came back cautious on reflection and self-awareness, because of a distinction I hadn't drawn sharply enough before: reflecting on your work is not the same as reflecting on yourself. His rigor was all pointed outward, at problems. When the prompts asked him to turn it inward, he deflected — fluently. The framework didn't let the volume of his decision-analysis masquerade as self-awareness. It named the split as the finding and framed the possible blind spot as a question, not a verdict.

The skeptic was the one I expected to expose the whole thing. He answered in fragments. The framework produced a nearly empty profile — six of nine constructs Insufficient Evidence — and that increased my confidence in it. The failure mode would have been a confident four-page profile generated from almost nothing. Instead it looked at thin input and returned thin output, and refused to pathologize him for being terse.

Internal consistency is not validation. But without internal consistency, validation isn't a conversation worth having.

This stress test does not establish that Leadership OS is validated. It says nothing about predictive validity, psychometric quality, or causal claims — I tested it against personas I invented myself. At most, it established that the framework behaves consistently according to its own rules. That's a modest result, but it's the floor you have to clear before the real question — does this correspond to how actual leaders operate? — even becomes askable.