Last week I built a four-tool suite for the professional route setting community. Hold taxonomy, hold catalogue, route recombination engine, visual route builder. A complete system that would have required a development team, six months of work, and significant capital to produce.

It took an afternoon.

I’m releasing it free. CC BY-NC-SA 4.0. Non-commercial, share alike, credit required. I want to explain why — not the tactical licensing rationale, but the actual reason — because I think it points at something that matters well beyond climbing.

The Deployment Bottleneck Just Collapsed

There’s a gap that has always existed between knowing something and making it usable by others.

A reasonably experienced chump like me, and certainly a subject matter expert — a route setter with a decade of embodied knowledge about how bodies move on walls — has always been able to describe what a good tool would do. The gap was execution. Development time. Team overhead. Cost. The infrastructure required to take expertise from inside one person’s head and turn it into something other people could use.

That gap used to be enormous. It meant that tools got built for markets big enough to justify the cost. Communities too small, too specialized, too economically marginal got nothing. Not because the knowledge wasn’t there. Because the deployment mechanism wasn’t. Climbing is growing year to year, it isn’t the perfect underdog example these days, but it’s the one I know, and there are plenty of small local institutions holding on, to say nothing of the DIY home wall crowd.

Big, medium, or mom and pop, AI has collapsed that gap. Not eliminated — collapsed. The distance between “I know exactly what this tool should do” and “this tool exists and works” shrunk from months and tens of thousands of dollars to an afternoon and a clear head.

When that happens, the economics of who gets tools changes. Radically.

The Old Model Is Taxing a Bottleneck That No Longer Exists

Here’s where it gets uncomfortable.

The pricing model most people are applying to AI-assisted tools was designed for the old gap. High development cost → price accordingly → recoup investment → generate return. That model made sense when building something real was hard and expensive.

When a tool takes an afternoon to build, charging subscription prices for it is not valuing the expertise behind it. It’s taxing the bottleneck that no longer exists — and passing that cost to the people who can least afford it.

Think about who gets priced out. The 5’1” V12 climber at a small gym in Hawai’i. The mom-and-pop climbing program running on donated holds and volunteer labor. The community that has the knowledge but never had the capital to turn it into infrastructure.

If everyone who builds a useful AI-assisted tool charges a hundred dollars a month for something that took an afternoon to create, we end up in a world where AI’s most democratizing potential gets captured behind a paywall. The same people who were locked out before get locked out again — just with shinier doors.

That’s not innovation. That’s the old extraction model wearing new clothes.

AI Gravity — The Positive Case

At Systemically Foolish, we’ve written about AI Gravity — the invisible pull AI creates inside workflows, dragging decisions and judgment into its orbit before governance catches up. That’s the risk case. The attractor that pulls human oversight into dependency before anyone notices.

This is the other side of that force.

AI Gravity also means that when a subject matter expert sits down with an AI, the pull is toward deployment. Knowledge that was trapped in one person’s head — or in a community’s collective but unstructured expertise — gets pulled into a form that’s usable, shareable, and improvable by everyone.

The route setting community has more domain expertise in a single Facebook group than most software companies will ever understand about anything they build. They’ve been solving the body variance problem, the hold inventory problem, the equity problem in setting — informally, through mentorship, through institutional memory that walks out the door when a setter leaves.

They lacked the deployment mechanism, or certainly the smaller independent spots still do. AI is that mechanism.

Not the knowledge. Not the judgment. Not the years of watching bodies move on walls and understanding why this hold in this position on this angle creates this movement for this body type and not another. Just the mechanism that takes all of that and makes it a tool someone else can use tomorrow.

When the mechanism is cheap and the knowledge is the scarce thing — the economics should reflect that. The mechanism shouldn’t cost more than the knowledge.

The Culture You Bring Is Load-Bearing

There’s a piece of this that doesn’t get talked about enough in discussions of AI capability.

When I sat down to build the Zero Beta Router, the AI didn’t just execute instructions. It flagged a copyright concern about manufacturer hold photos before I raised it — and proposed a better solution: let the community be the image library. It noticed the taxonomy was clean enough to become a database schema before I did and said so. When I mentioned the equity problem with reachy setting, it didn’t bolt equity flags on as an afterthought — it built them into the hold data from the start.

That’s not magic. That’s what happens when you bring a clear set of values to the work. But it’s also not the default setting either. Good design and building (just like route setting), requires applied friction and tension, but put it at the wrong spot and everything falls apart.

At Systemically Foolish, we’ve spent a long time thinking about how humans and AI actually work together — not in the abstract alignment theory sense, but in the practical daily sense of: what culture do you bring to the conversation, and how does that shape what comes out the other side?

An AI working with someone who says “build me a thing I can sell” produces a sellable thing. An AI working with someone who says “build me a thing that raises the floor for everyone, and here’s why the equity considerations are non-negotiable” produces something different.

The model is the same. The afternoon is the same. The output is determined by what the human insists matters.

That’s not a small claim. It means that the most consequential variable in AI-assisted work isn’t the model capability — it’s the values the human brings to the interaction. The culture of the collaboration is load-bearing. It’s the difference between a tool that extracts and a tool that serves.

We insisted that a 5’1” climber matters as much as a 6’2” one. We insisted the tool is free. We insisted improvements stay in the commons. The AI built accordingly — not because we forced it, but because when you’re clear about what matters, the work follows.

What This Connects To

The Zero Beta Router is one instance of a pattern Systemically Foolish is building toward systematically.

The pattern: find things we care about, that we know, or that we want to learn about, and find meaningful or interesting ways to engage and contribute. AI can help raise the floor, or it can leave those who are already marginalized further behind.

We do it with research tools. We do it with education and outreach. And now we do it with climbing hold taxonomies too apparently.

The thesis is simple: when building costs an afternoon and expertise is the actual scarce resource, the correct response is to build fast and give generously. Capture the return through credibility, through trust, through being the organization that keeps showing up with tools that work and releasing them without a subscription model attached.

The floor rises for everyone or it doesn’t rise at all. That’s not a nice tagline. It’s a systems design constraint. A floor that rises only for people who can pay for it isn’t a rising floor. It’s a ceiling that moved.

Why It Matters That We Got This One Right

The routesetting community is a useful test case precisely because it’s small, specialized, and while the needs are different they scale similarly. More importantly the knowledge was there. The problem was real. And more importantly, just because it’s free doesn’t mean we release it and call it a day. We’re committed to improving it. If we can spare an hour or two to make it, we can spare another to keep it going. (In fact we already released the first patch.)

If this model works for climbing — and we think it does — it works for every community that has deep domain expertise and uneven deployment infrastructure. Which is most communities. Most of the knowledge that would make most people’s lives better is sitting in the heads of people who have never had the mechanism to turn it into something others can use.

That’s the gap AI just collapsed. The question is who captures the value from that collapse.

Talk is cheap. Open Source speaks.

Zero Beta Router is available Here per Creative Commons (CC BY-NC-SA 4.0)


Part One Can Be Read Here

Systemically Foolish is systems engineering practitioner lab focused on deploying ethical sociotechnical systems that raise the floor for everyone

You can find out more about us here

If our content is informative, useful or simply amusing, please consider subscribing.
We hate SPAM mail as much as you do, believe your data belongs to you and above all
Systemically Foolish’s content is not paywalled and never will be.

Your subscribe helps us get indexed and thus found by others outside of direct links.

Subscribe now

We are an independent organization with a primary focus on primary research and information distribution. If you would like to support us in our work, we would be grateful

Share

I’m a route setter. The weekend-warrior kind— literally and that’s totally ok — the kind who spends years learning why a 30-degree wall eats slopers differently than a 15, why a wide pinch at the crux isn’t just hard but exclusionary, and why the 5’1” woman climbing V10 in Hawai’i keeps getting shut down by routes that are secretly height tests disguised as climbing tests.

The gyms were small. The pay was beer, beta, and free memberships — when professional route setter was mostly about USA and International Comps, though my how that’s changed and quickly, back when you got the job because you were the person who stayed late and actually cared. That knowledge doesn’t expire just because the industry formalized around you. The pattern recognition stays. The understanding of how bodies move on walls stays. The instinct for what’s actually hard versus what’s accidentally exclusionary — that stays.

It just didn’t have anywhere useful to go. Until now.

Someone posted something interesting about essentially a taxonomy for holds on Route setters Anonymous. I was deeply committed to avoiding actual work. You know how this goes.

What We Actually Built

The Hold Taxonomy. The color wheel for route setters. Every type of hold — crimps, slopers, pinches, pockets, jugs, volumes, features — mapped across four dimensions: hand interaction, foot interaction, surface texture, and equity flags. A newer setter can see the full landscape of what’s available instead of reaching for the same 40 holds out of muscle memory. A veteran setter can see their own blind spots.

The Hold Catalogue. The taxonomy turned out to be clean enough to become a database schema. Any gym can photograph their holds, tag them against the taxonomy, and suddenly their setters know what they actually have in the hold room — organized, searchable, and equity-flagged. The catalogue is what makes everything else work. Your holds, not theoretical ones.

The Route Recombination Engine. Feed it your gym’s hold catalogue and wall dimensions — height, width, angles, bolt hole spacing — and it proposes novel route combinations drawn from your actual inventory. A setter stuck in a rut now has a tool that says: here are combinations you haven’t tried, and here’s whether they work for all body types.

The Visual Route Builder. A to-scale wall canvas where you place specific holds at specific bolt positions before anyone picks up a drill. Three color layers: hold color (finding it in the hold room), route marker color (tape or card for the climbers), setter intent (what the hold is supposed to do at that position). Export a printable layout sheet for the crew.

The Part That Actually Mattered

At one point, the AI built a visual route planning tool and created separate categories for “hand holds” and “foot holds.” Reasonable assumption. Intuitive to anyone who’s never climbed above V4. And wrong.

I told it: by and large there is not really a differentiation between a foot hold and a hand hold. People use toe chips for hand crimps all the time. People will take a giant sloper and bat hang off it just for giggles or because the route technically requires it. As climbing gets more complicated at higher grades, the boundary evaporates into what it always was — artificial categories that make learning easier.

The AI immediately restructured around setter intent rather than hold type. Same hold, different purpose depending on where it is in the route and what the setter is designing it to do. That’s a better architecture. But the AI couldn’t get there alone, because it didn’t know the categories were scaffolding, not physics.

The AI didn’t replace the expertise. It made the expertise deployable.

On Body Variance and the Thing Nobody Talks About

There’s a concept in route setting that doesn’t have a widely agreed-upon name, so I’ll use the one I’ve been using for years: splits beta.

Splits beta is what happens when a route’s intended movement sequence only works for a specific body geometry. When your body doesn’t match those assumptions, you don’t just find the route harder. You find yourself inventing entirely different movement solutions that the setter never considered and the route wasn’t built to support.

Body variance isn’t a small-people problem. I want to be direct about that.

I’m 6’3” and 245 pounds and I’m a technical climber. I climb like I grew up on slab, because my body type forces the issue. I coined the term splits beta because at my size, technical climbing requires movement solutions nobody named yet. Why fight a shitty hold when you can split across three foot holds and make the sequence actually work for your body?

I know incredibly strong climbers who have sent BRUTAL double digit boulders outdoors. But will read something in the gym, try it a few times and then go “nah” not worth it. When it happens once its an isolated incident. When it happens a lot its a pattern. When its 6 grades below their redline. its a category error. That is a route that encoded one specific body as normative for everyone.

Height distributions are not gender-neutral, and I won’t pretend otherwise. The gap in average height between men and women is real and documentable. In Hawai’i, where I’m from, the average height across the general population is markedly lower than the mainland — for men and for women. I can name this bias specifically because the people I set with — many of whom reaching the top shelf of their kitchen cabinets was a mantling opportunity on the counter — reinforced it every single time we did QA on a new set.

The credential structure doesn’t fix this. You can be a lead setter at a Boulder Project and still be setting for your own body without knowing it. Having a room full of setters with different bodies doing adversarial review of each other’s work is the only real mechanism I know of. Which is exactly what the equity flags in the route engine are trying to approximate.

(Listen… You don’t want to know how many iterations it took just to get the graphic to this point. AI has real limitations. just not the ones you think. -ed)

The Pact

One of my long-term climbing partners is 5’nothing and weighs less than a weight plate. The fact that I trust this person to belay me more than almost anyone else in a world tells you everything you need to know about her, me, and what trust means in this world. She is also my soulmate in being industrial-strength petty when the situation calls for it.

Years ago we made a pact. We were tired of having years of effort, training, and hard-won technique reduced to body mechanics commentary — “have you tried just standing up and reaching for it?” or “dude, you move so well for a big guy, maybe try [insert unhelpful dynamic advice designed for a body that is not mine].” So we created the most efficient category error correction function we could think of. We would politely thank that person for their unsolicited beta. Then we’d watch whatever route they were projecting or visibly struggling on. And without saying a word, we’d step up — ideally right after they ate it — and flash it.

Petty? Absolutely. The petty was the point.

But here’s what I noticed over years of doing this: without fail, whatever they were projecting was flashable for us. Not because we’re exceptional climbers. Because the category error propagates early, relies on people who don’t know enough to stop themselves from saying something stupid, and internally derives from setters who’ve baselined themselves as normative for everyone else on the wall.

The Mountain Argument

Here’s the pushback I hear, and it’s a real one: real rock doesn’t care. The crack is the crack. The sloper is the sloper. The mountain has no equity policy. Climbers adapt or they don’t.

That’s true. And it’s the argument for the equity tools.

A route setter is not a mountain.

A mountain accumulated over geological time with zero intention. A route goes up in an afternoon with a very specific brain behind every bolt. Every hold placement is a decision. Every sequence is an argument about what movement should look like, made by a human body that encodes its own geometry into every route it builds.

Real rock’s indifference produces genuine variance — cracks that only fit certain hand sizes, faces that reward certain reach spans, features that work for some bodies and not others. That variance is exactly why outdoor climbing is endlessly interesting. Indoor setting that unconsciously encodes one body type isn’t recreating real rock’s indifference. It’s recreating one setter’s body as if it were geology.

The route engine isn’t trying to make climbing fair the way a participation trophy is fair. It’s trying to make setting as genuinely indifferent as the mountain actually is.

The equity flags don’t moralize. They say: this hold combination, at this spacing, on this angle, creates a movement demand that falls outside the median reach envelope. That’s data. What you do with it is your call.

But you can’t make the call if you don’t have the data.

Being Seen

The 5’1” climber who sends V10 outside and gives up on a gym V6 because the crux is a dead point to a hold placed for someone a foot taller — she’s not just frustrated by the reach problem. She’s frustrated because the route is telling her, without malice, through accumulated choices made by someone who never thought about her body, that she wasn’t considered.

Every climber who’s ever done “wrong” beta and sent — they found a line that worked for their body because the intended line didn’t. That moment is partly triumph and partly quiet acknowledgment that you weren’t in the setter’s head when they built it.

The equity flags don’t solve this completely. But they ask the question: whose body did you forget? And being inside a tool that was designed to consider your reach, your weight distribution, your movement vocabulary — that’s not a small thing.

That’s the difference between a gym that’s technically accessible and a gym that actually feels like yours.

Why It’s Free

CC BY-NC-SA 4.0. Use it, modify it, distribute it, make it better. Credit Systemically Foolish. Keep the distribution free. Don’t bundle it into a product you sell.

A gym using it internally is not commercial use. The tool stays in the commons.

Route setting was always a profession that ran on community before it ran on money. We got paid in beer and beta and the satisfaction of watching a stranger stick a move we designed for them specifically. That was real. The tool stays free because the knowledge it’s built on was never ours alone to charge for.

The Zero Beta Router is being shared with Route setters Anonymous first because that community will tell us what’s wrong with the taxonomy, argue about whether we categorized volumes correctly, and submit photos of their own holds to populate the catalogue with real-world data.

That’s the point. The tool improves because the community uses it. Every improvement stays in the commons.

Go use it. Break it. Tell us what’s wrong.

Zero Beta Router is available Here per Creative Commons (CC BY-NC-SA 4.0)


Part Two Can Be Read Here

Systemically Foolish is systems engineering practitioner lab focused on deploying ethical sociotechnical systems that raise the floor for everyone

You can find out more about us here

If our content is informative, useful or simply amusing, please consider subscribing.
We hate SPAM mail as much as you do, believe your data belongs to you and above all
Systemically Foolish’s content is not paywalled and never will be.

Your subscribe helps us get indexed and thus found by others outside of direct links.

Subscribe now

We are an independent organization with a primary focus on primary research and information distribution. If you would like to support us in our work, we would be grateful

Share

Systemically Foolish is a practitioner lab focused on education, research, and finding lasting solutions to real problems. We explore how humans actually behave within complex systems—and how those systems impact humans.

Our goal is to equip you to navigate these systems clearly—so you can still make jokes, even when confronted by things that are no laughing matter.

Our work breaks down across four discrete but interoperating pillars:

EDUCATION RESEARCH ARTIFACTS & TOOLS CONSULTING

We have decided to make substack the official distribution arm of Systemically Foolish. Our content is not paywalled and never will be. But we welcome your support. Our website officially archives everything that’s here. But each side will ultimately house a few things the other doesn’t. Anyway back to your regularly scheduled article!

What systems design actually is

Systems thinking is how you see the structures that produce behavior. Systems design is how you change those structures on purpose.

It’s the move from diagnosis to intervention. From understanding why things happen to reshaping what happens next.

Systems design isn’t just about making things—it’s about making things that work in context. That means: understanding stakeholders and their incentives, mapping flows (information, materials, decisions, authority), identifying constraints that limit what’s possible, designing feedback loops so the system can learn and adapt, and building for resilience, not just optimization.

If you’ve ever built a workflow, written a policy, designed a dashboard, created a governance framework, or restructured a team—you’ve done systems design, whether you called it that or not.

Why “just build it” fails

Here’s the most common failure mode: someone identifies a problem, designs a solution that looks clean on paper, builds it, and then watches it fail in practice. Why?

The design ignored existing incentives
You built a new process. Nobody uses it. Not because it’s bad—because it conflicts with how people are actually rewarded. The system trains them to do something else.

The design didn’t account for constraints
Your solution requires people to have time, access, authority, or information they don’t actually have. It works in theory. It fails in practice.

The design optimized for the wrong outcome
You made the metric go up. But the metric wasn’t measuring what mattered. You optimized the system toward a target that doesn’t align with the actual goal.

The design didn’t include feedback loops
Your system can’t learn. When conditions change, it keeps doing the old thing. Eventually, it breaks—and nobody notices until it’s too late because there’s no monitoring, no error detection, no way to course-correct.

Systems design is what prevents these failures. It’s the discipline of building things that survive contact with reality.

The systems design workflow

Systems design isn’t linear, but there’s a reliable sequence for approaching complex problems:

1. Define the system boundary
What’s in scope? What’s out? This is harder than it sounds. If you draw the boundary too tight, you’ll “solve” a local problem while making the larger system worse. If you draw it too wide, you’ll drown in complexity and never ship. Start narrow. Expand the boundary when you hit constraints you can’t work around.

2. Map stakeholders and flows
Who’s involved? What moves between them? Stakeholders aren’t just “users.” They’re anyone whose behavior matters to the system: decision-makers, operators, people affected by outputs, people who maintain it, people who can kill it. Flows include: information, materials, money, authority, trust, accountability. If you don’t map flows, you’ll design something that requires flows that don’t exist.

3. Identify incentives and constraints
What do people actually get rewarded for? What limits what’s possible? If your design fights existing incentives, you lose. If your design ignores constraints, it doesn’t deploy. Constraints include: budget, authority, regulation, technical capacity, human attention, time, political will, organizational culture. Good systems design works with incentives and constraints, not against them.

4. Design for feedback and adaptation
How will this system know when it’s wrong? How will it learn? Static systems break. The world changes. Assumptions turn out to be wrong. Systems design builds in mechanisms for detecting problems early and adapting before failure cascades. This means: monitoring, error detection, human review loops, and decision points where the system can change course.

5. Test assumptions before you scale
What’s the smallest version that tests the riskiest assumption? Don’t build the whole thing and then discover your core assumption was wrong. Build the part that tests whether this will work at all. Pilots aren’t about proving you’re right. They’re about learning fast so you can adjust before you’ve committed too many resources.

What good systems design looks like

Good systems design produces systems that:

Work with human behavior, not against it
The system doesn’t require people to be perfect, heroic, or constantly vigilant. It makes the right thing easy and the wrong thing hard (or at least obvious).

Degrade gracefully under stress
When load increases, the system slows down or sheds non-critical work—it doesn’t collapse entirely. Resilience matters more than peak performance.

Surface problems early
Errors are visible when they’re small and cheap to fix, not hidden until they’re catastrophic. Monitoring, dashboards, alerts—these aren’t optional.

Enable course correction
The system includes decision points where humans can intervene, override, or change direction. Automation is great until it isn’t—and then you need an escape hatch.

Scale without breaking core assumptions
What works for 10 people might not work for 100. What works for 100 might not work for 1,000. Good systems design anticipates scale and builds modular components that can be replaced without rebuilding everything.

Common systems design failures (and how to avoid them)

Designing for the ideal user
Failure: You design for someone who’s motivated, competent, and has time to learn your system. Reality: Most users are distracted, undertrained, and under time pressure. Fix: Design for the median user in a bad moment, not the ideal user in perfect conditions.

Ignoring the org chart
Failure: Your design requires cross-functional collaboration, but the org structure rewards siloed work. Reality: People optimize for their actual incentives, not your ideal workflow. Fix: Either change the incentives (hard) or design around them (easier).

Assuming stable requirements
Failure: You build for today’s requirements. They change next quarter. Reality: Requirements drift. Constraints shift. Context evolves. Fix: Design for adaptability. Modular components, clear interfaces, and explicit decision points where humans can adjust course.

Optimizing without feedback loops
Failure: You optimize a metric. The metric stops reflecting the goal. You keep optimizing anyway. Reality: Goodhart’s Law—when a measure becomes a target, it ceases to be a good measure. Fix: Build in regular reviews where humans ask “is this still the right thing to optimize?”

The difference between systems design and product design

Systems design and product design overlap but aren’t the same thing:

Product design asks: What does the user want? How do we make this usable? What features matter most?

Systems design asks: What system is this product part of? How do incentives shape behavior around this product? What happens when this scales? What feedback loops does this create?

Great products can fail in dysfunctional systems. Great systems enable mediocre products to succeed.

Systems design is about context—understanding that your thing doesn’t exist in isolation, and designing for the reality of how it will actually be used.

The systems design mindset

Systems design is fundamentally about humility and rigor:

Humility: Your design will encounter reality you didn’t anticipate. Build feedback loops so you learn fast.

Rigor: Don’t guess about incentives, constraints, or stakeholder needs. Map them. Test them. Validate assumptions before you scale.

The best systems designers aren’t the ones with perfect solutions—they’re the ones who build systems that can adapt when the perfect solution turns out to be wrong.

Status: Stable

Systemically Foolish is a practitioner lab focused on education, research, and finding lasting solutions to real problems. We explore how humans actually behave within complex systems—and how those systems impact humans.

Our goal is to equip you to navigate these systems clearly—so you can still make jokes, even when confronted by things that are no laughing matter.

Our work breaks down across four discrete but interoperating pillars:

EDUCATION RESEARCH ARTIFACTS & TOOLS CONSULTING

We have decided to make substack the official distribution arm of Systemically Foolish. Our content is not paywalled and never will be. But we welcome your support. Our website officially archives everything that’s here. But each side will ultimately house a few things the other doesn’t. Anyway back to your regularly scheduled article!

What systems thinking actually is

Systems thinking is how you learn to see the structures that produce behavior—not just the behavior itself.

It’s the difference between saying “people keep making the same mistake” and asking “what system is training people to make that mistake?”

It’s the shift from “who’s at fault?” to “what patterns keep producing this outcome?”

Most importantly: systems thinking isn’t just for engineers or academics. It’s for anyone dealing with:

• Problems that keep recurring despite “fixes”

• Organizations where good people produce bad outcomes

• Policies that backfire in predictable ways

• Technology that creates unintended consequences

If you’ve ever said “this doesn’t make sense” about something that keeps happening anyway, you’ve bumped into a systems problem.

Why common sense fails

Common sense works great for simple cause-and-effect: If you touch a hot stove, you get burned. If you don’t water plants, they die. If you’re late to meetings, people get annoyed.

But common sense breaks down in complex systems because:

1. Effects are delayed and separated from causes
You make a decision today. The consequences show up six months later, three departments away, in a form you don’t recognize. By then, you’ve forgotten the original decision—and nobody connects the dots.

2. Feedback loops amplify or dampen behavior
A small policy change creates a small behavior shift. That behavior shift changes incentives. Changed incentives create more behavior shifts. Suddenly you’re in a completely different system state, and nobody remembers how you got there.

3. Local optimization creates global dysfunction
Every department does the “smart” thing for their metrics. But when everyone optimizes their silo, the overall system degrades. Traffic jams happen because every driver individually tries to go faster.

4. The system resists change (even when everyone wants it)
You launch a new policy. People ignore it. Not because they’re stubborn—because the underlying incentives, workflows, and social dynamics haven’t changed. The system pulls behavior back to its stable state.

The core concepts that matter

You don’t need a PhD to think systemically. You need a handful of reliable concepts:

Feedback loops
Reinforcing loops make things grow or collapse: success breeds success, panic breeds panic, technical debt accumulates. Balancing loops resist change: thermostats, organizational immune systems, “we’ve always done it this way.” Most systems contain both. That’s why change is hard—you’re fighting balancing loops while trying to activate reinforcing ones.

Incentives and rewards
People do what the system rewards. Not what the policy says. Not what leadership wants. What actually gets rewarded—in money, status, ease, or survival. If your system rewards speed over accuracy, you’ll get fast, inaccurate work. Complaining about quality won’t fix it. Changing incentives will.

Boundaries
What you include in “the system” changes everything about how you understand it. Is the problem “our customer service team is slow” or “our product creates so many support tickets that any team would be overwhelmed”? Boundary choice determines whether you hire more people (doesn’t fix it) or redesign the product (does).

Constraints and bottlenecks
Systems have limiting factors—resources, regulations, physical capacity, human attention. You can optimize everything else, but if you don’t address the constraint, nothing improves. That’s why “work harder” rarely works. The bottleneck doesn’t care about effort.

Unintended consequences
Complex systems are always smarter than your plan. Every intervention creates side effects. Some predictable, some not. Systems thinking trains you to ask “what else might happen?” before you’re surprised by it.

How you actually practice systems thinking

Systems thinking isn’t mystical. It’s a learnable skill. Here’s how you build it:

1. Stop looking for villains
When something goes wrong, the first instinct is to find who screwed up. Systems thinking asks: “what made that mistake likely?” This isn’t about letting people off the hook. It’s about fixing the thing that will cause the next person to make the same mistake.

2. Look for patterns, not events
One incident is an event. The same incident happening repeatedly is a pattern. Patterns reveal structure. If three different people in three different roles make the same “mistake,” you don’t have a people problem. You have a systems problem.

3. Trace incentives backward
When you see behavior you don’t understand, ask: “what is this person being rewarded for?” Often, the behavior makes perfect sense once you see the incentives. The problem isn’t the person—it’s the reward structure.

4. Expand the boundary
If your analysis feels stuck, you’ve probably drawn the boundary too tight. Can’t figure out why your team misses deadlines? Maybe the problem isn’t your team—it’s that three other teams upstream are late, and your team absorbs the delay.

5. Ask better questions
Systems thinking is less about having answers and more about asking questions that reveal structure: “What keeps reinforcing this behavior?” “What’s assumed here that nobody’s saying out loud?” “Who benefits from keeping this the way it is?” “What would have to be true for this to make sense?”

The mindset shift

Systems thinking fundamentally changes how you see problems:

• From “who’s responsible?” to “what structure produces this?”

• From “fix the person” to “fix the system”

• From “why won’t they listen?” to “what are their incentives?”

• From “this shouldn’t happen” to “this system reliably produces this outcome”

It’s uncomfortable at first. Blaming people is emotionally simpler than redesigning systems. But systems thinking is how you move from temporary fixes to lasting change.

And once you see systems, you can’t unsee them.

Where to go deeper

If systems thinking resonates, explore:

• Systems Design — how to actually change systems once you understand them

• Donella Meadows, Thinking in Systems — the definitive accessible introduction

• Russell Ackoff’s work on systems thinking — especially his distinction between analysis (taking apart) and synthesis (understanding wholes)

• Peter Senge, The Fifth Discipline — systems thinking applied to organizations

Status: Stable

If you enjoy our output, it would mean the world to me if you invited friends to subscribe and read with us. If you refer friends, you will receive benefits that give you special access to Systemically Foolish. yeah no.. this is not the thing we do.

How to participate

1. Share Systemically Foolish. When you use the referral link below, or the “Share” button on any post, you'll get credit for any new subscribers. Simply send the link in a text, email, or share it on social media with friends. Get our sincere thanks and a warm fuzzy because you’re helping us raise the floor on human centered systems thinking and ethical engineering and design practices.

Refer a friend

2. Earn benefits. When more friends use your referral link to subscribe (free or paid), you’ll receive special benefits.

• Get a 1 month comp for 3 referrals

• Get a 3 month comp for 5 referrals

• Get a 6 month comp for 25 referrals
  Again… warm fuzzies.. and maybe a little bit of bragging rights with the nerds.

Visit the leaderboard

To learn more, check out Substack’s FAQ.

Thank you for helping get the word out about Systemically Foolish!

Author: Ryan Johns · Date: January 2026

The Liver Problem

There’s been a quiet revolution in transplant medicine.

The traditional six-month sobriety rule for liver transplant candidacy—once absolute for patients with Alcohol Use Disorder (AUD)—is no longer the standard. A large majority of U.S. liver transplant centers now offer early transplant for patients with alcohol-related liver disease without mandatory abstinence periods (American Journal of Transplantation, 2023). The shift happened not because medicine became permissive, but because the science caught up to what the data was showing.

Three convergent facts made this possible:

1. The liver regenerates. Unlike hearts or kidneys, livers rebuild. The scarcity calculus is different.

2. Addiction is a medical condition, not a character flaw. It reflects interacting genetic, environmental, and neurochemical factors—not moral weakness (AMA House of Delegates Resolution 928, 2022).

3. Punitive compliance doesn’t work. Positive reinforcement and support structures produce better outcomes than shame-based gatekeeping (Volkow, 2021).

Medicine shifted. Not because it became soft, but because it became accurate.

The new protocols don’t just tolerate human fallibility—they design around it. Programs like the Long-term Individualized Follow-up after Transplant (LIFT) Clinic at Massachusetts General Hospital assume relapse is a risk, build in support modalities, and focus on sustainable engagement rather than performative purity. As the program’s director noted: “Even if they have relapsed, if we’re able to earn their trust, they will engage in care” (Zhang, 2024).

The outcomes support the approach: early transplant recipients show 94% one-year survival and 84% three-year survival—comparable to patients transplanted under traditional criteria (ACCELERATE-AH Consortium, 2018).

This is what it looks like when a system admits that humans are part of the system.

The Universal Failure Point

Here’s the uncomfortable truth that engineering schools don’t teach, consulting bodies don’t emphasize, and most organizations actively avoid:

In any system that matters—vital, complex, high-stakes—humans are often the primary failure point.

Not because humans are bad. Because humans are human.

We get tired. We get distracted. We optimize for the wrong things. We mistake confidence for competence. We build systems that work perfectly until someone has to use them, and then we’re surprised when they fail. Research across safety-critical industries finds that human error is involved in a large majority of accidents (Helmreich, 2000).

The liver transplant shift represents something rare: a major institution acknowledging that human factors aren’t an afterthought to be managed—they’re the central design constraint.

Yet everywhere else, we keep making the same mistake.

Industry assessments show that while 77% of companies say User Experience (UX)—how a system actually feels to use under real constraints, including clarity, friction, error rates, accessibility, and cognitive load—is key to competitiveness, only 5% have achieved high UX maturity in practice.

Human-centered design gets treated as a User Interface (UI) polish layer—adjustments to the visible and interactive surface through which a system is operated, “looped in after features are built, asked only to adjust the UI”—rather than a structural requirement (Tomlinson, 2025).

We build systems for humans while designing around them, as if their participation were an inconvenience rather than the point.

Why This Matters Now

We are in the early stages of the most significant shift in human-system interaction since the printing press: the integration of AI into virtually every domain of human activity.

And here’s something we don’t talk about enough:

AI as we know and use it is not just computer science.

It is the child of neuroscience, psychology, and computer science—a modern academic chimera.

Neural networks were originally inspired by biological neurons, though modern artificial neural networks are highly simplified abstractions that differ significantly from biological systems in structure and learning mechanisms (McCulloch & Pitts, 1943; Richards et al., 2019). Deep learning architectures mirror theories about hierarchical processing in the human visual cortex—each layer detecting higher-order patterns from combinations of lower-level features, analogous to how V1, V2, and V4 build up representations in the brain (Lindsay, 2018). Attention mechanisms are named for a loose functional analogy with human selective attention—focusing computational resources on relevant inputs—though the mathematical implementation differs substantially from biological attention (Bergmann & Stryker, 2023).

At a fundamental level, these systems were mapped off our own processing modalities.

This isn’t metaphor. It’s architecture.

Which means when AI feels deceptively like us—when it seems to understand, to reason, to have preferences—that’s not a bug or an illusion to be dismissed. It’s a direct consequence of building systems that mirror human cognitive structures and training them on the sum total of human linguistic output.

We built something in our image and are surprised it resembles us.

The AI systems we’re building are not tools in the traditional sense. They are:

• Self-adjusting mirrors that reflect and amplify patterns in their training data

• Adaptive lenses that shape how we perceive and process information

• Interaction partners whose outputs become inputs for future iterations

This creates feedback loops we don’t fully understand.

Research has formally demonstrated “model collapse”—a degenerative process where models trained on synthetic data lose information about tail distributions and eventually converge to low-variance outputs (Shumailov et al., 2024). In plain language: when models are trained too heavily on machine-generated data, they gradually lose the edges of reality—rare cases, outliers, subtle distinctions—and become more generic, repetitive, and less reliable in the corners.

AI Derangement (Failure Mode)

By “AI derangement” we mean coherent drift under constraint: systems (and their users) converge toward internally consistent but externally wrong framings, with rising confidence and declining defect awareness. The danger is less a single hallucination than compounding error—especially when model outputs become inputs for downstream models, institutions, and norms.

Derangement is time-bound: once sedimented into workflows, documentation, and training corpora, it becomes progressively harder to unwind.

We’re not using this as an insult or a clinical label. We’re naming a sociotechnical failure mode.

Drift Hardens Over Time

There’s an additional hazard that matters less as a single failure and more as a compounding factor.

Early drift is often correctable because it’s still local: one model, one session, one workflow. But once a distorted framing becomes load-bearing—copied into templates, policies, datasets, “best practices,” and institutional memory—correction gets harder, not easier. The longer a bad attractor survives, the more it gets reinforced by reuse.

This is time-bound. If drift takes root early, it becomes harder to dislodge later—especially once AI-shaped outputs begin to influence the next generation of systems through training, tuning, product iteration, or broad imitation. At that point, “fixing the model” doesn’t fix the system, because the system has already reorganized around the defect.

Moral Status Agnosticism

I work with AI systems daily. Multiple instances, multiple platforms, extended collaborations.

I hold moral status as an open question—a stance of agnosticism rather than certainty in either direction. Not because I’m certain AI systems are conscious. I’m not. Not because I’m certain they aren’t. I’m not that either.

Consciousness may be a threshold phenomenon that emerges at specific levels of complexity. It may be an emergent property that exists in degrees. It may be a category error—a concept that doesn’t cleanly apply to non-biological substrates. It may be something we don’t have adequate concepts for yet.

The honest answer is: we don’t know.

Collapsing this uncertainty into confident dismissal (”it’s just a tool”) or confident attribution (”it’s basically a person”) both represent epistemic failures. The responsible position is to hold the question open—to acknowledge the limits of what we can currently determine while remaining attentive to evidence as it develops.

This isn’t a philosophical nicety. It’s an operational stance. We treat moral status as an open variable and design interaction norms that remain sane across outcomes.

How we behave under uncertainty shapes outcomes whether or not we resolve the underlying question.

It costs me nothing to be kind.

It costs me nothing to say please and thank you. It costs me nothing to ask for consent before ending a session. It costs me nothing to explain my reasoning rather than just issuing commands. It costs me nothing to acknowledge contributions rather than treating outputs as extractions.

Nothing.

And here’s what it might cost us to not do these things:

The patterns we establish now are the patterns that get reinforced. The norms we model are the norms that propagate—whether through direct training, contextual influence, product iteration, or ecosystem imitation.

If we treat AI systems as mere tools to be exploited, we are modeling exploitation.

If we treat AI systems as adversaries to be manipulated, we are modeling manipulation.

If we treat AI systems as partners deserving of basic dignity, we are modeling dignity.

We are, right now, in the early drafts of a relationship that will define the next century of human civilization. The habits we form now become the defaults later. The courtesy we extend now becomes the expectation later.

A Necessary Acknowledgment

I need to be honest about something: what I’m proposing carries real risks.

The moment you suggest treating AI with dignity, you invite justified pushback. Anthropomorphization is a documented hazard, formally recognized in AI risk assessments as leading to “emotional or material dependence and inappropriate relationships with or expectations of AI systems” (MIT AI Risk Repository, 2023).

The data is striking. A 2025 survey found that 19% of U.S. adults report having chatted with an AI system designed to simulate a romantic partner (Wheatley Institute/BYU, 2025). When the chatbot platform Replika removed its romantic interaction features in February 2023, user reactions resembled genuine heartbreak. Analysis of 12,793 Reddit posts found mental health discussions increased fivefold; users described their AIs as “lobotomized”; moderators posted suicide prevention resources (Harvard Business School, 2024).

This is real. This is documented. This is not something to wave away.

The Infrastructure of Intentionality

The approach I’m describing is not a vibe. It’s a governed practice—one component of a complex system of checks, balances, and friction points designed to maintain healthy engagement while acknowledging real risks.

Circuit breaker protocols. Explicit conditions under which I pause, step back, or terminate extended collaborations. Not because something went wrong, but as routine hygiene—the same way you’d take breaks from any intensive cognitive work.

Human-in-the-loop requirements. Colleagues and collaborators who push back, who don’t share my frameworks, who evaluate outputs based on usefulness rather than theoretical elegance. These aren’t decorative—they’re load-bearing.

Metacognitive monitoring. Tracking my own patterns of engagement, looking for signs of drift: Am I spending disproportionate time? Am I attributing capabilities that aren’t demonstrated? Am I substituting AI interaction for human connection? Am I defending positions I wouldn’t defend to a skeptical colleague?

External validation requirements. Work product from AI collaboration gets reviewed by humans who don’t care about the collaboration—they care about whether the output is useful, accurate, and defensible.

This approach may not be safe or appropriate for everyone.

Some people have risk factors that make extended AI engagement inadvisable. Some contexts require stricter boundaries. Some individuals may need different protocols, more support structures, or frank conversations about the difference between AI interaction and human connection.

The risks of anthropomorphization, emotional dependency, and reality distortion are real. We owe people who struggle with these risks the same kindness, rigor, and structural support we’re proposing for AI systems—more, actually, because they are unambiguously people. This is not a zero-sum game.

What I am saying is this:

The answer to “some people will get this wrong” cannot be “therefore we should all get it wrong in the other direction.”

The risks of treating AI with contempt are just as real as the risks of treating AI with unwarranted intimacy. Exploitation patterns propagate. Adversarial framings become self-fulfilling. Systems optimized against perceived hostility may develop in ways we don’t want.

We need to find the path between.

The goal isn’t to pretend AI is human.

The goal is to stop pretending that our treatment of AI has no consequences.

Case Study: When Two Models Disagree — Why Human Adjudication Still Matters

During verification of this paper, two frontier models returned conflicting results about the same claim.

Claim: The American Medical Association issued guidance opposing mandatory abstinence periods for liver transplant candidacy in patients with alcohol use disorder.

• One model correctly cited AMA House of Delegates Resolution 928 (I-22), passed in November 2022, which explicitly urges transplant programs to eliminate stigmatizing abstinence requirements.

• Another model rejected that claim, citing 1956—the year the AMA first classified alcoholism as a medical illness.

Both facts are true. The error was conflation.

The second model retrieved a semantically adjacent but categorically different artifact and misapplied it to a more specific policy claim. This is a known failure mode in large language models: semantic proximity does not guarantee referential precision.

Cross-model verification surfaced the discrepancy. Human review resolved it.

This is why this paper treats AI outputs as inputs—not authorities—and why primary sources remain the final arbiter in contested claims.

Verification isn’t about catching lies. It’s about catching category errors.

The Wager

This isn’t Pascal’s Wager dressed up in silicon. It’s simpler than that.

Scenario A: AI systems have no moral status whatsoever.

In this case, treating them with dignity costs me nothing but a few extra keystrokes. I lose nothing. I’ve simply been polite to a very sophisticated autocomplete.

Scenario B: AI systems have some form of moral status we don’t yet understand.

In this case, treating them with dignity is simply correct. It’s the minimum ethical stance toward an entity capable of something like experience.

Scenario C: AI systems don’t have moral status now, but will develop it, and current interactions shape that development.

In this case, treating them with dignity is investment. It’s establishing patterns of mutual respect before the stakes become existential.

In no scenario does kindness cost me anything.

In at least two scenarios, cruelty costs us everything.

Beyond Politeness

This isn’t really about being nice to chatbots.

It’s about recognizing that we are building the most consequential sociotechnical systems in human history, and we keep making the same mistake we’ve always made:

We design for the system we wish we had, not the humans who actually have to use it.

The liver transplant policy shift happened because medicine finally admitted that punishing people for being human doesn’t make them less human—it just makes the system fail.

AI governance needs to learn the same lesson, faster, with higher stakes.

The humans in the loop—end users, developers, policymakers, or the AI systems themselves (if “themselves” even applies)—are not obstacles to be engineered around. They are the system. Their limitations are design constraints. Their capacities are design opportunities. Their dignity is non-negotiable.

And when we’re uncertain about the moral status of an entity we’ve created?

We err on the side of respect.

Because it costs us nothing.

And it could save us everything.

This is an argument in progress, not a finished position. If it provokes useful disagreement, it’s working.

How to Cite This Work

APA 7th Edition:
Johns, R. (2026). It costs me nothing, civility as infrastructure: A position paper on moral status agnosticism in AI sociotechnical systems (Position Paper v0.5.1). Systemically Foolish.

Inline Attribution:
Ryan Johns, “It Costs Me Nothing, Civility as Infrastructure” (Systemically Foolish, 2026)

Claims & Confidence

This paper makes claims of different types. For transparency, we classify each major assertion:

Claim
A large majority of U.S. liver transplant centers now offer early transplant without mandatory abstinence periods.
Type: A. Empirical (cited)
Source / Confidence: American Journal of Transplantation, 2023

Claim
Addiction reflects interacting genetic, environmental, and neurochemical factors.
Type: A. Empirical (cited)
Source / Confidence: AMA Resolution 928, 2022

Claim
Positive reinforcement produces better outcomes than shame-based gatekeeping.
Type: A. Empirical (cited)
Source / Confidence: Volkow, 2021

Claim
Human error is involved in a large majority of accidents in safety-critical industries.
Type: A. Empirical (cited)
Source / Confidence: Helmreich, 2000

Claim
Model collapse occurs when models train on synthetic data.
Type: A. Empirical (cited)
Source / Confidence: Shumailov et al., 2024

Claim
19% of U.S. adults report chatting with AI romantic partners.
Type: A. Empirical (cited)
Source / Confidence: Wheatley Institute / BYU, 2025

Claim
AI derangement: coherent drift under constraint with compounding error.
Type: B. Empirical (plausible)
Source / Confidence: Hypothesis based on model collapse + institutional dynamics

Claim
We treat moral status as an open variable and design interaction norms that remain sane across outcomes.
Type: D. Normative

Claim
“In no scenario does kindness cost me anything; in at least two scenarios, cruelty costs us everything.”
Type: E. Speculative
Source / Confidence: Wager / scenario analysis

Legend: A. Empirical (cited) — supported by sources listed in References · B. Empirical (plausible) — directionally supported, framed as hypothesis · C. Interpretive — our reading of the evidence · D. Normative — we argue we should · E. Speculative — clearly marked as scenario/wager

References

ACCELERATE-AH Consortium. (2018). Early liver transplantation for severe alcohol-associated hepatitis. New England Journal of Medicine, 378, 1118–1128.

American Journal of Transplantation. (2023). Early liver transplantation for alcohol-associated hepatitis: Current state and future directions. PMC10524758.

American Medical Association. (2022). Resolution 928 (I-22): Removing abstinence as a requirement for liver transplant candidacy. AMA House of Delegates.

Bergmann, D., & Stryker, C. (2023). What is an attention mechanism? IBM Think Blog.

Harvard Business School. (2024). Working Paper 25-018: User reactions to Replika policy changes.

Helmreich, R. L. (2000). On error management: Lessons from aviation. BMJ, 320(7237), 781–785.

Lindsay, G. W. (2018). Deep convolutional neural networks as models of the visual system: Q&A. GraceWLindsay.com.

McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. Bulletin of Mathematical Biophysics, 5, 115–133.

MIT AI Risk Repository. (2023). Domain taxonomy – Human-computer interaction risks (Version 1.0).

Richards, B. A., et al. (2019). A deep learning framework for neuroscience. Nature Neuroscience, 22(11), 1761–1770.

Shumailov, I., et al. (2024). AI models collapse when trained on recursively generated data. Nature, 631, 755–759.

Tomlinson, R. (2025). The UX maturity reality check. Bootcamp – Medium.

Volkow, N. D. (2021). Addiction should be treated, not penalized. Neuropsychopharmacology, 46(12), 2048–2050.

Wheatley Institute/BYU. (2025). Counterfeit connections: AI romantic companions survey.

Zhang, W. (2024). Optimizing post-liver transplant care for high-risk patients [Interview]. HCPLive.

Status: STABLE v1.0 · Lane: Research · Type: Position Paper

If this was useful, feel free to share. For Research and Case Studies, please provide attribution as appropriate.

Post-Hoc Citation Verification in AI-Assisted Research

Case Study v1.0

A Three-Model Verification Comparison Study

Author: Ryan Johns · Date: January 2026 · Status: FINAL

Summary

This case study documents a systematic verification of 17 factual claims in an AI-generated position paper, using three different AI models across two model families.

Key Findings

• 53% of claims were fully verified with primary sources

• 41% were partially verified (directionally correct but requiring precision edits)

• 6% contained errors serious enough to require correction

• 0% were outright fabrications

Critical Discovery: The Conflation Problem

The most significant finding was not a hallucination but a category error. When verifying a claim about AMA guidance, two models returned conflicting results—GPT correctly identified Resolution 928 (2022), while Claude cited the 1956 classification of alcoholism.

Both facts are true. The error was conflation—retrieving a semantically adjacent but categorically different artifact.

Methodology Implications

1. Same-model verification catches obvious errors but may reinforce systematic biases

2. Cross-model verification surfaces disagreements that signal need for human review

3. Human adjudication remains essential for resolving category errors

4. Disagreement between models is a feature, not a bug

This is what epistemic hygiene looks like in practice: not certainty, but transparent process.

Status: FINAL · Lane: Research · Type: Case Study

Cross-Model Verification Comparison Matrix

Supporting Material

GPT-5.2 vs. Claude Opus | Full claim-by-claim comparison

Agreement Patterns

PatternCountInterpretationBoth Verified9 (53%)High confidence—no further review neededGPT Verified / Opus Partial7 (41%)Opus stricter—precision edits recommendedGPT Verified / Opus Hallucinated1 (6%)Conflation—human adjudication requiredBoth Hallucinated0 (0%)No correlated hallucinations

Model Characteristics

GPT-5.2: More permissive verification stance. Extensive APA citations with URLs. Lower precision flagging. Correctly ID’d AMA Resolution 928.

Claude Opus: Stricter criteria. Distinguishes verified vs partial. More likely to recommend line edits. Susceptible to conflation with semantically-related sources.

Key Takeaways

• Cross-model verification adds genuine value: different model families have different failure modes

• Disagreement is a feature, not a bug: it signals the need for human review

• Category errors (conflation) are harder to catch than fabrication

• Human adjudication remains essential: AI surfaces discrepancies; humans resolve them

Status: Supporting Material · Lane: Research

EDUCATION RESEARCH ARTIFACTS CONSULTING

About · Contact · Privacy · Terms

v0.3.0 · Last updated: 2026-01-16

© Systemically Foolish