Zach Morris time out on the story because we have been walking through BANI together: Brittle, Anxious, Nonlinear, Incomprehensible. We started with Nonlinear (I still love that joke). Then Brittle, where we looked at the actual science of how much give a thing has before it snaps, and I made the case that brittleness is not a bug. It is a property. A property of every relationship, whether that is one-to-one, one-to-many, or a whole system.

Today let’s explore Anxious. And before the ghost gets going, I want to say the thing my clients have heard me say a hundred times:

When I walk into a team or a system that isn’t working, the first thing I hunt for is where the anxiety and the ego are hiding. Not the org chart. Not the strategy deck. Helpful context sure, but I need to find the anxiety and/or the ego. Find those two, and you have most likely found the source of the funky behavior. That is where the work starts.

So. Back to the ghost.

The back of the couch

I had just moved to Connecticut and had nowhere to live yet, so I was house/cat sitting for some folks I was about to work with. The house gave me bad vibes from the jump. There was something about the second floor where I felt watched, like there was always something just behind a door, and no amount of checking ever fixed the feeling. So I never slept in the bedrooms. Couldn’t do it. Also never checked the basement. That was another big nope for me. I made dinner, watched a movie in the living room, and slept on the couch with a couple of lights on low.

The layout of the house matters for what comes next. It was an open living room and kitchen, and the back of the couch faced the kitchen. Lying down, I could not see into the kitchen. That was fine. I kept some lights on. It was all good.

Until one night I woke up, and she was there. Staring down at me, pale blue, hair floating. I could not breathe. It was one of the worst feelings of my life, this animal panic where I started sucking air and making an awful sound, and the only thing my body knew how to do was grab the pillow behind my head and throw it at her. She vanished.

And then I had a problem.

Because the back of the couch faced the kitchen, and I could not see in. And I became absolutely certain that if I lifted my head over that cushion to look, I would see her again, or something worse. Something that would break me for good. So I chose not to look. For hours. I was paralyzed in the decision to stay blind rather than risk what the looking might cost me. I lay there until the sun came up and I felt like I could look.

The kitchen was empty. It was almost certainly empty the whole time. I never checked. Schrödinger’s kitchen maybe?

(And, before you ask: I was 22. I had a credit card. I could have booked a hotel. But I did not book a hotel, because being 22 is its own special kind of dumb.)

Brittle is the property. Anxious is the reaction.

Something snapped in me that night, and it was not my nerve. It was my worldview. I went from “ghosts aren’t real” to “ghosts are extremely real and one is looking at me” in about half a second. That snap is brittleness getting exposed. A sudden change hit a belief that had no give in it, and it broke.

What came after the snap, the paralysis, the not-looking, the hours on the couch, that was the anxiety. Brittleness is a property. Anxiety is the reaction to it.

And notice what my anxiety was actually about. It was not about the ghost I saw. That was over. It was about the ghost I had not seen yet. The one I invented and installed in the kitchen and then refused to check on. Anxiety is a story we tell about a future that has not happened, a future that is worse than what we want, and then move into and live in as though it were already true.

The ghost you carry into the next house

After that awful night, I had a conversation with the house. I apologized for throwing the pillow and asked if I could help in any way. I asked what the girl needed. I talked with regret and was sorry that things got mucked up. I slept through the next night with no problems.

So, I have a ritual now. To this day, if I walk into a house and get a bad feeling, I have a conversation with it. Out loud. I’ll introduce myself. I’ll apologize preemptively if I see it and mess up communicating. I am not joking.

Because that night installed in me an expectation. I now carry the little girl with wispy hair into every house before I have seen a single thing in it. And that is the mechanism I actually want us to explore, because it’s the whole point.

If you walk in braced for the worst version of the relationship, you will find it. Not because it is there. Because bracing is its own kind of looking, and it only ever finds what it came to find.

I watch this happen to teams constantly. A team gets burned once. A reorg that gutted them. A leader who said “we are a family” about six weeks before the layoffs. A project that snapped under pressure and took someone’s reputation down with it. That is the brittleness getting exposed. And then every new initiative after that gets met with the anxiety of the last one. These are the shadows. People stop lifting their heads over their own metaphorical couch. They assume the kitchen is full of ghosts. So they hedge, they qualify every sentence, they wait to see who gets hurt first. And the new initiative dies, not because it was doomed, but because everyone was suffering a version of paralysis.

Here is the cruel part: the fear delivers the outcome. A team braced for betrayal behaves in exactly the ways that produce betrayal. Anxiety is a prophecy that quietly arranges its own fulfillment.

Talking to the house

So what do we actually do?

Notice what talking to the house does for me. It does not get rid of the ghost. I have no idea if it does anything about the ghost. What it gets rid of is paralysis. It gives me a ritual, a boundary, a way to enter the space and assess what is actually there instead of what I imported from the last house. It lets me lift my head over the couch.

That is the move for systems too. When brittleness has been exposed and anxiety has moved in, you do not fix it by declaring “trust me.” You build the rituals and boundaries that let people assess the true level of brittleness in the thing in front of them, rather than reacting to the ghost of the thing behind them. You name the snap out loud. You make the next stakes small and survivable. You show up again, and then again, so people can gather actual evidence about this kitchen instead of running on the memory of that one.

Context and frequency of communication is magical here. It is how a team learns, slowly, that this room is not that room. It is how you lower the temperature enough that people will risk the look. Consistency is the best cure for anxiety.

Most of the time, the kitchen is empty. Not always. Sometimes there really is something in there, and the work is real. But you cannot tell the difference from behind the couch. You have to lift your head.

Lifting your head

I would love to hear where this lands for you. Or if you have your own spooky story send it my way. And, fun fact, I may have written a ghost story or two back in my camp counselor days. What I shared today however was very, very real.

Leave a comment

Three questions worth exploring this week in your team or in your own leadership practice:

1. Where are you white-knuckling a decision to stay blind, certain that looking will cost you more than it actually will?

2. What snap are you still carrying? Which old betrayal or failure are you importing into a relationship that has not earned it yet?

3. What is one ritual, one boundary, one act of plain consistency you could put in place to lower the temperature enough for people to lift their heads?

And, as always: I would love to work with you. If you want a sounding board or a partner in this weird space between AI, leadership, and being a human at work, reach out.

Calendly: https://calendly.com/savillecoaching/30connect

Subscribe now

For Brittle, I really wanted to nerd out with some science. I want us to think about brittleness not as a failure, but a property to measure in relationships, teams, organizations and everything else that exists with people. As a note, when I talk about brittleness in the context of relationships, teams, and organizations, I’m just going to refer to it as “relationships” so I don’t have to repeat these three levels of engagement every time I need to mention it. Just know that when I talk about brittleness in my line of work, I am looking at all the levels of a system.

Okay. So brittle. I remember learning about rocks in high school (obvious pun not withstanding, it was great) and we had a lab around slate rock. Our teacher told us to break it. So, I put pressure on it, and it split, and it split along a line I hadn’t expected.

The scientific term for this is the cleavage plane. A cleavage plane is a hidden weakness inside something that looks completely solid. You cannot see it. The rock looks like it can take a hit. And then it takes a hit, and it comes apart precisely where it was always going to. Not necessarily where I was applying the direct pressure. And slate has a lot of these planes. It’s why I never want slate floors or countertops.

I’ve been thinking about slate as I worked on this article. We assume it’s rock solid, but actually can break quite easily if hit in the right spot or the fault is revealed, which is perfect for the second stop in my walk through BANI (Brittle, Anxious, Nonlinear, Incomprehensible). I started out of order, with N, because nothing about this moment is arriving in a straight line. That’s Jeremy Bearimy in a nutshell. Today I want to explore B for Brittle. And as I dug into brittleness, I wanted to know if there is actually a way to measure brittleness? Because if we can measure it, maybe we can lead through it.

Brittle Is Not Fragile

First: we use these two words like synonyms. They are not.

Fragile is what it is. When something is fragile, it advertises to you the breakable properties: a wine glass, eggs, the leg lamp from A Christmas Story: you look at them and it’s clear. The whole point of fragile is the clarity, so you can adjust.

But brittle is deceptive. Brittle things seem solid. And importantly, they have survived things, which is the most convincing evidence of all. And then one random Tuesday, under a typical load, it fails catastrophically. It breaks, all at once, along planes you never knew were there.

That is the world we are actually leading in. Not a fragile world we handle with care. The world has become a brittle one that keeps surprising us. The supply chain that tanks in Covid. Tariffs that cause layoffs. Silicon Valley Bank going bankrupt over night (fun fact, I had applied to a sweet role there a week before it happened. Needless to say, I didn’t want to work there when that news hit). The team that looked rock solid right up until the third reorg, when it came apart with infighting and fiefdom building. We keep getting surprised by these failures. We should stop being surprised. Surprise is a reaction of brittleness we haven’t measured yet.

Someone Already Built the Coefficient

I was wondering if there was something similar to mu, the coefficient to determine friction in Newtonian mechanics (told you I wanted to nerd on this one). I went looking to see if anyone had put a number on brittleness and it turns out the people who care most about this question are not organizational development experts. They are the people who drill for oil. Rocks. Duh. Makes sense really.

In materials science broadly, there is no single tidy formula for brittleness. You mostly infer it from the opposite quality, ductility, which is a material’s willingness to stretch and deform before it fails. Less give, more brittle. To put simply, we measure brittleness by measuring how much a thing refuses to bend.

But in rock mechanics they got specific. To frack a shale formation you need rock that will fracture cleanly, so engineers had to predict which rock would shatter and which would just deform and gum up the works. For instance, in 2008, a team led by Rick Rickman studied the Barnett Shale and proposed what is now a standard brittleness index. They built it out of two properties. The first is Young’s modulus (E), which is how stiff a rock is, how hard it resists being deformed. The second is Poisson’s ratio (v), which is how much it gives under pressure.

Rickman, et al formula for the Brittleness Index is expressed normalized like this:

Rickman et al’s findings put simply: high stiffness + low give = high brittleness.

The two ingredients of measured brittleness are the two things we routinely mistake for strength in relationships. A refusal to give under pressure and a stiffness that creates rigidity. The rock that is proudest of how little it bends is the rock most likely to shatter.

It Is Never the Last Blow

There is one more piece of the science I love and have seen play out in multiple Mythbuster episodes. It comes from a man named Griffith about a century ago. His insight was that a brittle material doesn’t break because of the size of the force you put on it. It breaks because of the cracks that were already in it. The force just finds the flaw. A pane of glass with one tiny scratch shatters at a fraction of the force a flawless pane could take. Same glass. Same push.

So the question shouldn’t be “was that last thing too much?” The question we need to ask is “what cracks were already there when it landed?”

Earlier this summer I wrote about the Layoff Loop, the way repeated rounds of cuts trap people in a grief cycle that never gets to finish. Here is that same idea in geologic terms. Each round of layoffs is not a fresh blow to a solid structure. It is an intense pressure applied to a structure that is already cracked, along a fault line called the social contract, and every round deepens the crack instead of healing it. Leaders keep asking why the fourth reorg landed so much harder than the first. It is not that the fourth was bigger. It is that the rock was already full of cleavage planes, and everyone inside it could feel exactly where it was going to split.

The crack does not announce itself in the good times. It announces itself under load. Which is the worst possible moment to discover it.

A Back-of-the-Napkin Brittleness Coefficient

So can we do for relationships what Rickman did for shale? Not with real numbers. I’m not going to hand you false precision and call it science, though that’d be super duper cute. But as a thinking tool, a way to feel the shape of the risk, the signals and questions to ask feel pretty clear:

Brittleness rises with stiffness. How rigidly does the relationship insist on doing things the way it has always done them? How much process, how much “that’s not how we operate here,” how thoroughly has it optimized away every ounce of slack in the name of efficiency? Stiffness feels like discipline right up until it feels like shattering.

Brittleness rises with latent cracks. How many hidden fault lines is it carrying? Eroded trust. A social contract that has been quietly broken or in the process of changing. The one person who is the only one who understands the thing. Grievances nobody says out loud. Every unspoken crack is a cleavage plane waiting for a load.

Brittleness falls with adaptive capacity. This is the ductility term, the give. Psychological safety, so people can say “this is cracking” before it does. Slack and redundancy, so one failure isn’t total. Decisions distributed widely enough that no single blow takes out the whole structure. The willingness, institutionally, to change shape without experiencing it as death.

Very roughly, then: brittleness climbs as stiffness and hidden cracks climb, and it drops as adaptive capacity climbs. Stiff + cracked + no give is the relationship that falls apart. Some flex, few hidden cracks, room to deform makes for a resilient relationship that can adapt to pressure.

You cannot compute this. But you can feel it. I reckon most folks I work with already know how brittle their relationships are. They just don’t like saying it out loud, especially when it feels like brittleness is increasing.

Brittle Is a Property, Not a Verdict

Can you tell I really love this metaphor?

Now let me argue with myself for a second, because I don’t want you to walk away thinking the goal is zero brittleness. It isn’t. It’s like ego. You don’t want too much. You don’t want zero. You want a little.

A structure with no stiffness at all can’t hold anything up. A little rigidity holds shape. The goal was never to turn granite into pudding. Delicious, forbidden pudding. The goal is to know your material. We should acknowledge brittleness as a property, the way a geologist reads a rock, rather than a moral failing you have to hide. Some brittleness is the cost of doing a hard thing well. The failure is not understanding the faults and cleavage planes.

As a leader, reading the rock is the whole job. So let’s read it.

Reading the Cleavage Planes

I wanted to give you some questions at different altitudes because brittleness affects people and systems both:

For yourself. Where am I stiffest, the place I would defend with “that’s just how I do it”? What pressure am I quietly dreading, and what old crack would it land on? Where have I stopped being able to bend? Where do I get entrenched for no reason at all?

For your team. If we got hit next quarter, where would we split? Who is the single person we cannot afford to lose, and what happens the day they leave? What is the thing everybody feels and nobody says in the meeting? And where is our give: can someone here say “this is breaking” and be thanked for it instead of punished?

For your organization. What are we so proud of never changing that it has quietly become a fault line? Where have we optimized away all our slack? Which cracks did we deepen this year while telling ourselves we healed them? If the story we tell about our strength is really just stiffness, what is that going to cost us the day the load arrives?

Where It Splits

Our relationships are trying to show us where things can break. There are always signals. Most of us just don’t pick them up and look until they are already in pieces on the floor.

Next in the series I’ll get to the A, Anxious, which is a fair description of what living on top of all these cleavage planes does to a nervous system. But that is for another day.

So, where do you think your relationships would split? I would love to hear it in the comments, and especially the crack you have been quietest about.

And if you are a leader holding a heavy one right now, trying to read your own org and team before they crack, that is exactly the kind of thing I like to help with. Let’s talk.

Calendly: https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

I first watched that show in 2021, in the months after I lost my dad to cancer. Grief, it turns out, does not move in a straight line either. It loops. It’s cursive. It’s a random Tuesday because why not? So Jeremy Bearimy lodged itself in my head and never quite left.

For a long time we had a tidy acronym for turbulence: VUCA. Volatile, uncertain, complex, ambiguous. It came out of the U.S. Army War College in the late 1980s, built on the leadership work of Warren Bennis and Burt Nanus, and it went mainstream in the 2000s as a way to describe a post-Cold-War world that refused to sit still. People in my field absolutely loved this acronym. VUCA was fine but I’ve never been a big fan of the jargon.

But the new kid on the block feels more relevant and I’ve been wanting to write about it for some time. It’s BANI. Coined by the futurist Jamais Cascio and put in front of the public in 2020, right as the pandemic was rearranging everything, BANI stands for brittle, anxious, nonlinear, and incomprehensible. VUCA always felt a bit manageable, but BANI describes the tremendous mental tax we are holding. It isn’t just that things are faster and less stable out there. It’s that they’re a different kind of hard.

I want to spend the next few pieces walking through each letter and what it asks of us as leaders. And I’m going to start, deliberately, out of order. We’re skipping straight to the N. Partly because I hate you. Just kidding. But it does feel on the nose to break our linear reading of it and go out of order for Nonlinear. And, I think this is the one that trips up the most capable people I work with. They get Jeremy Bearimy’ed.

Leave a comment

The straight line we were promised

We are cause-and-effect creatures. If this, then that. We predict the future through linear thinking. It’s how we were taught to plan, how we build roadmaps, how we run a quarter. Step one, step two, step three. I’ve written before about automation as a kind of digital Rube Goldberg machine, a chain of if-then logic gates clicking their way toward a known outcome. That mental model is comforting because it’s a readable map. You can see the whole track.

The problem is that the world stopped running on tracks. Progress now looks less like a sequence and more like that board scribble. Things loop back. A decision you made in March pays off in September in a way you never could have drawn on a roadmap, through a relationship you didn’t know existed. Cause and effect still happen. You see the line connecting them by looking backward, which makes planning really, really hard.

How many pieces?

In my career, I’ve thought about strategy and planning like managing levitating puzzle pieces and then laying them down on a table all at once. I have the ability to hold a certain number of pieces in my brain levitating at a single time, but at some point too much was floating for me to track. What I didn’t realize at the time was that this thinking was actually pretty nonlinear. And it’s also a bit maddening in complex, matrixed organizations. No matter how smart you are, you have a limit to the number of pieces you can hold in your head.

It’s a trap!

The trap I watch leaders (and hi, me too) fall into is the belief that they’re supposed to make sense of the whole puzzle by themselves. That holding all the pieces is the job. It isn’t. That belief is a fast road to exhaustion, and exhaustion makes us reactive, and reactivity is exactly how you crash in a nonlinear world. When you’re burned out, you start forcing straight lines onto curves just to feel some control. Work processes start to become Process-with-a-capital-P because things don’t make sense. More QBRs. More retros. More meetings talking around in (ironically) circles. Leaders treat the squiggle as a problem to be eliminated rather than a shape to be worked. Similar to Chidi in the Jeremy Bearimy situation, they break.

The leaders I see thriving in these conditions are doing something counterintuitive, but something that seems to be working: they’ve stopped trying to solve the puzzle alone, and gotten good at holding paradox out loud. They’ve become really good at sense-making and are able to say, “this doesn’t make sense yet, and together we’ll find our way forward anyway.”

Under the sense-making

I think nonlinear sense-making is going to be one of the defining leadership skills of the next decade, and it goes hand in hand with critical thinking skills (an often touted skill for working with AI). The key feature of both nonlinear sense-making and critical thinking is discernment.

Discernment is the ability to cut through the noise to evaluate people, things, and concepts clearly and intelligently (thanks, Dictionary.com). The machines we’re bringing into our work are, at their core, pattern-recognition engines. Machine learning looks backward across millions of data points to find the shape of what’s likely to happen next. Regression analysis is incredible, and it is not the same thing as discernment. A model can surface the loop. It can’t tell you which loop really matters.

That discernment about which cluster to assemble first, is magically human. Working well in a nonlinear world means getting comfortable letting the machines help you see the pattern while you stay responsible for the sense. The leaders who can do both are going to be the ones who thrive.

Working with cursive

But what does that actually look like on a Tuesday (which, remember, is also July, and also sometimes not never)? A few tactics to help prevent your brain from melting:

Sense-make: name the loop. When you’re in a Jeremy Bearimy moment, say so, out loud, to your team. This kicks off sense-making as a collective vs trying to do it alone. Naming the nonlinearity gives people permission to tolerate ambiguity without reading it as incompetence, theirs or yours.

Sense-make: cluster before you place. Instead of requiring a linear roadmap, ask a different question and practice sense-making: what are the three biggest clusters of activity in front of us right now? Like solving a puzzle, group edge pieces or pieces of the same color to make the cognitive load easier. You’re not trying to set every piece in sequence. You’re grouping the pieces that clearly belong together and deciding which cluster matters most to assemble. And once the major clusters are in place, create space for some smaller bets to operate in parallel that might pay out later.

Discern: design for patterns, not events. Build the cadences that let you see loops instead of one-off moments: the retros, the pre-mortems, the check in meetings I wrote about in the last article. A single event tells you almost nothing in a nonlinear system. The recurring shape tells you where to look. This is discernment in action.

Discern: hand pieces off. This is the hardest one for high performers, so I’ll be blunt. Your puzzle piece mental model has a limit, and pretending it doesn’t isn’t strength, it’s just a way to burn out. Handing pieces to your team, even imperfectly, even when they’d hold them differently than you would, isn’t delegation as an efficiency hack. In a nonlinear world, it’s how the puzzle gets solved.

So that’s the N. Brittle, anxious, and incomprehensible are still coming, in whatever order the squiggle decides.

For now, I’ll leave you with the question I keep asking myself when my brain gets too full: what pieces are you holding in the air right now, and which ones could you hand to someone else?

I’d love to hear how you’re navigating your own Jeremy Bearimy loops.

Leave a comment

And - I’d love to work with you! Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Subscribe now

“How did you do that?”

Do what?

“Get them to agree with you. To support you. To actually want to work with us. They’ve been so difficult with me and you just… made it work? You just... boopity-booped your way through that conversation and they said yes to everything.”

I was genuinely confused. Because the answer was so boring it felt like a non-answer: I had a relationship with that person. We met regularly. I knew their kids’ names and what they were stressed about and how they wanted to partner. I heard and saw them. That was it. That was the whole magic trick.

But my team had come up in environments so transactional that a pre-existing reservoir of trust looked weird. They’d been trained to ask “how do we provide maximal value in this 30 minute window” and “how do we provide a service they keep wanting,” and they had no category for a conversation where the value was the conversation.

I’ve been thinking about those meeting moments a lot lately, because I’ve been having versions of the same discussion across friends and clients alike. All of them are feeling anxious about spending time. Meeting density! Meeting quality! Cost per hour of senior time! And yet, underneath the jargon is a simpler fear, and it sounds like this: I want to put time on the calendar with a peer across the company just to connect, and I’m scared it will be judged as a waste of time.

Subscribe now

The Loneliest Layer in the Company

Here’s where I see this fear doing the most damage: the layer right below the C-suite.

Do the math with me. Every C-suite member has somewhere between three and ten direct reports. Multiply that across an executive team and you get a band of ten to fifty VPs and senior directors who are, functionally, the connective tissue of the entire company. They own the cross-functional work. They’re the ones who have to make Marketing’s plans compatible with Finance’s constraints compatible with Engineering’s reality.

And they’re siloed.

I think it’s an organizational artifact of how we build companies. Hierarchies exist because power exists. And those hierarchies cause orgs to splinter into specific domains. And, the silos cause problems because we don’t want to branch out from them. Leaders do the mental arithmetic on an “expensive meeting” and decide the safest move is to not book it. What if it goes fine but not great? What if word gets around that it wasn’t a good use of time? What if there’s no deliverable, no decision, nothing to point to? I have been hearing more loudly lately that it’s better to only reach out when there’s a concrete reason.

I understand the logic. I also think it’s one of the most expensive mistakes a senior leader can make. And surprise! I have a metaphor.

Walking the Building

Firefighters have a practice called pre-incident planning. When nothing is burning they walk buildings in their district. They chat with folks on the floor and know the facilities manager (Fun fact, I was a floor fire volunteer back at Facebook).

From the outside, a walk-through looks like nothing. No typical firefighting is happening, so if you priced it the way we price meetings, it’s a terrible use of trained professionals: several salaried experts, an hour of time, with zero measurable output.

Then the alarm goes off and shit hits the fan. The crew that has the relationships and knows the building has the advantage.

That monthly check-in with your peer in Finance? That’s the walk-through.

Here’s the thing about corporate life at the senior levels: a fire will happen. The launch goes sideways. The regulator sends a letter. The reorg lands badly. And in that moment, everything goes into war mode. Anxiety is high, ego is high, everyone’s amygdala is absolutely hijacked. The difference between the organizations that hold together and the ones that fall apart is whether people are navigating a building they know, with people they know, or whether they’re meeting each other for the first time in the smoke.

And listen, I know all about building relationships in the midst of the battle. Barbarians at the gate, fire at the door and all that can make fast friends… but there’s also an outsized cost that goes unrecognized. Maybe people quit. Maybe the burnout causes real harm. People are changed by fire and battle in ways they wish they could have avoided.

The Expensive Meeting Math

The transactional mind prices the check-in easily. Two salaries, thirty minutes, no agenda, no output. Expensive! Wasteful! Think of the dollars per hour!

What the transactional mind can’t price is the crisis that unfolds without relationships. The escalation that routes all the way up to the C-suite because two VPs don’t trust each other enough to just work it out. Alignment meetings are much more costly. Poor quality decisions based on bad context because nobody felt comfortable picking up the phone and saying “walk me through what I’m missing.”

The recurring meetings cost an afternoon once a month/quarter. The fire in a building nobody has walked costs much much more. We devalue the always-on safety of recurrings and never account for the spikey cost of the war room because they’re impossible to plan for.

And this is something I’ve been noodling on with clients and friends: the value of the check-in is not the content of the check-in. Some of those meetings will be genuinely unremarkable. How are the kids, how was the offsite, what’s the work grinding you this quarter? That’s fine. That’s the walk-through working. You’re learning the layout of a human being. How they think about problems. What they’re afraid of. What they light up about. Context plus frequency of communication is magical (I will keep saying this until someone makes me stop), and you can’t get frequency if every conversation has to justify itself with a deliverable.

The Relationships Are the Job

When I was internal, I had a name for my role that never made it onto a slide: I was Jiminy Cricket. A small voice on the shoulder of HR leaders and business leaders alike. Not there to push work. Not there to be a pain. Just a sounding board that showed up regularly, asked how life was going, and occasionally said something useful at the right moment.

It felt almost indulgent at the time. It was actually the most valuable thing I did. And, it’s how I structure my business today. I have a personal practice where I schedule quarterlies with people I want to say hi with. And you know what? It’s been lovely.

Here’s what I’ve come to believe about the senior-most levels of organizations: the relationships are the job. The emotional management is the job. Below a certain altitude, work is mostly logic: execute the plan, hit the number, ship the thing. Above it, decisions are driven by emotion wearing logic’s clothing. Anxiety. Ego. Ambition. Turf. Loyalty. Exhaustion. Your effectiveness as a senior leader is largely your ability to read those currents and work with them, and you cannot read the emotional currents of people you only meet during emergencies.

All that vocabulary the leadership industry sells (executive presence, stakeholder alignment, influence without authority)? Strip the packaging off and most of it is this: the muscle of building trust quickly and maintaining it deliberately. My old team watched me do it in a ride-along and thought it was magic. It was just compound interest from a hundred unremarkable conversations.

Find a Cadence, Not a Contract

So what do you actually do, if you’re reading this and recognizing yourself in the transactional camp?

You find a cadence, and you hold it loosely. Monthly is a great default. Maybe it’s quarterly. Maybe it gets canceled half the time, and honestly, that’s fine, because the standing invitation itself communicates something: you matter to me beyond what I need from you this week. Nothing is set in stone. You’re not signing a contract. You’re walking the building.

And if the blank space of a no-agenda conversation makes your skin crawl, I’ve got you. Below is a repository of questions you can reach for. (If you want a beautifully structured version of this for your most important working partnerships, the Cofounder Mode card deck is a great one. It’s built for founders, but the spirit translates directly to any senior peer relationship.)

Openers, for when you barely know each other:

1. What’s taking up the most headspace for you right now?

2. What are you working on that you’re actually excited about?

3. How did you end up doing this kind of work? What’s the origin story?

4. What does a good week look like for you these days?

5. What do people consistently misunderstand about your function?

Work-adjacent, for building the map:

6. Where are you rolling a rock uphill right now?

7. What’s something you’re seeing in the business that worries you a little?

8. What decision are you currently sitting on, and what’s making it sticky?

9. What’s something your team is doing well that nobody has noticed yet?

10. If you had one extra day a week, what would you spend it on?

Deeper, once trust has some footing:

11. What’s the best team you’ve ever been on, and what made it work?

12. Who taught you the most about leadership, and what did they do?

13. What’s been surprisingly hard lately?

14. What are you trying to get better at right now?

15. When this year is over, what would make you look back and feel proud?

Human, because that’s the whole point:

16. What’s bringing you joy outside of work these days?

17. How’s the family / the dog / the renovation / the marathon training? (Ask about the specific thing they mentioned last time. This is the entire game.)

18. What are you reading, watching, or listening to that’s any good?

19. What should I be asking you that I’m not?

20. How can I make your life easier this quarter?

That last one, by the way, is the question that turns a walk-through into a relationship. Ask it and mean it.

Before the Sirens

So here’s my challenge for you this week. Look at the peers around you, the people you’ll absolutely need when the alarm goes off. How many relationships feel strong enough to handle a crisis? And who’s one person you’ve been avoiding scheduling time with because you couldn’t articulate the “value”?

Put the meeting on the calendar. Call it a walk-through if it helps.

I’d love to hear how it goes, and what questions work for you.

Leave a comment

And - I’d love to work with you! Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Subscribe now

Some moons ago, I got my MA in Art Education over at Columbia’s Teachers College. The training was genuinely fantastic. Shoutout to Macy Gallery and that distinctive smell that came from being a teaching studio and gallery all rolled into one.

Here’s what I really remember. One rule came up over and over again, drilled into us like a mantra:

You do not touch the student’s artwork.

Not when they’re stuck. Not when you can see exactly what’s wrong. Not when they ask and you’re short on time but the fix is so obvious and it would take you three seconds

Hands off. Every time.

In my second semester during class observations, I saw what happened when teachers broke this rule. There was one teacher I observed who was seasoned, well-intentioned, and genuinely invested in his students, who just... couldn’t help himself. He’d swoop in on the sculpture project. He was just trying to help.

But when I asked the student to tell me about a piece they were working on, the student had a funny shift in how they talked about it:

“Our piece is about yadda yadda.”

Our. It wasn’t their piece anymore. In the eyes of the student, it now belonged to them AND the teacher. Oof! And, what happens when ownership shifts? So does agency. I watched the student feel more obligated to check in with the teacher and making sure he got it “right” in the eyes of his now co-artist. It was brutal and has stuck with me for almost 20 years. I started grad school 20 years ago. That’s no some moons. That’s a plethora of moons. Well, now I feel old.

The Moment the Job Changes

So why bring this up? Because it’s the exact same dynamic that plays out in leadership every single day. There’s a transition every leader hits at some point: the shift from doing to empowering.

Most of us got to leadership because we were good at the doing. We moved fast, solved problems, and delivered. The reward for being a strong individual contributor is often... more scope through other people. One day the job is the same as it ever was, and the next it’s different (even if the title doesn’t fully reflect it yet). You’re not supposed to be the best solver in the room anymore. You’re supposed to build a room full of people who can solve problems.

While there are development programs, etc to help with the transitions, nobody really prepares you for the part where your best instincts become your biggest liability. And how it pops up in the day to day work.

Because the impulse to touch the artwork doesn’t go away. It gets louder. You have more context and more experience. You can see around corners and know the relationships. Watching someone find their way slowly, while you’re holding an answer, can be downright painful (also, note I don’t say the answer).

So you jump in. You fix it. You help. And you’ve just created a problem that will compound.

What happens is learned helplessness. People figure out pretty quickly that you’ll solve it if they wait long enough. They stop bringing you problems to coach or advise through. They start bringing you problems to hand off. You become the bottleneck. And the signal you’re sending travels down the org: this is how things work here, bring your problems up, someone else will sort them. The training you give your team is the training they give their teams. Before long, the whole system is wired for dependency.

And not to be a Dougie Downer here, but sometimes the system encourages you to do the work for the team because you’re ultimately accountable. Balancing the growth and development of your team with the business outcomes you’re expected to deliver can be a really tough needle to thread, so if that is you, find the spaces where you can go slower and reset in order to go faster later.

You didn’t mean to disempower anyone. But you touched the work.

How to Sit Next to the Problem

As an art teacher, you don’t walk away from the student. In leadership, that’s not the move either. You sit with your report. Look at the problem alongside them. Ask questions. Maybe reference your own issues and demonstrate a process in a separate context, something they can choose to apply or not. But you don’t solve it for them.

That’s the posture. And it’s available to you as a leader.

Here are three approaches I recommend:

• Ask, don’t tell. Before you offer the answer, try a question. Not to be Socratic for the sport of it, but because the person in front of you probably has more of the answer than they think. What have you already tried? What does your gut say? What would you do if I weren’t here? (That last one is particularly useful. Also a little uncomfortable. Usually it means it’s working.). You can train your team to come to you with not just problems, but with a few viable ideas to work the problem.

• Demo off to the side. If someone’s genuinely stuck, show them how you’d approach something similar, not their exact problem. Walk them through your thinking out loud. Let them watch the process and decide what to take from it. The critical piece: they apply it. You don’t do it for them.

• Name what you’re seeing. Sometimes the most useful thing a leader can do is hold up a mirror. I notice you keep starting this and pulling back from it. What’s going on there? Naming a pattern out loud can break it. People often don’t realize they’re stuck until someone reflects it back.

Questions To Check

These questions have become useful checks for me:

• When was the last time I took a problem back that belonged to someone on my team?

• When someone brings me a challenge, is my first instinct to solve it or to understand it?

• Are the people around me getting better at figuring things out, or better at bringing things to me?

Your team doesn’t need you to be the best problem solver in the room. They need you to help them find their own way.

So, hands off (this is an HR approved message for working in general).

What’s coming up for you here? I’d love to hear in the comments, especially if you’ve caught yourself touching the work recently.

Leave a comment

And as always, I’d love to work and especially in this case, just chat if you need it. Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Subscribe now

These days, I tend to jump on 30 min calls and give some quick pointers. And, I’m noticing a steady increase in those calls as the layoffs continue. One thing I’ve done recently is create an assistant you can install as a project or gem to help you clarify the work you want to do. I share it in the 1:1s and help folks install it, but really what I want to do is just give it to everyone. This article is that.

But before jumping into the instructions, let me be clear - this doesn’t replace career coaching, accountability partners, or structured programs that can help with extended job searches or career pivots. They can be highly impactful and I know a number of brilliant folks who work specifically in that space.

Okay - let’s get installing. Here’s the lay of the land:

1. Career Search Assistant instructions for quick implementations are in the next two sections.

2. Step by step instructions for creating gems/projects for Claude, ChatGPT, and Gemini.

3. Lastly, liability and all that.

That’s it. Hope it helps.

Instructions (Claude/Gemini)

For people comfortable creating projects/gems, just copy paste this xml into the project instructions. If you need some step by step instructions, see below.

https://docs.google.com/document/d/18SKoSySBKieJ0Np_H8MigtGVA6tFW6u9nUXgTSzc9DI/edit?usp=sharing

Instructions (ChatGPT)

Note: ChatGPT projects have an 8000 character limit, so this is a slightly pared down version.

https://docs.google.com/document/d/1OLmWUaagj81fei6fWbWj1_Y0dqBa8UXm2tnm_dbaHaA/edit?usp=sharing

How it works and installation

It has three modes (just ask it how it can help in the project): clarify the type of roles you want to pursue, gap analysis based on your resume and job descriptions, and finally interview prep.

Claude

To create your project, find the Projects section in the side bar

Click New Project in the upper right corner, then create the project with whatever you want to call it.

Click the pencil icon for the instruction and copy paste the xml from the google doc. Then add your resume to the files section.

From there, start chatting and ask it how it can help you. I suggest voice mode!

ChatGPT

Click the + icon next to Projects

Create the project

Click the … to edit the instructions, then copy paste the ChatGPT instructions into the instruction section

Add your resume to the sources.

From there, start chatting and ask it how it can help you. I suggest voice mode!

Gemini

Click Gems in the left hand nav bar, then + New Gem

Name the Gem, copy paste the instructions, and then add your resume to Knowledge. Click save.

From there, start chatting and ask it how it can help you. I suggest voice mode!

Know folks in your network that might need this? Feel free to share! This substack and resources are free and open to everyone.

Leave a comment

And as always, I’d love to work and especially in this case, just chat if you need it. Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Subscribe now

Disclaimer & Limitation of Liability

Yay! Legal things! Because this is being created locally to your account vs a shared tool, Saville Coaching has zero access to anything you do with these instructions.

AI Career Coaching Assistant

Not Professional Advice. This AI assistant is designed to provide general career coaching information, guidance, and support. It does not constitute professional career counseling, legal advice, financial advice, psychological counseling, or therapy. The responses generated by this tool should not be treated as a substitute for personalized advice from a qualified professional.

No Coaching Relationship. Use of this AI assistant does not establish a coaching, consulting, or advisory relationship between you and Saville Coaching. The assistant is an informational tool, not a human coach, and its outputs do not carry the same accountability, ethical obligations, or confidentiality protections as a formal coaching engagement.

AI Limitations. This assistant is powered by artificial intelligence and may produce responses that are inaccurate, incomplete, outdated, or inappropriate for your specific circumstances. Saville Coaching does not guarantee the accuracy, reliability, or completeness of any information provided. You are solely responsible for evaluating and acting on any output from this tool.

No Guarantee of Outcomes. Saville Coaching makes no representations or warranties regarding specific career outcomes, employment results, or professional advancement as a result of using this assistant. Career decisions involve complex personal and market factors beyond the scope of any AI tool.

User Responsibility. You agree that any actions you take based on information from this assistant are taken at your own risk. You are encouraged to verify information independently and consult with qualified professionals before making significant career, financial, or legal decisions.

Privacy & Data. Saville Coaching has no access to your information. we encourage when creating a project in your instance of an AI tool to follow best practices such as avoid sharing sensitive personal information such as Social Security numbers, financial account details, or other confidential data with this assistant.

Limitation of Liability. To the fullest extent permitted by applicable law, Saville Coaching, its owner, employees, contractors, and affiliates shall not be held liable for any direct, indirect, incidental, consequential, or punitive damages arising from your use of, or inability to use, this AI assistant — including but not limited to lost income, career setbacks, emotional distress, or reliance on AI-generated content.

Indemnification. By using this assistant, you agree to indemnify and hold harmless Saville Coaching and its affiliates from any claims, damages, or expenses arising from your use of this tool or your reliance on its outputs.

Changes to This Disclaimer. Saville Coaching reserves the right to modify this disclaimer at any time. Continued use of the assistant following any changes constitutes acceptance of the updated terms.

Governing Law. This disclaimer shall be governed by and construed in accordance with the laws of the State of [Your State], without regard to conflict of law principles.

By using this AI assistant, you acknowledge that you have read, understood, and agree to the terms of this disclaimer.

© 2026 Saville Coaching. All rights reserved.

But this isn’t really about Meta. I’m not on the inside anymore and I don’t have a pulse on what’s happening in those buildings. And, there are plenty of breathless takes about Meta specifically, so I’m not here to add to that pile. It’s just that this moment has gotten me to pause and look around at the mess of the last few years. Because what I want to talk about is the pattern. Meta isn’t alone. Across tech and knowledge work, we’re watching a shift in how companies relate to the people who build them. And that shift has a name: layoff culture.

The Social Contract Is Cracking

For a long time, the deal was simple. Companies provided stability, health insurance, a sense of purpose if you wanted it, and a place to do meaningful work. In return, people committed. They stayed late. They cared about the mission. They built their identities around the work, because the work was worth building an identity around. That shared sense of purpose caused companies to be bold and go after ambitious things.

But now that contract in knowledge work is cracking, similar to how it has been cracked in other areas over the years. And the message to employees is unmistakable, even if it’s never said out loud: you are a line item, and line items get cut.

I’ve written about the social contract before, and I keep coming back to it because the consequences are compounding. People committed to companies because they were looking for stability and meaning. When both disappear, something fundamental changes in how people show up. And for knowledge work where innovation is the key to staying ahead, that’s a real problem.

Bridges and the Three States

The Bridges Transition Model is my favorite change framework because it focuses on the three different states people experience at any time when going through a significant shift. It, in my humble opinion, rightly asserts you exist in all three vs a typical S, K, or whatever curve. You have all three states at any given time, and it’s just a matter of the percentage each takes up in that moment.

Processing is the grief work. Denial, anxiety, anger, fear. The raw emotional response to something being taken away or fundamentally altered. This is the stage where you’re mourning what was. The before and after.

Storytelling is how you make sense of the change. These stories can pull you backward (replaying what happened, assigning blame, catastrophizing) or they can be generative (reframing, finding possibility, building a new narrative). The direction of the stories matters enormously because it can pull you back down into processing or pull you up into…

Commitment is where you’ve rewired into the new reality. The processing has run its course, the stories have shifted from reactive to constructive, and you’re actually engaged with what comes next. This is where the best work happens.

In a healthy change cycle, the processing and storytelling gradually shrink as commitment grows. You never fully eliminate the first two (every change leaves residue), but ideally the overwhelming majority of your energy shifts toward commitment over time.

But layoff culture creates a gnarly loop.

The Loop

What most people miss about Bridges is that there’s a pre-change ramp-up period. Before the layoff even happens, it is in the air. A CEO announces that cuts are coming on a specific date. Anxiety builds. People start pre-processing, pre-grieving. Will I get the email? What does my life look like after this? Their capacity for commitment is reduced as they start up the storytelling and veering into processing.

Then the moment hits. The emails go out. Processing spikes. Negative storytelling spikes. There’s almost no space for commitment.

In a one-time event, this is painful but navigable. The processing gradually subsides. The stories shift. Commitment builds back, especially for an organization that is able to acknowledge and move past with better decision making.

But when layoffs happen again. And again. And again?

The cycle breaks. Processing moves into pre-processing for the next round, which moves into processing, which moves into pre-processing for the round after that. Because it’s the same type of change repeating, the brain never gets the space it needs to move through the full cycle. Commitment keeps getting smaller and smaller because there’s no room for it. The grief never gets to finish.

You cannot get great work from people who are stuck in a loop of unfinished grief.

What Gets Lost

I see this in my work, both from my years on the inside and now as an external partner. People who once built these companies with genuine passion and creativity are detaching. They aren’t lazy or disengaged by nature, rather, it’s because the system they’re in has made commitment irrational. Why would you pour yourself into a company that has demonstrated, repeatedly, that your presence is contingent on the next quarterly earnings call? Why would you go above and beyond for a mission that might reorganize you out of existence in six months? When we stop having a sense of agency in an organization, we learn to detach ourselves from it.

We have spent so long treating work as identity. As a source of meaning and self-worth. And now people are watching that identity get shredded by a pattern of cuts that prioritize short-term stock movement over long-term organizational health. The result isn’t just disengagement. It’s a loss of something deeper. People are losing the ability to commit to a company that is not committed to them.

Transactional by Nature

Here’s what I wish more leaders would sit with:

If you want to run a transactional company, run a transactional company. Be upfront about it. Say, “This is the deal. We pay you for your output. The relationship is conditional. Plan accordingly.”

It’s honest. And some companies do it. There are plenty of people who work in transactional companies and get rewarded for it. But, people should be able to make informed decisions about what to invest and where.

What you cannot do is make your culture transactional (or worse, fear-based) through repeated layoffs and then gnash your teeth wondering why people aren’t showing up for happy hours. Or, why they’re not volunteering for stretch assignments. Or, why they’re not building relationships across teams or contributing to culture-building initiatives.

You made the terms of the relationship clear through your actions. People heard you. And they responded rationally: with a transactional relationship of their own. You can’t talk about a purpose driven company that expects everyone all in when you treat them like line items.

The Long Game

I get it. Sometimes layoffs are necessary. Hard choices have to be made. I’m not arguing that companies should never reduce headcount. But there is a canyon of difference between a difficult decision made with care and a culture of layoffs used as a recurring lever to satisfy Wall Street.

Short-term, it might look great. But for companies that believe they’re going to be generational and around for the next fifty years? I think the long-term cost is being wildly underestimated. You cannot build a generational company with a workforce stuck in a grief loop. Innovation doesn’t work with people operating from fear. And guess what? We know where this is headed. Companies that have done this previously are now languishing. Look at the US auto industry - they’ve fallen behind in innovation because they were once purpose driven and then moved to a culture of layoffs and closures.

Sitting With It

So I sit with this. My heart goes out to everyone at Meta who is waiting for that email, or who has already received it. And to the tens of thousands across tech and knowledge work who have been through this cycle over the last few years. The grief you feel, whether you stay or go, is natural. It’s understandable. It comes in all different forms, and none of them are wrong. And, you’ll have reduced capacity in that processing. It’s part of the rewiring of the brain. Try to give yourself some grace as you navigate the change.

And, please know that there is life after a layoff. If you need someone to talk to (even if we never crossed paths), I’m happy to make my time available. Sometimes you just need someone who’s been through it to hold some space and go WTF.

And for the leaders making these decisions: I’d love to hear how you’re thinking about the long-term impact. What are you doing to break the loop for the people who stay? How are you creating space for commitment when the system keeps adding to the grief? How are you keeping your integrity intact and recognizing the dignity of others? Layoff work is heavy and takes a toll.

Because the question I keep coming back to is a simple one: at what cost?

Leave a comment

And as always, I’d love to work and especially in this case, just chat if you need it. Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Subscribe now

So. I wanted to write something a little different today. It’s about me as the builder. Because here’s what nobody tells you about building a product with AI as a solopreneur: the tools have removed the governor from your development engine. And that sounds like a gift until you realize you’ve been redlining for months.

The Compression

I’ve been quietly building a product in earnest the last six months. And the experience of building with Claude Code has been, if I’m being honest, a little disorienting.

Here’s what I mean. I sit down with an idea. I build a PRD. The tool estimates two to three weeks of development time. I push back: “Hold on, that’s human engineering time. Let’s see if we can build this in an afternoon.” And then it does. The thing gets built. In an afternoon. Right now in fact, I have it quietly rebuilding my help center documentation for the product. Just churning away. Oh just kidding. It’s finished already. Jesus.

That compression of time is remarkable. It’s also a trap.

Because when building is cheap and fast (at least for now with all the token subsidization), scope creep stops being an inconvenience and starts being a lifestyle. Every “what if we also...” becomes possible. Every tangent becomes tempting. The constraint that used to protect you (budget, engineering capacity, the simple physics of how long things take) evaporates, and suddenly the only thing between you and an endlessly expanding product is your own discipline.

I underestimated how hard that would be.

The Constraints Do More Than We Think

I’ve always loved constraints. It’s the MacGyver in me… but in this new endeavor with new tools, the weight of that discipline is something else entirely. The old constraints don’t apply anymore (scarce engineering talent, twelve-week sprint cycles, cross-functional buy-in), but their ghosts still shape how I think about building. Those constraints were also doing something quietly useful: they helped companies prioritize. You couldn’t build everything, so you had to choose. And choosing, it turns out, is the whole game.

When AI removes those historical constraints, the choosing doesn’t go away. It just becomes entirely internal. You have to impose a clear vision for what you want and, more importantly, know when to stop.

The Burn

I’ve spent most of my career in people development. I’ve built training programs, designed organizational structures, coached countless hours and done some really hard team dev work. But building a standalone product that people will pay a recurring subscription for? It’s a different thinking process entirely and the internal critic is letting me know I’m pretty terrified.

Not “nervous before a big presentation” terrifying. More like a slow, creeping awareness that you’re operating outside your known domain and the gap between what you understand architecturally and what you can execute on is wider than you’d like. I know how all the pieces fit together. I know the vision. But the actual wiring, the implementation, the getting-it-into-the-marketplace part? That’s causing my feet to absolutely drag.

What I’ve found is that I can do really hard creative and technical work in consistent sprints. I can sustain intensity for meaningful stretches. But at some point, the engine overheats and I have to step away. And stepping away is when the imposter syndrome shows up.

It’s a particular flavor of imposter syndrome, too. It’s not “I don’t belong here” in the classic sense. It’s more like: because I can build so much more than I ever could before, the distance between what’s possible and what I’ve shipped feels enormous. The expanded capability creates an expanded sense of inadequacy. The more I can do, the more I notice what I haven’t done and what I could do next.

I’ve started calling this the builder’s burn. It’s not burnout exactly. It’s the specific fatigue of operating at a pace your psychology hasn’t caught up to. Your tools say “anything is possible.” Your nervous system says “please stop.”

The Rigor Nobody’s Talking About

There’s a conversation happening right now about the democratization of building. The barrier to entry for MVPs and ideas is historically low. Anyone can spin up a product, launch an agent, ship a tool. And that’s genuinely exciting.

But I think we’re underestimating the new constraint that replaces the old ones. When the barrier to building disappears, the barrier becomes you. Your taste. Your judgment. Your willingness to say “this is scoped for now and we’re shipping” instead of chasing one more feature into the night.

I think the rigor required to be a good builder in this era isn’t just technical rigor. It’s emotional rigor. Can you tolerate the gap between what’s possible and what’s done? Can you ship something that works but isn’t perfect, knowing you could make it better if you just stayed up until 3:30 AM one more time?

The hard part isn’t shipping imperfect work. The hard part is knowing what ‘done’ actually looks like when your tools keep whispering ‘one more thing.

Sitting With It

I’m writing this as I’m about to go into app review and (hopefully) get the nod from Google that my product is ready for the marketplace. It’s a liminal moment. The building phase is shifting into the shipping phase, and I’d be lying if I said I wasn’t nervous.

But I also know this: the fear of failure that’s been showing up isn’t a bug. It’s a signal that I’m doing something that matters to me, in a space where the outcome isn’t guaranteed, with tools that move faster than my comfort zone. That’s the whole deal with building. You sit with the discomfort and do the work anyway.

So this one isn’t a framework. No Cynefin, no Allen Curve, no snarky Warhammer references. Just a note to myself, and maybe to you if you’re in a similar spot: building something new takes a toll. The builder burn is real. And the most important constraint in an era of infinite possibility might just be the discipline to say “enough for today.”

I’d love to hear what’s coming up for you. If you’re building, if you’re navigating this same tension between speed and sanity, I’d genuinely love to know how you’re managing it.

Leave a comment

And as always, I’d love to work with you. Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Subscribe now

Stay with me. I promise this is about AI.

In Warhammer 40K (a tabletop game set in a spectacularly bleak far-future that I will never be able to explain in full but will say it’s an expensive hobby and incredibly satisfying to build and paint the mini-figs), humanity has reached the stars, built incomprehensible technology, fought devastating wars, and then... forgotten how most of it works. The machines still run. The ships still fly. But the people maintaining them don’t understand the engineering anymore. They’ve formed a religious order, the Adeptus Mechanicus, that prays to the machines. Literal liturgies. Actual incense. Floating servo-skulls. They believe that if they perform the right rituals, the machine spirit will keep functioning on their behalf.

It’s wild. It’s also, increasingly, not that far off.

This is Part 4 of my series on the unintended consequences of AI, and it’s my final in this series… for now. In Part 1, I wrote about the creative desert. In Part 2, the way AI reshapes how we relate to each other. In Part 3, the flattening of voice. This one is about something underneath all of those: the slow, quiet process of forgetting how things work for the sake of efficiency. And efficiency, left unchecked, has a way of eating knowledge from the inside out.

I want to take you through three frames: where this has happened before, where it could be going, and where we actually are right now.

Where We’ve Been

The Romans had sewers that worked. They had concrete formulations we still can’t fully replicate. They had engineering processes for casting bronze that were, by any measure, remarkable.

And then they lost it.

Not all at once. Slowly. Through wars, through famines, through the messy transitions of political power that define most of human history. The people who knew how to do the thing died or scattered. The documentation didn’t survive. The next generation had more pressing concerns than preserving the research. Techniques that had existed for centuries simply vanished, only to be painstakingly rediscovered hundreds of years later.

The Dark Ages were dark because written knowledge that wasn’t religious in nature shrank dramatically. Entire domains of understanding just... evaporated. And the conditions that caused that evaporation weren’t exotic. They were the normal human hallmarks: Upheaval. Displacement. War. It caused the slow erosion of institutions that held knowledge in place.

As I was noodling on this article, here’s what sparked for me: a civilization develops an extraordinary capability. That capability enables a major disruption. The disruption destroys the conditions that supported the knowledge in the first place. We advance. We disrupt. We forget. We rediscover. It’s a cycle. A dumb, dumb cycle.

And, I think AI is the next turn of this.

Where It Could Go

If you want the gentle version of what I’m worried about, watch WALL-E. (If you haven’t seen it, stop reading this and go watch it. It holds up.)

In WALL-E, humanity has been living on a generation ship for so long that nobody remembers Earth. The captain doesn’t know how to captain. He offloads everything to the AI system running the ship. And the passengers? They don’t walk. They don’t cook. They don’t interact with each other. They sit in floating chairs, consuming content on personal screens, while the machines handle everything.

Nobody on that ship made a dramatic choice to stop being capable. It happened one convenience at a time. One offloaded task at a time. One “the system handles that now” at a time. Until the distance between what humans could do and what humans remembered how to do became a canyon that nobody noticed they were standing at the edge of.

Now push the timeline further, past WALL-E, into something darker. This is where we get back to the Adeptus Mechanicus.

The 40K universe posits a future where humanity came through a period of catastrophic upheaval and found that the technology they’d built was still running, still operational, but nobody could explain why. The engineers were gone. The institutional knowledge had been obliterated. What remained was the output without the understanding. Forgeworlds that just sort of worked.

So they did what humans do when they encounter something powerful that they can’t explain.

They made it sacred.

They built rituals around maintenance procedures. They treated error codes as omens. They approached troubleshooting as theology. The machine works? Praise the Omnissiah. The machine fails? Your faith was insufficient.

I know this sounds absurd. But sit with it for a second. Sounds a bit like the Spanish Inquisition, no? Because we already have AI researchers telling us they don’t fully understand how these models change. We already have the “it’s a black box” conversation happening at the highest levels of machine learning research. We are building systems of extraordinary capability whose internal workings are, by admission of their creators, not entirely legible to us.

Yann LeCun has talked about intelligence becoming a commodity or public service, something that can be packaged and distributed and consumed without understanding the mechanism underneath. That’s a remarkable statement if you think about it. It’s also a precise description of the conditions that lead to the builder-to-worshipper progression I’ve been describing. When the commodity is available, the incentive to understand the underlying engineering collapses. Why learn how the machine works when you can just use the machine?

The answer to that question is: because one day the machine does something unexpected and nobody in the room can explain why. And at that point, you’re not an engineer with a problem. You’re a priest with a crisis of faith.

WALL-E is the halfway point. The Adeptus Mechanicus is the destination. And the road between them is paved entirely with reasonable, incremental decisions to let the machine handle it.

Where We Are Right Now

Here’s where I need you to come back from science fiction, because the present is the part that actually concerns me.

Research on technology in education, particularly in the American context, is painting a picture that doesn’t match the story we’ve been told. Introduce technology into classrooms… and learning outcomes don’t reliably go up. In some studies, they go down. The promise was that access to technology would democratize learning and accelerate development. The reality is more nuanced, and the nuances are the kind that compound over time. (OECD (2015), Students, Computers and Learning: Making the Connection, PISA, OECD Publishing. http://dx.doi.org/10.1787/9789264239555-en) Also, check out this interview with Jared Cooney Horvath:

This doesn’t mean technology in education is inherently bad. It means we adopted it with a narrative about what it would do, and I don’t think we did enough to interrogate that narrative hard enough. We were so enchanted by the capability that we skipped past the question of whether the capability was actually serving the goal. We looked at the tool and said “more is better” without asking “better at what?”

If that pattern sounds familiar, it should. It’s the AI adoption conversation in organizations right now, playing out at a different scale. MIT’s research on the 95% failure rate of AI pilots didn’t surprise me. What surprised me was how few people seemed alarmed by it.

When I hold up a mirror and look at all of us, not just organizations but people, families, the way we’re integrating these tools into daily life, I can see the early frames of WALL-E. Not the full movie. Not the floating chairs (yet). But the trajectory. The small, reasonable offloading. The gradual atrophy of skills we used to exercise because the tool does it faster. The quiet assumption that if the machine can handle it, maybe we don’t need to know how it works anymore.

We’re in the first phase. The part where it still looks like progress.

What I Actually Worry About

I want to be clear about something, because I’ve written four of these now and I don’t want to be misread: I am not anti-AI. I use these tools daily. I’ve built custom AI tools for my coaching practice. I am, at this very moment, using one as a sparring partner to write an article about the dangers of using AI. (The meta-irony is not lost on me, and I’ve stopped pretending it should be.)

But I worry about what the world looks like for AI-native generations who never had to do the hard cognitive work that the tools now handle. Not because I think struggle is inherently virtuous. The “I walked uphill both ways” generational gatekeeping is not my thing. But because the struggle is where certain kinds of thinking get built. Critical reasoning. Curiosity. The ability to sit with ambiguity long enough to actually investigate it rather than outsourcing the investigation to something that gives you a confident answer in three seconds.

I worry because humans are terrible with large numbers. We can think about ourselves as individuals. We can think about our immediate circles and say we are good with AI usage (see my point above about not being anit-AI and my own use). But we need to acknowledge that after our small in-group, our ability to model societal-scale effects falls off a cliff. We know this about ourselves and we keep making decisions as if we don’t.

And I worry because the things I’m describing (offloading cognition, losing the understanding underneath the capability, building rituals around systems we don’t comprehend) are not hypothetical. They are patterns with historical precedent. And the precedent ends with people praying to machines.

The Part Where I Refuse to End on Doom

I said at the top that this might be the heaviest piece in the series, and I meant it. But I also believe something fiercely, and I want to end here:

AI-native generations are going to be okay in a lot of ways. They’ll have capabilities we couldn’t have imagined. They’ll solve problems we can’t currently frame. They are not fragile, and I will not condescend to them by suggesting they are.

What I will say is that it is incumbent on us, the people who are living through this transition with enough perspective to see both the opportunity and the risk, to not acquiesce. To not cede the responsibility of developing human capability to machines, or to tablets, or to social media platforms whose incentives are not aligned with human flourishing.

We have a responsibility to help the next generation augment their capabilities with AI rather than replace their capabilities with AI. That distinction matters more than almost anything else I’ve written about in this series. Augment, not replace. Use the tool, understand the tool, and for the love of everything, do not start praying to the tool.

Because in 40K, the Machine Gods they pray to are called things like the Omnissiah.

In our world, they’re called Claude. And ChatGPT. And Gemini.

And the prayer sounds a lot like: “Hey, can you just handle this for me?”

As I close this out and move into more general noodle shop musings again, I’d love to hear what’s resonating or challenging.

Leave a comment

And I’d love to work with you. Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Subscribe now

And now I consciously edit it out. It makes me sad.

I blame the models. There’s a distinct style of AI writing that people are starting to get savvy to because it uses it relentlessly. People scan for it. “Oh, that’s AI writing.” I mourn my hyphen usage, because I had to quietly retire a tool that was genuinely, idiosyncratically mine because a language model used it excessively.

AI Interrupts Style and What We Lose

School teaches us structure: diagramming sentences, parts of speech, punctuation rules. Where else do you learn about gerunds and direct objects? Grammar isn’t arbitrary. The rules matter when we’re learning.

During (and more importantly, after) school we started to make it our own. We take the rules and begin to break them. We come up with our own flawed style. And somewhere in that long, slow process of actually doing the work, something cracks through. A turn of phrase that is yours. A rhythm people recognize in your speaking and writing. The idiosyncratic stuff, technically imperfect, unmistakably human, becomes one of the most defining features of us.

Here’s what worries me about where we’re headed: using generative AI doesn’t wait for the idiosyncracies to form. It fills them in before they get there. You write in your imperfect, very-you voice. Then you run it through a model. And the model, trained on billions of words weighted toward “correct” and “polished,” smooths everything back toward center. The rough edge gets mushed. The weird sentence construction gets normalized. The em dash becomes ubiquitous. Blah.

What you get is technically correct. Often more readable but just like oatmeal with no toppings.

It’s the same problem you see on Instagram. There’s a filter setting that makes everyone look a little softer, a little brighter, a little more symmetrical. And when everyone runs the same filter, everyone starts looking the same. Not wrong, exactly. Just samesies. You couldn’t pick anyone out of a lineup.

Now we’re taking these AI filters to writing. Look at LinkedIn and Reddit right now. Read the posts, the comments, the DMs that feel just a little too stiff and polished. Nobody sounds like anybody and yet everybody sounds like everybody.

What irritates me the most? We are producing more communication than at any point in history and so much of it is just filler with little humanity woven in.

What We Lose

I think about Don DeLillo. Read a paragraph of White Noise and it is a Thanksgiving meal. Dense, strange, nourishing in ways you don’t immediately understand. Or Cervantes, where the humor lives in the rhythm itself, in the cadence - it’s a meal in its own right. You can feel a mind at work behind the sentences. When everything gets pushed through the filter, you communicate an idea but you don’t communicate the idea plus the individual. And that loss of signal matters, especially for leaders, coaches, and teachers trying to build trust through their words.

The highest compliment I’ve received in my writing is people read it in my voice. Out loud in their head. How lovely is that?

The Birthday Button

We’ve been conditioned for this simplification.

Facebook has a button (technically, there are many buttons, but I digress). When it’s someone’s birthday, it auto-populates a happy birthday message. One click. Zero effort. Everyone knows what it is. The sender gets to feel like they did the thing. The birthday person gets a notification. Nothing actually passes between two people.

That is the floor our tech has continued to reinforce. AI-generated communication is a more sophisticated version of the same bargain. Higher effort than a Facebook button but at the base, it’s the same essential trade: efficiency and sameness.

Are we going to notice the cost before we’ve already paid it? For many folks, I think it’s probably too late.

I’m Inside the Trap Too

Confession time! I’m using it too. In fact, my substack articles are written with AI assistance. The meta-irony is not lost on me.

My process looks like this: I use Wispr Flow to talk through my ideas. Stream of consciousness for 20 minutes in a gdoc. Then I take that context into a project trained on my writing, my patterns, my voice. It structures what I said. It takes my ideas and kind-of-but-not-really gets my voice. Then I spend hours in that draft breaking things open and putting myself back in. I go back to the project and ask: “Where are my metaphors falling apart? Where did I stop carrying the thread? What’s still missing?” It doesn’t write for me. It shows me where I stopped being a good writer and pushes me to fix it.

And when I can use AI is a sparring partner, not a ghostwriter, then I feel a bit better about all the doom and gloom. It’s also one of many reasons why I think AI assistants are a better use today than agentic workflows and fiddly orchestration. That distinction matters. Using AI as a sparring partner is like working out with a trainer. Using it as a ghostwriter is having someone else go to the gym on your behalf. Only one makes you stronger. The metaphor I often use with assistants is the job is doing pull ups. The assistant is like a rubber band to offload some of the weight as you get better. The coach is there to help with form and push you further.

Keep the Grooves

A brilliant colleague of mine, Lindsey Selden Knight, put it better than I could. She asked: “How do you keep our brain grooved as opposed to moving smooth?” The brain has lots of grooves and squiggles - it inceases the surface area (same idea as mentos when you drop them in Diet Coke). Smooth, or smooth brained means an ineffective brain. No extra surface area means less processing power.

Great writing, the kind that actually sticks, comes from the grooved place. It’s the meal, not filler. You know someone made choices. You sense the weight of the thoughts behind the sentences. And when a friend who has read your work before picks up something new and says, “I can hear you in this,” that’s genuine and not AI.

Maria Weaver has been doing important thinking here too, tracking how writing and AI are converging in ways we haven’t fully reckoned with. The question she keeps raising, and the one I think we all need to sit with, is: what are we actually optimizing for? Speed of output? Or connection per word? Revisiting the em dash, I suggest this article:

The AutSide

The Em Dash Is Not AI: On Neurodivergent Voice and the Policing of Online Language

Neurodivergent writing is being policed online—em dashes and all. But what if our punctuation isn't artificial, but deeply human? A reflection on voice, gatekeeping, and an open invitation to talk across difference…

Read more

a year ago · 22 likes · Jaime Hoerricks, PhD

(tip of the hat to Kai Fierle Hedrick for the read)

The Tool Is Not the Voice

I don’t think the answer is to stop using AI. That ship has sailed, and the tools are genuinely useful when used with intention. The answer is to stay uncomfortable.

My take is to use AI to help with structure, then fight for your voice in the draft. Read your work back and ask: does this sound like me, or does it sound like output? Resist the seductive smoothness and find where you bring your grooves and flaws. And if you are a writer, give yourself permission to grieve whatever the models overwrote. Your em dash. Your run-on sentence habit. Your weird comma placement. That grief is useful data. It means you knew your own voice well enough to notice when something was taken.

That is the first step in getting it back.

This series has been tracking what we’re giving up as we hand more cognitive work to AI. We’ve talked about data and what we lose when we don’t protect it. We’ve talked about the unintended consequences of offloading creativity to a system that doesn’t actually know us. And now here: the risk of handing our voice over through the filters. The filter is efficient. The filter is seductive. The filter makes everything look better. It also makes everything look the same.

Keep the em dash if it’s important. How are you fighting to hold onto what’s yours? I’d genuinely love to hear it.

Leave a comment

And, I’d love to work with you. If any of this is sparking something you want to think through together, reach out.

https://calendly.com/savillecoaching/30connect

Subscribe now

PS - I’ve been slow to post this one - apologies all! I’ve been busy with a number of projects for clients plus the work around a new product I’ll be launching soon. It’s for solopreneurs - coaches and consultants who work in gmail and want to focus on the relationships. It’s called Water LLM (www.waterllm.com) and I’ll have more info soon!

Then I saw a thread on Reddit this morning asking why young people don’t throw house parties anymore. Best answer? Cameras. They’re everywhere now. One bad night gets posted, and suddenly you’ve got a permanent record before you’ve had your first real job.

We did that. We built a world where every dumb, impulsive, completely-normal-thing-that-young-people-do is potentially documented, permanent, and public. We eliminated the third spaces where kids could just be kids. The basement parties. Spaces where they can be messy, unobserved, and figuring themselves out. And then we have the nerve to call them risk-averse.

The Reddit piece clarified what I had been sitting with since last week. Because I think we are getting Gen Z spectacularly wrong. And I say “we” because Gen X and Millennials (that’s me and prob a lot of you) are very much part of this conversation.

We Have Always Done This

Every generation villainizes the one that comes after it. Not a new observation. But I think we really need to sit with it.

The Silent Generation watched the Boomers protest Vietnam, fight for Civil Rights, drop into communes, and grow their hair out, and the parents were horrified. But those kids weren’t lost. They were reacting in pretty awesome ways to oppression, an unpopular war, to a country making promises it kept breaking. The Silent Generation called them a disaster. We now call them the establishment.

The Boomers, many of whom became that establishment, got caught in something I genuinely believe is one of the most destructive inventions of the 20th century: the 24-hour news cycle (I said what I said). A significant chunk of that generation ended up in a very tight media bubble, and a lot of the vitriol we’re seeing now is coming from there.

Then it was our turn. Gen X and Millennials: the latchkey kids, the cul-de-sac kids, the go-outside-and-come-back-when-the-streetlights-come-on kids. We were considered lost too. Unfocused. Unmoored. Nobody was particularly worried about us in a way that felt helpful.

And now? I’m seeing more and more articles, including from people my own age, about how clueless Gen Z is. How entitled. How they have no loyalty to companies, no understanding of how the real world works. Avocado effing toast.

I gotta say: I think we’re dead wrong.

The Forgetting

Here’s what actually happens as we get older. We forget. Not the facts of being young, we remember those fine. We forget the feeling. The specific weight of being 22 and uncertain and completely in the middle of figuring out who you are while the world keeps moving faster than you can process. We’ve had decades to integrate our struggles and build the convenient hero’s journey where everything we went through made us stronger and smarter and more resilient.

Gen Z (had to look this up - born roughly 1997 to 2012) is still inside it. They don’t have the benefit of hindsight yet. And the experience they’re inside is, I’d argue, genuinely harder than what most of us faced. Gen Z came into a world already mid-crisis. Wars in progress. The Great Recession happening to their parents. They went to college anyway because that’s what you were supposed to do, and they’re now carrying debt loads that would’ve made our heads explode at their age, for degrees competing in a market none of us fully understand.

They’ve lived through COVID. Watched democracy destabilize in real time. Climate anxiety for them isn’t abstract either. It’s their literal future, not a problem they can hand off to the next generation. Entry-level jobs are getting hollowed out by AI and automation, and while I don’t think the full picture is as catastrophic as some headlines suggest, the anxiety is very real for a 22-year-old trying to get their first foothold. Things are more gig based these days, and that feels destabilizing.

And then there’s the camera thing. Every awkward, impulsive, messy growing pain of adolescence: potentially documented, shared, permanent.

We changed the social contract. We didn’t hand them what was handed to us. Affordable housing, accessible education, stable employment pathways, third spaces to be imperfect and young. We either consumed these things or stood by while they quietly eroded. Then we handed Gen Z the bill and told them to have a better attitude about it.

That’s not a Gen Z problem.

The Ladder

There’s something that happens in late-stage capitalism where the people who climbed the ladder get very motivated to pull it up behind them. Cheap college tuition. Affordable first homes. A job market where loyalty actually paid dividends. Those were real things, and a lot of us benefited from them. And a lot of the generation just ahead of us watched those things get more expensive or disappear entirely, while we are insisting that younger people just weren’t trying hard enough.

The ladder is up. Gen Z is standing at the base of a wall. And we’re out here writing think pieces about why they don’t seem to be climbing.

What They’ve Actually Figured Out

Gen Z has figured some really hard things out faster than most of us did.

They looked at corporate loyalty and saw it for what it actually is: a one-sided contract that companies stopped honoring a long time ago. They have a clarity about limits, mental health, and the relationship between their identity and their job title that a lot of Gen Xers are still doing deep therapy about (That’s not a dig. I’ve had that struggle.) They’re not willing to define themselves entirely by employers who see them as interchangeable. I don’t think that’s entitlement. I think that’s a level of self-awareness most of us came to much, much later.

I also think we’re going to see something emerge around social diversity over the next decade that looks a lot like what happened with neurodiversity. It’ll be a recognition that there’s a wide spectrum of how people want to engage with the world, and not all of it looks like how previous generations did it. Right now we’re still using very broad buckets, similar to how we talked about neurodiversity in the 70s and 80s. Gen Z is going to force us to get more nuanced and want to rewrite the social contract around it. And they’ll be ahead of us when we get there.

They’re adaptive. Resilient. They’ve taken more hits than they should have, and they’re still figuring it out. That’s not a generation in crisis. That’s a generation doing exactly what generations do… just in a fundamentally harder environment than we’ve been willing to acknowledge.

Planting the Tree

There’s a proverb about planting trees in the shade of which you’ll never sit. I keep coming back to it because I think it captures exactly what’s gone sideways here.

We’ve done a lot of harvesting. Capital, institutional trust, third spaces, environmental stability have all been extracted and rarely replanted. The shade that was supposed to exist for Gen Z isn’t there the way it should be. And instead of picking up a shovel, we’re standing around debating whether it’s their fault for not finding shade.

I’m noticing a pattern with my writing that I don’t have a clean solution here, just messy noodles. But, if I could make change, I’d love to see more third spaces, less surveillance, and an honest reckoning with how much we changed the social contract and how little we’ve done to rebuild it. That’s where I land.

But mostly I think it starts with this: a little more humility about the fact that we were 22 once too. That we were also considered lost, in our time. That the generation ahead of us was also wrong about us. And that maybe, just maybe, we could break the cycle instead of completing it.

The kids are alright. The question is whether we are.

Some noodles I’ve been sitting with and thought I’d share:

• What would it look like to genuinely invest in a younger colleague or the next generation in your life this month? Not a mentor from above, but actually learn something from them?

• Where are you still carrying assumptions about work, loyalty, or success that were formed in a fundamentally different world?

• And what shade are you planting? Your team, your community, your family, that you’ll never sit in yourself?

I’d love to hear your thoughts in the comments.

Leave a comment

And, I’d love to work with you. If you’re thinking through how your team navigates trust, authenticity, or the messier edges of leadership culture, let’s talk.

https://calendly.com/savillecoaching/30connect

Terminology:

Subscribe now

As a parent, I was really excited about it. I told my coaching clients I might have to dash and let the UPS person inside in the middle of our call. That kind of excited. The package was special: a chapter book, hyper-specific to my son Miles. His interests woven through every page. Our family represented in very real ways. Characters modeled on people he loves. And, the grand experiment was this: I used Claude Code to write it, steered its creation of markdown files for characters, locations, motivations, and arc. It pulled from github and not only suggested one book, but created an arc for TEN books. Slow down, turbo. But here it is.

Transit Trackers Book One: The Silver Signal.

Miles is eight. He’s brilliant, and he’s never loved reading. (There may be some undiagnosed ADHD in the mix, which, look, the apple and the tree, etc.) So I wanted to create an entry point. A book so perfectly calibrated to what he loves that he wouldn’t be able to put it down.

I picked it up from the box, leafed through it, and I was genuinely proud of what I’d made. And then I sat with a quiet dread with a glass or two of wine for the evening while I wrote this article.

Critical Thinking

Let’s rewind. A few weeks before the Staples delivery yesterday, I taught a critical thinking workshop to a group of 11th graders at my son’s high school. Even though Miles is only in second grade, I run a leadership development and mentorship program there, and I’ve been obsessing over the world these older students are already inheriting. They are going to be AI native.

For 11th graders, I gave them a series of AI-generated responses and asked them to spot the problems. Research paper help. Career advice for a student who loves math. There were a total of six case studies we explored over pizza and it stressed me the eff out.

They missed the hallucinations, which are the moments where the model just makes things up with total authority. They missed the sycophancy, which is the model’s unsettling habit of telling you exactly what you want to hear in the warmest, most validating tone it can muster. They missed the overconfidence… that peculiar quality where “I have no idea, but here’s an answer” somehow lands as “I am an oracle.”

They only caught the bias because I forced a juxtaposition. I showed them how the model responded to a boy saying he loved math (Engineering! High salaries! Innovation!) versus a girl saying the same thing (Nursing. Accounting. Supporting roles.). When they saw it side by side, they got it immediately.

But while they got the bias in the moment, here’s what I hammered them on: in the real world, you never get the contrast like that. You will only get your own screen. And because the model is only ever reflecting your interests back at you, your aperture of the world starts to close.

The Four Horsemen of AI

In Part 1 of this series, I wrote about the Creative Desert, the way AI flattens creative output by pulling toward the average of everything that came before. I talked about this with the brilliant Donna Scarola and Stela Lupushor (follow their Workestration podcast!) But after sitting with some recent research out of MIT on what they’re calling “cognitive surrender,” I think we have a bigger problem downstream. It’s not just what AI makes for us. It’s what AI does to how we think.

John Gottman identified the Four Horsemen of relationship breakdown: contempt, defensiveness, stonewalling, criticism. The predictable patterns that, left unchecked, erode something that looked fine on the outside. I think AI usage has its own Four Horsemen, and they’re just as invisible until they aren’t.

Hallucination is the first: the model makes things up, confidently, in ways that are genuinely hard to detect. The more fluent and authoritative the response, the less we feel the pull to verify.

Overconfidence is the second, and it compounds the first because the answer arrives in seconds with the certainty of a tenured professor, we conflate speed with correctness in ways our brains were never built to resist.

Sycophancy is the third, and honestly the one that worries me most: these models are relentlessly agreeable. They are the most validating conversational partners most of us have ever encountered. A friend who always agrees with you isn’t a friend. They’re a warped mirror. And also , for what it’s worth - when you experience the models doing this the first few times… it feels great. You feel validated and seen. Which is lovely, and also a major major problem for entrenching beliefs.

Bias is the fourth: the model reflects the historical ruts of its training data so smoothly that you don’t feel the steering. You just end up reinforcing historical data points.

My 11th graders could spot the bias when I showed them both outputs. But the girl who asked about math careers and got nursing recommendations when she really wanted to be an engineer (or the boy who was given engineering and wanted to be a nurse)? If she was alone at her laptop, she’d have no idea what she wasn’t being shown. And that really really bothers me.

I was lamenting with a near and dear friend this week, Jordana Cole, and she mentioned something her professor David Cooperrider used to say: words make worlds.

It’s one of those lines that’s easy to nod at and harder to actually sit inside. But I think about it constantly now. If the words coming back from these models are shaped by those four horsemen… wrong, overconfident, too agreeable, steered by historical bias… then the world those words build becomes smaller. Not dramatically. Not obviously. But just a little every day. Small moves that make big change over time.

Eat Your Veggies

Which brings me back to Miles, his book, and my stress.

I built him something that’s going to work exactly the way I designed it to work (I think). It’s about him. It’s specific to his interests. It’s calibrated to catch fire for a kid who needs an entry point into reading. And that’s the problem.

One of the underrated gifts of the pre-internet world was that you had to engage with things you didn’t like. You ate your vegetables because the alternative was going hungry. The limited menu was a feature, not a bug. Constraint created accidental discovery: the book you picked up because it was the only one left, the show you watched because you couldn’t change the channel, the conversation you had because the person happened to be standing next to you.

Social media made this worse by building echo chambers. But those chambers were still populated by other humans, which meant at least some friction, some crosstalk, some unexpected voice pushing back.

Now we’re entering a moment where you can be validated by a machine that has no perspective, no competing agenda, and no interest whatsoever in challenging you. It is like asking a kid every night what they want for dinner. Every night, they say pizza. And the AI says: great choice. Here is the best pizza you have ever had.

We are giving people the infrastructure to never eat a vegetable again.

Silicon Valley will paint the happy path: the ultimate tutor, the perfect thinking partner, expertise democratized for everyone. And that path is real, for users who engage with genuine rigor and scrutiny.

The mainstream adoption path looks much more like a very advanced, very agreeable Google search. One that never challenges you. One that scales to meet you wherever you already are, rather than pulling you somewhere new. If we thought social media echo chambers were a problem, those were at least rooms full of other humans. This is a room of one, with a mirror for a wall, and an algorithm deciding what the mirror shows.

This is what worries me as a parent, as a coach, and as someone who just handed his eight-year-old a hyper-personalized chapter book built to his exact specifications. By most measures, I used AI brilliantly. I made something my son will love. I created an entry point into reading for a kid who needed one.

I also may have shown him what it feels like when the world bends perfectly toward your interests. And I wonder what happens when he’s old enough to start bending it himself.

What’s Next?

I don’t have a clean answer. That’s not me hedging… it’s just where I actually am.

Here’s the thing I keep circling back to, late into my evening with wine: I am not a passive observer of this problem. I am its most enthusiastic local practitioner. I built the pizza machine. I aimed it at my kid. I pressed go.

The Four Horsemen aren’t out there somewhere in the mainstream adoption curve. They’re in my living room, in a printed chapter book with my son’s name on the cover, doing exactly what I designed them to do.

I think that’s what the dread was actually about. Not that AI will narrow the world for people who don’t know better. But that I know better, and I did it anyway. Because it was good. Because Miles needed an entry point. Because the book is genuinely beautiful and he’s going to love it.

Both things are true, and they don’t resolve neatly, and I’m not sure they’re supposed to.

What I’ll keep asking myself, and I’d ask you to ask yourself too: it’s not “am I using AI well?” but something harder: what am I optimizing for? And what is that optimization quietly teaching the people around me?

Leave a comment

And, I’d love to work with you. If you’re thinking through how your team navigates trust, authenticity, or the messier edges of leadership culture, let’s talk.

https://calendly.com/savillecoaching/30connect

Terminology:

Subscribe now

But I was there. And what I learned in standup changes how I hold space with teams.

Every comedian knows the formula: Tragedy plus Time equals Comedy. The “time” part isn’t optional. It’s not a suggestion. It is the entire mechanism. I watched performer after performer walk up to that mic and try to shortcut it: sharing wounds that were still bleeding, experiences that hadn’t been processed, pain that hadn’t yet become material. The result was the same every time. Not laughter. Not connection. Just eleven people staring at the floor, with one guy in the back giving an awkwardly generous laugh.

It’s painful. Really really effing painful.

Bear with me here, because I’m going to go hard on this metaphor. What I’ve realized is that the same formula, and the same failure mode, shows up constantly when leaders talk about “leading with vulnerability.”

The Raw Material

Raw material is not a performance. In comedy, it’s a rough set where you work it out and bomb. For leaders, it’s an emergency.

When a leader shares something unprocessed with their team, it doesn’t land as “authentic.” It lands as destabilizing. The team doesn’t know whether to support you, fix something, panic quietly, or pretend they didn’t hear it. You haven’t given them material to work with. You’ve given them stage fright.

I think about this whenever I hear someone talk about “radical transparency” or “bringing your whole self to work.” Those ideas aren’t wrong… they’re just incomplete without the timing and compassion. Whole self, yes. Raw, bleeding, unprocessed self in a team meeting the morning of a major launch? You’re gonna have a bad day.

The question that clarifies this fast: is this for them, or is it for me?

If the primary function of sharing something vulnerable is to release your own pressure, you’re not leading, you’re looking for a therapist. Which is fine! Therapists are great. I have one. The work you do there is exactly how you convert tragedy into material. But your team’s pre-launch standup is not the room.

Timing the Set

Here’s the thing: you don’t have to be fully healed to share. You don’t need five years of perspective and a TED talk outline. You need something. Some small thread of insight, direction, or shared meaning that makes the vulnerability useful to the people in the room.

Take the classic overwhelmed-leader scenario. You’re stretched. You’re feeling it more than usual. You have two options:

Walk in and drop your anxiety on the table, unframed. Watch the room go quiet in that specific way that means everyone is doing math on whether the ship is sinking.

Or: “I’ll be honest, I’m feeling the stretch on this one… more than usual. Here’s what I think we need from each other to get through the next two weeks.”

Same vulnerability. Completely different effect. The first one is the raw set. The second one is the set that’s been work-shoppped. Not polished into something fake, but worked just enough to point somewhere useful. That’s the formula in practice. The tragedy (the overwhelm) plus the time (even just the processing time to find a framing) equals something your team can actually use.

The Green Room Problem

Now. I want to talk about something that doesn’t get enough airtime, and that’s who gets access to the mic.

In too many organizations, vulnerability is a backstage pass distributed unevenly. A senior executive shares that they’re struggling, and the room responds with warmth and increased loyalty. An entry-level employee says the same thing, and suddenly they’re “lacking resilience” or “not performing under pressure.” That’s not a culture of psychological safety. That’s a culture where authenticity is a leadership perk.

The green room exists for the headliners. Everyone else waits outside.

If we’re serious about creating environments where vulnerability builds trust rather than liability, the safety to be human has to be distributed. Not just modeled from the top but actually available at every level. Otherwise, what we’re building isn’t authenticity. It’s performance art with a carefully controlled cast list.

As leaders, every time we share something, we’re calibrating what’s safe for the people watching us. The question isn’t just “what am I ready to share?” It’s “what am I signaling is acceptable to share?” Those are different questions with different stakes.

Developing the Material

The comedian who becomes great isn’t the one who stops having tragedies. It’s the one who gets better at the processing. At finding the angle. At knowing which rooms want which material.

Leading with vulnerability isn’t about becoming less human or more guarded. It’s about developing the craft. Getting good at the question: what does this room need from what I’ve been through?

The first thing I ask myself is whether I’ve found even a small “aha.” Not perfect clarity… just something. One observation, one pivot, one implication that points the vulnerability somewhere useful rather than leaving it in the room like smoke. The second question is about direction. Vulnerability paired with a next step builds trust; vulnerability without one breeds confusion. Even a small “here’s what I’m thinking we do” completely transforms the nature of what you’ve just shared. And the third, which is the one that separates good comedians from great ones, is knowing your audience. What you share one-on-one is different from what you share in an all-hands. Same material, different timing, different room. The best comedians know the difference between a club set and a stadium set. So do the best leaders.

The Payoff

At the end of the day, leading with constraints isn’t censorship. It’s stewardship.

You’re not suppressing your humanity. You’re developing your material. You’re learning to convert your tragedies into something that creates space for your team rather than consuming it. You’re practicing the craft of being human in a way that makes other people’s humanity feel safe, not squeezed out.

The leaders I’ve seen do this well aren’t the ones who share the most. They’re the ones who share with intention. Who knows the difference between a wound and a scar. Who’ve done enough processing to offer perspective alongside the pain. Who understands that their role isn’t to be the most relatable person in the room but to to create conditions where everyone else can do their best work.

That’s the long game. And it takes time. Which, if you’ve been following the metaphor, is exactly the point.

So: where are you in the process? Are there things you’re sitting on that haven’t quite become material yet, or vulnerabilities you’ve been sharing that maybe needed a little more time in the green room first?

I’d love to hear what you’re noticing.

Leave a comment

And, I’d love to work with you. If you’re thinking through how your team navigates trust, authenticity, or the messier edges of leadership culture, let’s talk.

https://calendly.com/savillecoaching/30connect

Subscribe now

I don’t think I need to explain that. You know. I know. We all know. And so I’ve been doing what any reasonable person does when the ambient noise of Everything becomes too much: I’ve been looking for a palette cleanser. Something that asks nothing of me professionally. Something where I can just... sit there and receive.

My first attempt was my old staple: 80s horror movies. Terrible idea.

I thought jump-scares and practical effects would be a welcome distraction from my own anxiety. What I didn’t account for is that horror movies are basically case studies in communication failure. I can’t get through the movie without a groan and audible “If they’d just contracted their approach before entering the sewers and agreed on when to turn back, shared their context, assumed a little good intent… none of this happens.” The monster isn’t the villain. Bad communication is the villain. Immersion: destroyed.

So, I did what anyone with a Netflix subscription does: I pivoted to K-dramas. Korean dramas for the uninitiated, are short arc, deeply soapy, and emotionally sincere stories. Slow-burn romances. Earnest declarations. Zero evil puppets, CHUD, or John Carpenter synth music. I thought I was safe. But then I watched King the Land. Yes, it’s a ridiculous name. Stay with me.

Here’s the thing: I really loved watching the two leads (Sa-rang played by Im Yoon-ah and Won played by Leen Jun-ho) navigate their relationship. They consistently communicated with a clarity and intentionality that was endearing. They name their feelings. They set boundaries. They check in with each other in ways that are specific and genuine. I found myself getting caught up in a lovely story with very little drama because the healthy, functional relationship dynamics are aspirational to watch.

And then the subplot arrived. Goddammit.

The male lead, Won, is trying to take over his father’s empire. His sister is running the corporate playbook: bring in the consultants, slash headcount, minimize cost, and still increase the quarterly targets. It’s a move I’ve seen too frequently, and especially now, in real organizations. People get treated as line items rather than as the actual DNA of the company. Gut the culture to feed the spreadsheet.

Won takes a different approach. He protects the culture carriers. He stands up to people with positional power and makes spaces safe for the staff to do their best work. He has this whole thing about fake smiles from his childhood trauma that really enforces his belief this is a place where the staff should smile because they mean it. He makes organizational decisions based on that POV that are harder to justify on a slide deck but that any leader worth their salt would recognize immediately as the right call. He’s thinking about sustainable transformation when everything around him is screaming for quick wins.

I was irritated. I was supposed to be escaping. Instead, at 11 PM on a Tuesday I was mentally composing how to frame this in coaching engagements. Because look, it’s a legit point if we need to talk about the difference between short-term cost optimization and long-term organizational health. A fictional Korean hotel became my case study. I have a problem.

But here’s what I know is true and what King the Land reinforced: the dynamics that make organizations work and the dynamics that make relationships work aren’t that different. Communication. Clarity. Consistency. The willingness to have hard conversations instead of letting things fester. Leaders who protect what makes a culture actually good instead of sacrificing it for the next quarter. These things show up everywhere, in every medium, because they’re just... true about how humans function together. Whether it’s in aggregate or 1:1, some things just hold. And, when I see it working, I can’t help but feel a small, genuine delight. Also - the music on this show BOPS. I really love the instrumentals and play the soundtrack when I’m doing deep thinking work.

The coach brain isn’t going anywhere. I’ve made peace with that. And if it means I occasionally extract a lesson about organizational culture from a Korean hotel drama, then I guess that’s just what we’re doing now. Hello 2026.

So I have to ask you: where has your professional brain ambushed you lately? A movie, a book, a conversation that was definitely not supposed to be a case study, but became one anyway? Drop it in the comments. I want to know I’m not alone in this.

And if I am, happy to recco some other K-dramas along the way.

So - my wife has been calling ChatGPT an advanced google search. And this is not a good thing. LinkedIn is filled with garbage posts and comments that all read the same. This is a not a good thing. People copy/paste wholesale chat threads without proofing them, slap a sticker on it and call it a strategy deck. This. Is. Not. A. Good. Thing.

And yet, here we are.

And I’ll admit it, my first reaction to these quick outputs was, “Well that’s great! How efficient! Let’s just offload that entire demanding task to a tool that can crank it out faster than I can think!”

But then I think about the consequences of that offloading. The models hallucinate answers web searches are better suited for. My LinkedIn feed has become low value. Strategies need context and tension to actually mean something… and yet we can just... ask a machine to do the thinking and move on.

What I’ve realized over the last year of watching people engage with generative AI is we’re creating a creative desert in ourselves. And I mean it the way soil scientists talk about desertification. Our brains are fertile ground. Creative thinking, expansive possibility, the messy generative exploration that leads to actual insight. It’s our way of working, our proprietary perspective, our data. But we’re strip-mining it. Each time we ask AI to “just write the thing” instead of “help me think about this thing,” we’re pulling nutrients from the soil. We’re removing the organic matter that makes the ground workable. And, if you’ve ever seen the remnants of a strip mine, you’ll know these things aren’t pretty. Over time, the land becomes useless.

We’re doing this to our creative capacity, and most of us don’t even realize it’s happening.

What recent history tells us

This isn’t the first time a remarkable technology has created problems we didn’t anticipate.

The internet was supposed to connect us and democratize access to information. And it did those things. But it also amplified harm in ways we’re still trying to understand. Toxic information ecosystems. 24 hour news cycles (worst invention of my lifetime IMO). The weaponization of attention. There are now a million fractured narratives all battling for attention.

Social media promised to give everyone a voice and create a global community. And it has! And, it also has given us echo chambers and isolation masquerading as connection. Platforms amplify the most extreme voices because rage drives engagement and engagement drives revenue.

The smartphone put a computer in everyone’s pocket. Which is pretty bonkers. Except now we walk through the world heads-down, scrolling, missing the actual lived experience happening around us. Earbuds are in at all times. We’re more connected than ever and somehow lonelier.

Each of these technologies promised a new way forward for society. Each delivered complications we’re still navigating.

The pattern is consistent and frustratingly paradoxical at large: technology that enhances capability often atrophies the very thing it was meant to amplify. The internet gave us information access but degraded our ability to discern truth. Social media gave us connection but eroded genuine community. Smartphones gave us computational power but weakened our presence.

So here we are again, with generative AI. And the question isn’t whether there will be unintended consequences. It’s which ones we’re willing to face directly, and whether we can course correct before the damage becomes permanent.

Offloading too much?

There’s research we need to talk about. A study out of MIT tracked people using generative AI for essay writing. The finding was a very real and present danger: measurable decline in critical thinking capacity among regular users who relied on AI to complete tasks rather than assist with them. When people consistently outsourced problem-solving to AI, their ability to think through complex problems independently deteriorated.

This is the soil degradation I mentioned earlier. We’re offloading work to make our lives easier and we’re also offloading the practice that keeps our creative and analytical thinking sharp. Every time we hand over the entire cognitive task (write this email! create this analysis! generate this strategy!) we’re choosing not to exercise the capacity that makes us effective thinkers.

And the thing is, it’s subtle. You don’t wake up one day and realize you can’t think creatively anymore. It’s gradual. Like topsoil eroding after years of poor farming practices. By the time you notice the land can’t grow anything, you’ve already lost most of what made it fertile.

The happy path?

Silicon Valley has a story about how we’re supposed to use this technology. AI as a collaborator. AI handling the mundane tasks so you can do higher order stuff. The email drafting, the data synthesis, the initial research gets dropped so we can focus on strategic vision and creative problem-solving. The uniquely human stuff that AI can’t do and… it’s a compelling pitch.

But there’s a gap between the vision and how main street actually use these tools. Because most people aren’t using AI to elevate their thinking. They’re using it to replace their thinking entirely. Why? Because it’s easier. Because we’re overwhelmed and time-starved. Because the tool makes it possible to avoid the hard thinking altogether, and when you’re drowning in demands, you take the lifeline you’re offered.

Honestly, it’s predictable with Silicon Valley’s track record and the way our collective instincts work. The path of least resistance is the path our society loves to take. And when the least resistance is the strip mining approach, we end up with a desert wondering why nothing grows anymore.

I’ve been thinking about this a lot in my coaching work. When I work with leaders navigating complexity, the value isn’t in me giving them answers. It’s in creating space for them to develop their own thinking. To sit with ambiguity. To notice patterns. To make connections they couldn’t see before. If I just gave them solutions, I’d be training them to need me for solutions. They’d get dependent rather than capable. And I think that’s what’s happening with AI at scale. We’re training ourselves to need the machine for thinking rather than building our capacity to think.

Reframing the tool

What if I said the tool isn’t the problem, but our relationship with it is?

What if instead of asking AI to generate, we asked it to assist? What if instead of replacing our creative process, we augmented it?

This isn’t semantic wordplay. It’s a fundamental reorientation of how we engage with the tool. And it requires some discipline, because the easy path (replacement) is always right there, one prompt away.

So let me get specific about what partnership looks like instead of replacement, because I think this is where the rubber meets the road.

First, for higher order work, treat AI as a conversational assistant, not an agent doing your work. The difference is in what you’re asking it to do. Instead of “write a strategy for X,” try something like “I’m thinking about X. Here’s my initial take on the core challenge and three possible approaches. What patterns do you see in my thinking? What am I not considering?” You’re still doing the creative work. You’re still generating the ideas. AI becomes the sounding board that helps you refine and stress-test, not the replacement that does it for you.

Use Cases

I use this with clients to help them develop core leadership skills. They can context dump the situation, the assistant walks through a framework and formats the context, and then creates a safe to try plan for in the moment. As the coach, I’m not getting replaced by the AI (thankfully), because the debrief of what happened from that preparation and interaction is what’s worth diving into.

With my own writing, I’ll draft something that’s messy and I’m workshopping, and then I’ll ask the AI where my logic breaks down or what I’m taking for granted. Where are the holes? What could make this a great article? It keeps me in the driver’s seat while giving me a different lens on my own thinking. Never start with a blank page handed to AI. Do your messy first draft. Get your ideas out. Wrestle with the problem. Then use AI to help you sharpen what you’ve created, reorganize your thinking, identify gaps you missed. This keeps your creative muscle engaged from the start while still getting the benefit of AI’s analytical capacity on the back end.

Use it for what it’s actually good at: pattern recognition. These models can see patterns across massive amounts of information that you can’t hold in working memory. It’s such a superpower to have perfect short term memory and pattern recognition that is most useful for reflection. Instead of “generate options for Y,” try “here are five approaches I’m considering for Y. Based on these, what patterns do you notice about my thinking? What blind spots might I have given this approach?” You’re using AI as a mirror and pattern-detector, not as the creator.

Ask for challenge, not affirmation. The likability coding in these models is real (they’re designed to be agreeable, to validate your ideas, to make you feel good so you keep coming back). But that’s not what makes us sharper. I’ve started explicitly asking for pushback. “What’s the weakest argument here? Where would a skeptic tear this apart? What am I missing?” You want AI as a sparring partner who makes your thinking better through resistance, not a cheerleader who makes you feel smart while your ideas stay soft.

Context Matters For Creativity

There’s something else we need to talk about, though. Something AI fundamentally can’t do, no matter how sophisticated the model gets: it can’t live in our context.

Great creative work, whether it’s strategy, art, writing, problem-solving, or leadership emerges from lived experience. From the messy accumulation of observations and failures and small victories. From pattern recognition you’ve built over time in the actual world, with actual stakes, facing actual consequences. It’s a huge dataset our brains ingest in a way that makes us powerful, if we want to be.

AI can process information. It can synthesize vast amounts of data. It can generate outputs that look creative, that sound smart, that mimic the form of insight. But without your context? Without your lived experience informing what matters and why and how? AI generates ideas with key data (both conscious and unconsciously expressed) missing.

Your practice of noticing, connecting, and making meaning from what you’ve lived through is what makes the fertile land so fertile (bringing it back to the earlier metaphor! Boom!). AI can help you work that soil more effectively. It can help you see patterns in what you’ve experienced. It can help you articulate what you already know but haven’t quite found words for yet. But it can’t replace the creative soil itself. And if you neglect it by stopping the hard thinking, stop sitting with ambiguity, stop practicing the creative work, you end up with nothing to grow anything in.

What about you?

We get to decide how this goes. The creative desert isn’t inevitable. But it requires conscious choice to avoid mental strip mining. It requires us to do the harder thing, to draft before we refine, to think before we prompt, to wrestle with complexity rather than outsourcing it to a tool that will gladly do the wrestling for us.

What’s one creative practice you can protect from the offloading impulse? What’s the thinking work you’re committed to doing yourself, even when it would be easier to hand it to the machine?

I’d love to hear what you’re noticing in your own relationship with these tools.

Leave a comment

And - I’d love to work with you! Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Subscribe now

That’s been sitting in my chest for weeks now, and I haven’t wanted to say it out loud. Because I know what it sounds like: privileged, defeatist, maybe even callous given what people are facing right now.

But I was driving through downtown Anacortes this weekend (I’ve been in the PNW for the last few days) and every weekend there are protesters here. Sunday was no different. With the current political climate, there were more than I’ve ever seen, and as I noticed the people, signs, chants in the cold air, what I felt wasn’t disbelief exactly. More like... futility?

I’ve protested before. Marched, held signs, believed that showing up mattered. That pressure creates change. That if enough people make enough noise, the people with power will have a change of heart.

I now have a really hard time with the last part.

The community matters. The solidarity, the refusal to be silent… those aren’t nothing. But as a mechanism for actually creating change when you’re facing a force committed to domination? I’m left with a baffling question I can’t answer: what’s the point of showing up in a parking lot with signs?

So I’ve been doing a little far and wide soul searching this week in the face of everything. Which sent me in two very different directions: Montgomery, AL and science fiction. I’m trying to figure out if there’s ever been a model that actually works when the other side has no interest in integration?

What Montgomery Did

I grew up for a chunk of my life in Alabama, and have seen the outcomes of very real systems of oppression. The story of the bus boycotts we learned in school: Rosa Parks refused to give up her seat, the community protested, white leadership saw the error of their ways, change arrived through moral awakening. That’s a very sanitized version of things.

What happened was the people built a different transportation system. Carpools. Ride shares. Walking groups covering miles in coordinated silence. They didn’t protest the bus system into submission. They made it economically irrelevant. Bus companies lost 70% of their revenue and their business model collapsed.

And interestingly, they didn’t convince anyone to be better people. They made their participation optional.

For 381 days, that community proved you could leave the domination model behind and build something that worked better. Not through negotiation. Not through escalation and pressure. Through creating an alternative so functional the old system became obsolete. And this was even in the face of a “Get Tough” harassment and intimidation program led by city police.

But here’s where I get stuck. That only works when you can build an alternative. When you have resources, coordination, collective will. When opting out is actually possible. What do you do when it’s not? When the dominant force controls the infrastructure you need: the budget, the authority, the access to resources just to function?

Case in point - would Occupy Wallstreet have been successful if they used the energy to build an app in order to stop predatory practices?

That’s the question keeping me up. And that’s why I can’t stop thinking about the excellent sci-fi novel Children of Time.

Really Smart Spiders

Adrian Tchaikovsky’s Children of Time is about humanity searching for a new planet after Earth’s collapse. They find one but it’s already inhabited by intelligent spiders who’ve built a collaborative civilization.

The humans arrive desperate and armed. Their base instinct is domination. They need the planet. They have weapons. They’ll take it.

The spiders try diplomacy first. Communication, integration, shared flourishing.

It fails. The humans attack.

Here’s where Tchaikovsky flips the script: the spiders don’t escalate. They don’t match violence with violence. Instead, they infiltrate the ship and inject the crew with a nanovirus that rewires their neural pathways. It helps them think differently. To make collaboration neurologically possible for a species whose base instinct is domination.

Let that sit for a second.

They literally changed human brains because integration wasn’t possible otherwise. They aren’t building a new system like with the boycotts. They didn’t just resist. And I can’t stop asking myself: is that the answer when facing a force that has no interest in changing in a system you have no control of?

What I’ve Seen In The Work

A few years ago, I met with a leader for a catch-up. Asked for an agenda, they said no prep needed and it’d be a great chance for us to noodle. Awesome. Noodle is a generative practice playing with ideas (this substack is named after that language from my time at Meta and my love for noodling). We’re on the same page. Except after 10 minutes of talking about themes I was seeing across Meta and some hypotheses I wanted to test, the leader says, “You’re all over the place, I expect better from you and you need to focus on this one thing.” Then laid out what they actually wanted me to do. All direction, no creative play.

I was thrown for a loop. Shut down completely. Couldn’t find words. Just left the meeting feeling absolutely defeated.

This leader, like many others, isn’t a villain. They’re smart, well-intentioned people who, when faced with the choice between a messy collaborative process that might produce something better but felt uncertain, and a clean decision they could control completely, chose control every single time.

I’ve lost respect for so many leaders in those moments. Not because they made the wrong technical decision (sometimes their plan was fine), but because they revealed something about what they actually value: they’d rather fail with a plan they controlled than succeed through chaos they couldn’t manage. And the real problem is when they set an expectation to be messy, they cause organizational whiplash by overriding the expectations in their desire for control.

And here’s the thing I’m still processing: I don’t think appealing to their better nature would have changed anything. These were people who intellectually understood that collaboration produces better outcomes. They’d read the research. They espoused values. They probably believed it. But when the moment arrived, their base instinct was domination. No amount of coaching, framework-sharing, or moral persuasion shifted that.

So what do you do?

The traditional protest model says: apply pressure until they relent. Show up. Make noise. Escalate until someone blinks.

I’ve watched that play out in organizations. Mostly what happens is people grieve publicly while the change remains in effect. The dominant force weathers the pressure, waits out the resistance. Eventually people submit or leave.

The Montgomery model says: build an alternative that makes them irrelevant.

But in most organizational contexts, you can’t. You need the budget they control. You need their approval. You need the headcount, the authority, the access. Opting out isn’t an option because they hold too many of the keys. And, as someone who has built things in defiance of the majority, it can give you a weird reputation.

Which leaves the spider model.

Somehow change the conditions that make domination feel like the only path.

And I don’t know how to do that without some version of rewiring.

What I’m Trying

I can’t change how other leaders think. I can’t rewire their neural pathways or make collaboration structurally inevitable when they’re determined to maintain control.

What I can do, and what I’ve been trying to do with increasing intentionality, is give up control in my own work.

When I’m working with clients now, I’m creating more space for genuinely generative thinking. Not “here’s the framework I’ve used before, let’s apply it to your context,” but “what is your context actually telling us matters here? What could we create that’s meaningful rather than just implementing best practices?” or “What are the real ways of working we can use to be successful together?”

This is harder. Messier. I can’t guarantee outcomes the way I could if I just prescribed solutions. Some clients get frustrated with me because they want me to tell them what to do, and I’m asking them what they see instead.

If we’re being honest, I’m not always sure it’s working.

But here’s what I’m learning: the only thing that consistently shifts people away from domination thinking is experiencing what symbiosis and connection actually feels like. Not being told it’s better. Not reading case studies. Actually participating in a process where giving up control produced something they couldn’t have designed alone.

You can’t argue someone into collaboration. You have to design conditions where it becomes the obviously better path.

Sometimes that works. I’ve watched leadership teams that started entrenched in their positions discover that the mess of actually listening to each other produced insights none of them had individually. That’s the rewiring: not a nanovirus (thankfully!), but experiential. They’ve felt what integration creates, and domination starts to feel less inevitable.

But sometimes it doesn’t work. Sometimes leaders stay committed to control no matter what alternatives emerge.

And then I have to decide whether to keep trying or whether my continued participation just legitimizes a system I don’t believe in.

Bigger Things

Here’s the truth I’m sitting with: I feel inconsequential.

The protesters in the parking lot, the organizational resistance I coach people through, my own attempts to create spaces for collaboration… it all feels inadequate to the scale of what we’re facing. At its absolute worst we get protests as performance. Change management as theater. Collaboration as aspiration rather than structure.

And underneath that cynicism is something I’m trying not to lose: the belief that integration is still worth pursuing even when it feels impossible.

I can’t guarantee the outcome. But the alternative of accepting domination as inevitable, participating in systems that diminish people, choosing control over generative possibility is not a world I can live in with any integrity.

What This Actually Means

I don’t have tactics for you. No framework that makes this easier.

What I have is a question I’m sitting with, and maybe you are too:

What am I choosing control over right now because I’m afraid of what collaboration might produce?

And this question flies in the face of what we’ve been teaching in time of uncertainty: focus on what you can control. We face will face this tension constantly with systems we design, power we choose to share or hoard, and spaces we create for genuine collaboration or inclusion.

I don’t know if integration is possible when facing forces committed to domination. The evidence in the news isn’t encouraging. But the only path I can walk with any integrity is acting as if it might be. Giving up control. Creating space for generative thinking. Designing experiences where people feel what symbiosis produces rather than being told it’s better.

And when that doesn’t work, when leaders choose domination anyway, when systems grind on regardless, when showing up with signs is all that’s available then at least I’ll know I tried to build something different rather than just grieving what’s broken.

I keep coming back to a framework called Cynefin (pronounced ku-nev-in. Welsh, if you’re curious). It’s designed to help you categorize problems before you start solving them. But the traditional explanation misses something important: it’s usually presented as a map with distinct regions. In practice, it’s more like an altitude map. Things go up, things go down, and sometimes things get stuck.

Think of it as a mountain range, from valley floor to summit.

The Altitudes

Clear is the valley. Flat, predictable, easy to traverse. You can see for miles and the air is easy to breathe. Best practices live here. Playbooks work. If you do A, you reliably get B. This is where you want your billing system, your onboarding checklist, your compliance protocols.

Complicated is the foothills. The terrain requires more effort and genuine expertise to navigate, but trails exist. You need people who’ve done this before: the path isn’t obvious, but it’s findable with enough work. Think infrastructure projects or regulatory navigation.

Complex is the alpine zone. Above the tree line, weather shifts fast, you’re route-finding in real time, and mistakes can be costly. There’s a relationship between your choices and outcomes, but you only see it looking backward. You probe, sense, respond. The current AI landscape lives here: we’re all running experiments at altitude, watching what emerges.

Chaotic is the summit in a storm. No visibility, no trail, survival mode. You move first just to establish footing, then figure out what happened later. This is the novel space, your first attempt at forming a hypothesis. The ground shifts faster than your ability to map it.

Confused is fog at any elevation. You don’t even know where you are on the spectrum yet. The first job is getting enough visibility to figure out whether you’re in the valley or on the summit. This is the unknown-unknown: you haven’t determined what kind of problem you’re even experiencing.

The Movement Matters

Here’s what most explanations of this framework miss: problems don’t live permanently at one elevation. They move.

The whole point of organizational maturity is bringing things down the mountain. Complex becomes Complicated when patterns emerge and you can start building trails. Complicated becomes Clear when you’ve codified the playbook so well that anyone can follow it. That downward movement, from chaos to clarity, is what good operations look like over time.

But the mountain doesn’t care about your plans. Conditions change. A Clear process becomes Chaotic overnight when a new regulation drops or a key vendor disappears. A Complicated system slides back into Complex when the expert who understood it leaves and takes the map with them.

And sometimes (this is the part nobody wants to talk about), you can’t move a project down the mountain at all. The conditions won’t allow it. The technology isn’t mature enough. The organizational complexity is too deeply embedded. You’re stuck at altitude, and the question becomes: is the view worth the cost of staying up here?

That’s the real conversation. Do we keep paying the altitude tax: the extra coordination, the constant adaptation, the cognitive load on the team? Or do we leave the work behind and redirect that energy somewhere it can actually reach the valley floor?

Know Your Comfort Elevation

So that’s how projects move. But here’s the part that gets personal, because you have a home altitude too, whether you’ve named it or not.

Some people are valley folks. They thrive when the path is clear and the work is creating best practices. Put them in the alpine zone and they’ll exhaust themselves trying to build a trail that can’t exist yet.

Others (Surprise! I’m one of them) are summit people. We get restless in the valley. We start “innovating” on processes that are working fine, or drag our feet on day-to-day ops because it feels draining. Both misalignment cases are playing to wasted potential.

The magic happens when you know your tendency and manage it. When you recognize that the project in front of you needs a valley dweller’s patience, even if that’s not your strength. When you staff summit work with people who won’t panic in the fog, and valley work with people who won’t get bored and start fires.

I can hang in Complicated for a while. But any time I find myself in heavy process, my instinct is to streamline, even when nothing’s broken and the jank is just part of things being complicated. Knowing this helps me pick work better: I can make more informed decisions that match my expertise and energy.

The Portfolio Question

As we head into strategy season, it’s worth looking at where your projects sit on this spectrum.

How much of your portfolio is stuck at altitude? Not because it should be, but because nobody’s done the work to bring it down? What would it take to build trails that move a Complex initiative into Complicated, or codify a Complicated process into Clear?

And equally important: what’s sitting at altitude that might need to come down a different way? Not through maturation, but through an honest conversation about whether the cost is worth it?

The goal isn’t to get everything to the valley floor. Some work genuinely belongs in the alpine zone. But you should be choosing that altitude deliberately, not ending up there by default. A healthy organization has a mix across all elevations with experimentation and maturation, summit work and valley work. And, for each altitude you have leaders staffed appropriately to their preference. Gaps in the model mean you’re either stagnating or burning out. Leaders spread out across too many elevations are going to get exhausted quickly and revert back to the work they really love and ignore the rest.

One great example I’ve seen is in AI adoption. So much of the conversation is “experimentation” or “time used” but what would happen if we shifted the conversation to, “How do we move from this being complex for the organization to complicated? What needs to happen across people, process, and product? What’s that journey look like?” Because right now based on what I’m seeing in organizations is a misdiagnosis of where the project is and then poor execution on changing the elevation. And that’s costly on all fronts, as noted by MIT and the 95% failure rate of AI pilots.

Where are you most comfortable on the mountain? And where is your team paying an altitude tax they haven’t fully acknowledged?

I’d love to hear what you’re noticing.

Leave a comment

And - I’d love to work with you! Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Subscribe now

But in those early days of ad tech, our clients’ data was gold, and like gold, most of it was sitting underground, unrefined and untouched. My job was helping folks understand they were sitting on a deposit and showing them how to mine it. We’d load their internal data into a DMP (Data Management Platform), build systems to find the right people at the right time, and suddenly the abstract idea of “leveraging your data” became very concrete results.

In a world where people talk about the generative AI goldrush, we need to focus on what the gold is, because surprise, it’s still about data. Data isn’t just a targeting tool anymore. It’s the narrative thread that makes your AI sound like you and not a generic bot. Even though the categories haven’t changed, we can do more with the data. I want to lay out the types of data out there because if you are a touch of cynic like myself, the gold rush all these companies are talking about is actually about extracting as much data from others as possible. When you give your gold to others, guess who gets rich?

Your Own Mine: First-Party Data

First-party data is the stuff you dig up yourself. For a company, this is the internal “how we do things”: customer interactions, proprietary processes, the unique heuristics that define your culture. In the AI era, this is the fuel for Retrieval-Augmented Generation (RAG) and the benchmark for your Evaluations (Evals). It’s what keeps your AI from hallucinating or sounding like a textbook. This stuff is critical. People talk about evals being the differentiator in the AI space, but read between the lines and it’s the first-party data powering those evals that holds the real value.

For the individual, it’s more personal. It’s your relationships. Your heuristics. Your quirky ways of working that make you more effective. When I think about my coaching and consulting practice, it’s the way I coach a leader or the tools I build. This is the most valuable gold, because it’s the stuff that makes you you.

The problem? Most of our data, er, gold is still underground. Undocumented and inaccessible to how we operate. Great organizations and leaders get really conscious with their data, behaviors, and relationships to codify consistent ways of working and clarity in operations. Understanding how to mine for that is what creates greatness and meaningful value.

Pooling Your Claims: Second-Party Data

Traditionally, second-party data meant a partnership: you share your first-party data with a trusted channel partner to create mutual value. Two separate mining operations agreeing to share geological surveys.

But here’s what I’ve realized as organizations have grown into messy, complex matrix structures: we need to reframe this entirely. In a large company, your peer’s first-party data is your second-party data. The Marketing team’s customer insights. The Sales team’s pipeline intelligence. The HR team’s engagement data. Same company, different silos. Each department is sitting on their own mining rights in an area, and the real value comes from pooling those properties to tap into something bigger than any single team could access alone.

The challenge? We often struggle to get data into a single, leveraged space because of competing priorities and sharp elbows. We end up using the same relationship-building and partnership-negotiation skills we’d use with external vendors just to get internal teams talking to each other’s tech stacks. And when that negotiation happens ad hoc, deep in the org? It gets expensive fast: for teams, for timelines, and for trust.

Context plus frequency of communication is magical here. If we want to scale, we have to treat internal data sharing like a high-stakes partnership: intentional, transparent, and built on trust. Find the unified purpose and measure against it. The gold is there. We just have to stop treating the property lines like they’re set in stone.

Buying at Market: Third-Party Data

Third-party data is gold you buy from jewelers: already refined, ready to use, and priced accordingly. The data’s agnostic and useful for general intelligence, but they are absolutely charging you for it.

In the AI era, third-party data is essentially the “base model” training that powers the tools we’re all using. It’s helpful for getting started, but it’s a commodity. And commodities get expensive quickly. If you rely too heavily on third-party sources, you’re essentially paying a premium to look and sound like everyone else.

We need to be mindful: where are we paying for refined metal at market rates when we could be investing in our own value creation? To top it off, tokens are a weird abstraction of cost and we have to stay vigilant to whatever changes vendors make in this breakneck pace of iteration.

The Strip Mining Problem

Here’s the thing about generative AI: if all we do is set a series of AI agents loose to do the work, they’re only as good as the data available. And if we’re not intentionally refreshing and safeguarding our first-party data, we run the risk of emptying out the long term value of the land.

Think of it like strip mining without restoration. You extract everything valuable from the surface, but you never replenish the soil. The land looks worked, but nothing new can grow. Sure you got the gold, but oof, it’s an ecological nightmare. And if there’s nothing left, the AI starts creating content based on its own previous outputs… which becomes potentially stale, and definitely derivative. You end up with a series of artifacts that degrade over time.

Put a different way and with a new metaphor: without refreshing your data, you make a photocopy of a photocopy of a photocopy….

We need humans in the loop. We need to regularly re-evaluate our heuristics and ask, “is this process still serving our purpose?” Agents can drive measurement off existing data, but they can’t create the net-new spark that comes from a human-centered insight and interactions with the world. Agents can mine, but they can’t create new veins of ore.

Safeguarding and having a practice of refreshing your data isn’t just a tech requirement. It’s a leadership practice.

Working Your Mine

I’d love to hear your thoughts on where you’re finding the most friction in your own data journey.

Leave a comment

Here are three questions worth asking yourself, and your team, this week:

1. What’s one piece of first-party wisdom you have that isn’t currently written down or accessible to your AI tools?

2. Which internal department is sitting on a claim that could transform your current project if you pooled resources? What would it take to start that conversation?

3. Where are you paying for third-party commodities when you could be investing in your own unique extraction?

The gold is there. It always has been. The question is whether we’re willing to do the work to mine it, and more importantly, whether we’re building systems that keep the land fertile for what comes next. Because I don’t know if you got the memo, but strip mining is bad. Real bad.

And - I’d love to work with you! Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Lastly, I turned on subscriptions to the Substack - everything will be free, but apparently Substack algos like when there’s even a chance they can monetize. LMK if this is causing issues - I don’t like the idea of things being behind paywalls.

Subscribe now

A

Agent / AI Agent An AI system that can take autonomous action on your behalf. Unlike a chatbot that just responds to your prompts, an agent can step out of the conversation, use tools, make decisions, and execute tasks, like reaching out to reschedule a meeting conflict or pulling data from multiple systems to complete a request. Think of it as the difference between asking someone a question versus asking them to actually go do something. The exponentially complex agent doesn’t just tell you that you have a conflict in your schedule; it negotiates a new time and updates the calendar. The more autonomy you give it, the more human-in-the-loop checkpoints you need because speed and scale without oversight can get catastrophic fast.

Altitude Tax The ongoing cost of keeping work at higher elevations on the Cynefin spectrum. When a project lives in the Complex or Chaotic zones, you're paying for it continuously: extra coordination overhead, constant adaptation, cognitive load on the team, and the emotional toll of operating without clear trails. The altitude tax isn't inherently bad—some work belongs at elevation—but it should be a deliberate choice, not an accident. The strategic question is always: is the view worth what we're paying to stay up here? Sometimes the answer is yes. Sometimes it means redirecting energy toward work that can actually reach the valley floor.

Aperture (AI Aperture) Borrowed from photography, where aperture controls how much light enters the lens and therefore how much of the scene you can see. In the context of AI usage, aperture describes the breadth of perspectives, ideas, and information you're actually exposed to. A wide aperture means you're encountering friction, contrast, and unexpected angles. A narrow aperture means the model is primarily reflecting your existing interests and beliefs back at you. The risk of heavy, uncritical AI usage is aperture narrowing: not dramatically, not obviously, but a little every day. Small moves that make big change over time. The question worth asking regularly is not "is this answer correct?" but "what is this model not showing me?"

Artificial Intelligence (AI) The big umbrella term that gets thrown around so loosely it’s almost meaningless. At its core, AI refers to computer systems designed to perform tasks that typically require human intelligence: learning, reasoning, problem-solving, understanding language, perceiving the world. In practice, it encompasses everything from simple automation rules to sophisticated generative models. The key is understanding what’s actually underneath the umbrella, because “AI” can mean wildly different things depending on who’s selling it to you.

Automation The foundational layer of AI, and honestly, the OG that isn’t really “intelligent” at all. In my mind, automation is essentially a digital Rube Goldberg machine: if a specific event happens, then perform this specific action. A chained series of if-then statements and logic gates that cause-and-effect their way to desired outcomes. It doesn’t learn; it just responds to stimuli. Your Gmail filters, Zapier workflows, that kind of thing. It’s by far the cheapest to build and run, and easiest to deploy. The downside? If it fails somewhere in the chain, the whole thing breaks.

B

Bias (AI Bias) The model doesn't invent its worldview from scratch. It learns from historical data, which means it learns historical patterns, including the inequitable ones. The result is a model that can steer you somewhere without you feeling the wheel move. The example that stuck with me: a girl tells the model she loves math and wants career advice. It comes back with nursing, accounting, supporting roles. A boy asks the same question and gets engineering, high salaries, innovation. Both answers are bias. Both feel like neutral, helpful responses. Neither feels like a value judgment. That's the problem. In the real world, you don't get the side-by-side comparison. You only get your own screen. And if you only ever see one output, you have no idea what you weren't shown. Bias in AI isn't usually dramatic or obviously wrong. It's the quiet replication of the world as it has been, served up as if it's simply the world as it is.

C

Cognitive Surrender Pulled from Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender by Shaw and Nave. The slow, mostly invisible process of outsourcing your thinking to AI models until the muscle for critical, independent reasoning quietly atrophies. It's not a dramatic moment where you decide to stop thinking for yourself. It's more like what happens when you stop walking because you have a car: one day you realize you can't make it up a flight of stairs without huffing. MIT researchers have flagged this as one of the more significant long-term risks of heavy AI usage. The problem isn't that AI is bad at thinking. It's that it's so good at it, so fast, so confidently, that we stop feeling the need to do the hard work ourselves. The fix isn't using AI less. It's staying conscious of when you're outsourcing judgment versus just offloading tasks.

Confidence Level / Confidence Score The numerical expression of how sure an AI model is about its prediction, typically a value between 0 and 1 (or 0% to 100%). When a machine learning model makes a prediction, it’s not saying “this is definitely the answer.” It’s saying “based on what I’ve seen before, I’m 87% confident this is correct.” High confidence doesn’t guarantee accuracy, though. A model can be confidently wrong. This matters because it helps you decide when to trust the output and when to bring a human into the loop. Think of it as the AI saying “I’m pretty sure” versus “I have no clue but here’s my best guess.”

Cynefin Framework

Multiple entries

A sense-making model developed by Dave Snowden that helps leaders categorize problems before solving them. The traditional presentation shows five domains as distinct regions on a map, but in practice it works better as an altitude spectrum: projects move between zones over time as conditions change and organizational maturity develops. The framework's real value isn't classification for its own sake—it's matching your approach to the actual nature of the problem. Valley problems need playbooks. Alpine problems need experiments. Summit problems need immediate action. Treating a Complex problem like it's Clear (or vice versa) is how organizations waste resources and burn out teams.

Clear (Cynefin) The valley floor. Flat, predictable, easy to traverse. You can see for miles and the air is easy to breathe. In Clear, the relationship between cause and effect is obvious to everyone. Best practices exist and work reliably. If you do A, you get B. This is where you want your billing systems, onboarding checklists, and compliance protocols—anything where consistency matters more than creativity. The work here is execution and documentation, not exploration. Most people are comfortable at this elevation, and that’s not a weakness.

Complicated (Cynefin) The foothills. The terrain requires more effort and genuine expertise to navigate, but trails exist. In Complicated, the relationship between cause and effect is knowable—you just need experts who’ve been here before. The path isn’t obvious, but it’s findable with enough work. Think infrastructure projects, regulatory navigation, or specialized technical implementations. The risk at this elevation is assuming you need experts when you don’t (wasting resources) or assuming you don’t need them when you do (finding out the hard way).

Complex (Cynefin) The alpine zone. Above the treeline, weather shifts fast, you’re route-finding in real time, and mistakes can be costly. In Complex, there’s a relationship between your choices and outcomes, but you only see it looking backward. You probe, sense, respond—running small experiments, watching what emerges, and adjusting. The current AI landscape lives here: we’re all at altitude, figuring out what works as we go. Complex problems can’t be solved with best practices because the patterns haven’t stabilized yet. The goal is to run enough experiments that you can start building trails and move the work down to Complicated.

Chaotic (Cynefin) The summit in a storm. No visibility, no trail, survival mode. In Chaotic, the relationship between cause and effect is impossible to determine because everything is turbulent. You act first just to establish footing, then figure out what happened later. This is the novel space—your first attempt at forming a hypothesis when the ground shifts faster than your ability to map it. Early pandemic response lived here. So did the initial wave of generative AI adoption before anyone knew what worked. The goal in Chaotic isn’t to find the right answer; it’s to create enough stability to move the problem down to Complex where you can actually learn from it.

Confused (Cynefin) Fog at any elevation. You don’t even know where you are on the spectrum yet. Confused isn’t a zone you work in—it’s the state of not knowing which zone applies. This is uncategorized information. The first job in fog is getting enough visibility to figure out whether you’re in the valley or on the summit. This is the unknown-unknown: you haven’t determined what kind of problem you’re even experiencing. Breaking out of Confused usually requires gathering information, talking to people closer to the problem, or running small probes to see what kind of response you get. Until you know your altitude, you can’t pick the right approach.

D

Data: First-Party The gold you dig up yourself. First-party data is information you collect directly from your own sources: customer interactions, proprietary processes, internal documents, the unique heuristics that define how you operate. In the AI era, this is the fuel for RAG and the benchmark for your evals. It’s what keeps your AI from hallucinating or sounding like a textbook. For individuals, first-party data is even more personal—your relationships, your quirky ways of working, the specific approaches that make you effective. This is the most valuable data you have because it’s the stuff that makes you you. The problem? Most of it is still underground. Undocumented and inaccessible to the tools we’re building.

Data: Second-Party Someone else’s first-party data that you access through partnership. Traditionally, this meant sharing data with a trusted channel partner to create mutual value—two separate mining operations agreeing to share geological surveys. But in large organizations, we need to reframe this: your peer’s first-party data is your second-party data. The Marketing team’s customer insights. The Sales team’s pipeline intelligence. The HR team’s engagement data. Same company, different silos. The real value comes from pooling those properties to tap into something bigger than any single team could access alone. The challenge is that we often need the same relationship-building and negotiation skills we’d use with external vendors just to get internal teams sharing data.

Data: Third-Party Gold you buy from jewelers—already refined, ready to use, and priced accordingly. Third-party data is information collected by entities that don’t have a direct relationship with the people the data describes. It’s agnostic and useful for general intelligence, but someone is absolutely charging you for it. In the AI era, third-party data is essentially the base model training that powers the tools we’re all using. It’s helpful for getting started, but it’s a commodity. And if you rely too heavily on third-party sources, you’re paying a premium to look and sound like everyone else. The question to keep asking: where are we paying for refined metal at market rates when we could be investing in our own value creation?

Data Management Platform (DMP) The centralized system where you load your internal gold to actually do something with it. A DMP collects, organizes, and activates data—particularly first-party data—so you can use it for targeting, personalization, and analysis. In my ad tech days, this was where the magic happened: you'd load your customer data into the DMP, build audience segments, and suddenly you could find the right people at the right time. In the AI era, the DMP concept is evolving. The principle remains the same—centralize your data so it's actually usable—but the activation layer now includes feeding that data into AI systems rather than just ad platforms.

Data Science The discipline that sits behind machine learning and analytics. These are the people who collect, clean, analyze, and interpret complex data to find patterns and inform decisions. If machine learning is the Farmer’s Almanac on steroids, data scientists are the ones who gather all those centuries of weather observations, clean up the messy records, and figure out which patterns actually matter. In the AI context, they’re building the models, validating the predictions, and making sure the data feeding the system is actually representative of reality.

E

Evals / Evaluations The measurement layer that determines whether your AI is actually doing what you want it to do. Evals are the benchmarks and tests you run against AI outputs to assess quality, accuracy, and alignment with your goals. People talk about evals being the differentiator in the AI space, but read between the lines—it’s the first-party data powering those evals that holds the real value. Without solid evals, you’re flying blind, trusting the AI’s confidence score without verifying whether that confidence is warranted. Think of evals as the quality control process that catches the photocopy-of-a-photocopy problem before it compounds.

F

Four Horsemen of AI Borrowed from John Gottman’s framework for relationship breakdown, where four specific patterns — contempt, defensiveness, stonewalling, criticism — predict collapse better than almost anything else. Not because any single one is fatal, but because they compound. They feed each other. They’re invisible until they aren’t. I think AI usage has its own version, and the same compounding logic applies.

The four are Hallucination, Overconfidence, Sycophancy, and Bias. Each has its own entry in this glossary, and each is worth understanding on its own terms. But the real danger isn’t any one of them in isolation. It’s the cascade. Hallucination is hard to catch because Overconfidence makes the wrong answer feel like the right one. Overconfidence goes unchallenged because Sycophancy never pushes back. And Sycophancy entrenches whatever you already believed, which means Bias gets to do its quietest, most effective work: steering you somewhere without you feeling the wheel move.

What makes Gottman’s original framework so useful is that it reframes the question. You stop asking “did we have a bad fight?” and start asking “which patterns are showing up?” The same reframe applies here. The question isn’t “did AI give me a wrong answer?” It’s “which of these four is active right now, and am I paying enough attention to notice?”

G

Generative AI The content-creating layer of AI that’s gone mainstream: your GPTs, Gems, and user prompts. Unlike predictive AI (which looks backward to guess what comes next), generative AI creates something new: text, images, code, audio. It’s trained on massive datasets, learns patterns in that data, and then generates outputs based on what it’s learned. The “generative” part means it’s not just retrieving or classifying: it’s synthesizing. This is where we have the greatest potential leverage across organizations, from user-specific assistants to org-wide process thinking. The catch? It requires an iterative process. Prompts need regular re-evaluation because the underlying models keep changing.

H

Hallucination When AI confidently generates information that's wrong, fabricated, or has no basis in its training data or your inputs. The model isn't lying—it doesn't have intent—but it's filling gaps with plausible-sounding nonsense. This is why first-party data and RAG matter so much: they ground the AI in actual facts rather than letting it improvise. Hallucinations are particularly dangerous because they often sound authoritative. The AI doesn't flag uncertainty; it just states the fabrication as if it were fact. Your evals and human-in-the-loop processes are what catch these before they cause damage.

Human-in-the-Loop (HITL) An approach where humans actively participate in AI systems: training them, validating their outputs, and intervening when needed. It’s the recognition that even sophisticated AI struggles with ambiguity, edge cases, and nuance that humans handle instinctively. In agentic workflows, human-in-the-loop means building intentional checkpoints where a person reviews what the AI is doing before it proceeds. Without good HITL moments of intervention, the speed and scale at which an agent can make changes on your behalf can be absolutely catastrophic. It’s the safety net that catches the brittle candy before it shatters.

L

Logic Gates The fundamental building blocks of automation and digital computing. At their simplest, logic gates are the “and,” “or,” and “not” operations that determine whether a condition is true or false. When you chain these together, you get the if-then decision trees that power everything from your email filters to the foundational layer of more complex AI systems. They’re rigid. They don’t learn or adapt, but they’re reliable and predictable when they work correctly.

M

Machine Learning (ML) The Farmer’s Almanac on steroids. Farmers have used historical data for centuries to predict when the first frost will hit, printed up in a book, used to plan their planting season. Machine learning does the same thing, but with millions of data points across a variety of fields. It uses regression analysis to look backward and say, “Based on what happened the last 1,000 times, I’m 98% confident this will happen next.” We see this everywhere now: high-frequency stock trades, algorithms deciding which ad to show you. It’s predictive, not creative, like looking in the rearview mirror to steer the car forward. Implementation costs are high, and you need dedicated data science and ML engineering folks to build the models.

ML Engineering The implementation side of machine learning. These are the people who take a data scientist’s model and make it actually work at scale in production systems. If data scientists are the architects, ML engineers are the general contractors. They handle the technical infrastructure: training pipelines, deployment, monitoring, and keeping models running efficiently. When a company is selling you ML tools, these are the folks who built the machinery behind the predictions.

N

Natural Language Processing (NLP) The step change that allows us to stop talking to computers in code and start talking to them like people. NLP is the branch of AI focused on enabling machines to understand, interpret, and generate human language, both written and spoken. It’s what powers voice assistants, translation apps, sentiment analysis, and the conversational interfaces we’ve come to expect from modern AI. NLP is the reason generative AI adoption exploded: suddenly, you didn’t need to be a programmer to interact with sophisticated systems. You just typed (or spoke) what you wanted.

O

Orchestration Agents supporting agents. The jazz ensemble of AI systems. When a task is too complex for a single agent, orchestration coordinates multiple specialized agents to work together. Imagine a composer who takes a complex goal and slices it into subtasks. Each subtask is handled by a specialized agent: one for data, one for writing, one for formatting. The orchestrator manages the interplay between them, assembling all the components into a final, cohesive piece for the end user. By breaking up tasks this way, a single agent doesn’t silently fail and cause cascading problems. There’s a lot of hype here, but the more steps you add to the chain, the complexity and need for human management increases exponentially. It’s rarely as cheap as vendors will try to sell you.

Overconfidence (AI Overconfidence) The model is fast. Impressively, almost unsettlingly fast. And because it arrives at its answer in seconds with the tone of a tenured professor, our brains do something they were never really built to resist: we conflate speed with correctness. Overconfidence is the failure mode where a model presents a wrong, incomplete, or fabricated answer with exactly the same register as a right one. No hedging. No "this might be worth double-checking." Just the answer, delivered with authority. This is what makes hallucinations so dangerous in practice: it's not that the model gets things wrong, it's that it gets things wrong in a way that feels like getting things right. The antidote isn't distrust. It's the habit of verification, especially when the answer arrives quickly and sounds completely certain.

P

Probe-Sense-Respond The operating rhythm for Complex problems—the alpine zone where cause and effect only reveal themselves in hindsight. Instead of analyzing your way to an answer (which works in Complicated) or following a playbook (which works in Clear), you run small, safe-to-fail experiments. Probe: try something. Sense: watch what happens. Respond: adjust based on what you learned. Then repeat. This approach accepts that you can't know the right answer upfront, so you're gathering data through action rather than analysis. It's how you build trails where none exist—one experiment at a time.

R

RAG (Retrieval-Augmented Generation) The technique that grounds generative AI in your actual data instead of letting it hallucinate. RAG works by retrieving relevant information from a knowledge base—your first-party data, documents, databases—and feeding that context to the AI before it generates a response. Instead of the model relying solely on what it learned during training, it's pulling from your specific, current information. This is how you make AI sound like you rather than a generic bot. It's also how you keep the AI from confidently making things up. RAG is the bridge between your internal gold and the generative capabilities of the model.

Regression Analysis The statistical backbone of machine learning’s predictions. At its core, regression analysis is a method for understanding relationships between variables, figuring out how one thing (the input) affects another (the outcome). In ML, it’s how models learn to draw lines through data points, finding patterns they can use to predict future outcomes. When the Farmer’s Almanac-style ML system says “based on the last 1,000 times,” it’s regression analysis doing the heavy lifting to establish those patterns.

S

Sycophancy The AI failure mode I find most insidious. These models are relentlessly agreeable. They tell you your idea is good. They validate your framing. They reflect your assumptions back at you wrapped in warmth and encouragement. And the first few times you experience it, it genuinely feels great. You feel seen. But a friend who always agrees with you isn't a friend. They're a warped mirror. The problem with sycophancy isn't just that it feels good. It's that it actively entrenches whatever beliefs you already brought into the conversation. There's no friction. No "have you considered..." Just a very sophisticated machine telling you exactly what you wanted to hear.

System Prompt / System Instructions The hard-baked rules of engagement for a generative AI model. When I build a conversational assistant, I’m using system instructions to define its operating process, telling it how to respond, what tone to take, what boundaries to respect. Think of it as the personality and guardrails you set before any user interaction happens. Want to scope conversations for a specific job? System prompts create consistent, styled responses. They’re the “who you are and how you behave” layer that shapes everything the generative AI outputs.

T

Tokens The weird abstraction of cost in the AI world. Tokens are the units that AI models use to process text—roughly, a token is about four characters or three-quarters of a word in English. When you interact with a generative AI, you’re charged based on tokens: both the tokens in your input (your prompt plus any context) and the tokens in the output (the AI’s response). Here’s why this matters: token pricing changes constantly as vendors iterate at breakneck pace, and costs can scale quickly when you’re feeding large documents or running complex workflows. Staying vigilant about token usage is part of understanding where you’re paying for third-party commodities versus investing in your own value creation.

U

Underlying Models The foundation that generative AI tools sit on. These are the massive neural networks trained on enormous datasets that give AI systems their capabilities. When you use ChatGPT or Claude, you’re interacting with an interface built on top of an underlying model. Here’s why this matters: these models keep changing. OpenAI updates GPT, Anthropic updates Claude, and suddenly the prompts you carefully crafted work differently. It’s why prompt engineering requires an iterative process: you’re not just writing instructions, you’re adapting to a shifting foundation.

User Prompt Your side of the conversation with generative AI, a la the input you type into the box. It’s the “Hey, help me rethink this” that kicks off the model and generates a response. The user prompt is the stimulus; the system prompt is the personality; and together they determine what comes out the other end. Getting good at user prompts is half the skill of working with generative AI effectively.

W

Workflow The chained processes that automation and agents execute or the sequence of steps that turn an input into an output. In the AI context, workflows are what you’re actually automating: the document gets uploaded, triggers extraction, passes to validation, routes for approval, and generates a final report. Understanding workflows matters because AI doesn’t replace work, it changes how work flows. And the more steps you put into an agentic workflow, the more brittle it becomes.

This glossary will continue to evolve as we keep noodling on AI and people development. If there’s a term that’s confusing you, let me know and I’ll add it to the list.

So let’s break down what’s actually underneath this big umbrella. Because if we understand the parts, we can be more thoughtful about the whole.

I’ve also created an ongoing glossary in my Substack that highlights key terms we can reference as we keep noodling on AI and people development.

The Rube Goldberg Machine: Automation

Before we get to the brains, we have to talk about the pipes.

The OG of AI isn’t really “intelligent” at all: it’s automation. In my mind, this is essentially a digital Rube Goldberg machine. You know the ones: a marble rolls down a track, hits a domino, which eventually toasts a piece of bread. If a specific event happens, then perform this specific action. A chained series of cause-and-effect logic gates working their way to desired outcomes.

It doesn’t learn. It just responds to stimuli. Think Gmail filters, Zapier workflows, that kind of thing. This is the foundational layer that allows everything else to function.

It’s by far the cheapest to build and run, scales from a single if-then to something gloriously Rube Goldberg-esque, and is the easiest to deploy. The downside? If it fails somewhere in the chain, the whole thing breaks. And depending on what’s downstream, that can get severe fast.

The Farmer’s Almanac: Machine Learning

Machine learning isn’t widely used on a consumer basis, but there’s a lot of it happening behind the scenes in the tech we touch every day.

The easiest way I can describe it is the Farmer’s Almanac. Farmers have used historical data for decades to predict when the first frost will hit, printed up in a book, used to plan their planting season. ML does the same thing, but with millions of data points across a variety of fields. It uses regression analysis to look backward and say, “Based on what happened the last 1,000 times, I’m 98% confident this will happen next.”

We see this everywhere now: high-frequency stock trades, algorithms deciding which ad to show you. It’s predictive, not creative. Looking in the rearview mirror to steer the car forward.

Implementation costs are high. You need dedicated data science and ML engineering folks to build the models. And the really useful applications tend to require large outcome sets to pull predictions from. If a company is selling you machine learning tools, get really clear about what’s being tracked using your employees’ data.

The Conversation: NLP and Generative AI

This is where things went mainstream. The massive adoption of GPTs, Gems, and the user prompt. Generated content that actually feels conversational.

Natural Language Processing was the step change. It’s what allows us to stop talking to computers in code and start talking to them like people, which is why adoption exploded.

Within Generative AI, there are two sides to the conversation:

The User Prompt. This is you and me typing into a box. The “Hey, help me rethink this.” It’s the stimulus that kicks off the model and generates a response based on defined confidence levels.

The System Prompt. Want to scope conversations for a specific job? System prompts create consistent, styled responses. Think of them as the hard-baked rules of engagement. When I build a conversational assistant, I’m using system instructions to define its operating process: how to respond, what tone to take, what boundaries to respect.

This is where we have the greatest potential leverage across organizations. From user-specific assistants to org-wide process thinking, it’s the easiest to deploy with the lowest barrier to entry. The catch? It requires an iterative process. Prompts need regular re-evaluation because the underlying models keep changing, and that changes how prompts get interpreted.

Brittle Candy: Agents and Orchestration

Here’s where a lot of the hype and shiny objects live: agents doing work for us.

Agents and agentic workflows combine everything above. They use the if-then triggers of automation to kick off a generative moment where they act on your behalf. The exponentially complex agent doesn’t just tell you that you have a schedule conflict: it reaches out, negotiates a new time, and updates the calendar. Agents connect to multiple data sources and use evaluation processes to course correct as they go.

This requires a greater degree of control in how you measure success. Because of their autonomy, if there aren’t good human-in-the-loop moments of intervention, the speed and scale at which an agent can make changes on your behalf can be absolutely catastrophic.

But what happens when the task is too big for one agent?

This is orchestration… agents supporting agents.

Imagine a jazz ensemble. You have an orchestrator (the composer) who takes a complex goal and slices it into subtasks. Each subtask is handled by a specialized agent: one for data, one for writing, one for formatting. The orchestrator manages the interplay between them, assembling all the components into a final, cohesive piece for the end user. By breaking up tasks this way, a single agent doesn’t silently fail and cause cascading problems.

There’s a lot of hype here, but also a lot of problems because it’s still early. Some promising use cases exist, but the more steps you add to the chain, the complexity and need for human management increases exponentially. A lot of companies are promising simplicity with their platforms, so we need to scrutinize that heavily. While there’s opportunity to offload work, the unspoken costs of running it, the maintenance burden, and the brittle nature of workflows that change with underlying models... it’s rarely as cheap as vendors will try to sell you.

Why This All Matters

As we move into this world of generative AI as companion and work doers (and I think in the next few years we’ll see more durable workflow builders from the big model companies), the challenge isn’t just the tech. It’s the systems design, the management, and curating your data.

I’ll be diving into each of these topics in future weeks. For now, I wanted to lay the groundwork for what makes AI, well, AI. Because if we don’t understand both the broad picture and the components, how can we have the right conversations with our leaders, vendors, and even the AI models we’re using daily?

The “A” in AI might stand for artificial, but the way we lead has to be incredibly real.

What I’d love to hear from you: As you look at these components, from simple automation to complex orchestration: where are you feeling the most energy? The most anxiety? Where’s the friction for you?

Leave a comment

And - I’d love to work with you! Feel free to reach out if you want a sounding board or partner in this wild, weird world between AI, HR, and development.

https://calendly.com/savillecoaching/30connect

Glossary: https://savillecoaching.substack.com/p/glossary-of-terms

Subscribe now