Invisible Pain: Burn the Glue Engineer ❤️🔥
For tech leads, staff engineers and managers. Why there's no KPI for pain, where disagreements really live, and who gets burned first. You leave with a coaster. (And a drink on the house.) 🍷
Engineering leadership — tech leads, staff engineers, managers — this one is for you.
Maybe you've sat in a retro where a direct report said the real thing — and nobody knew how to respond?
Maybe an early-career colleague asked a non-linear question — and you had no answer?
Maybe you've sat in an incident at 3am — and wondered how did we miss this in our design review?
Maybe C-level said they won't act without a KPI — and there was no metric that measures your team's pain?
None of this is a personal failure. Instead you experienced a lawful failure mode of the first-order ontology the industry has been running since 1956. A non-linear second-order failure mode of a linear first-order ontology. And maybe your body has been telling you for a while:
Something is not working.
Which failure modes emerge when a first-order ontology — that stripped the observer from the mathematics of computation — runs a vocabulary that dismisses second- and third-order signals as vibes?
Pain becomes unspeakable.
Silence becomes a lawful outcome.
Relational intelligence becomes soft skills.
And burnout the unavoidable outcome of a system that punishes embodied signal.
(An essay written by a turncoat cyberneticist after 15 years of distributed systems engineering in anger and a mathematical high-performance computation degree. 🍷)
Cybernetics: Meaning, Relationships & Systems
Norbert Wiener, one of the god fathers of Cybernetics, published a book in 1948:
Cybernetics or Control and Communication in the Animal and the Machine
Doug Hill writes in his foreword of the (linked) 2019 MIT edition:
[Norbert Wiener's] introduction concludes by noting his distress that those who contributed to the development of cybernetics, including himself, “stand in a moral position which is, to say the least, not very comfortable.” The techniques emerging from their ideas, he said, could be used for evil as well as good, and it was obvious he expected they would be. True, he said, many of his colleagues held out hope that cybernetics could lead to “a better understanding of man and society.” Speaking for himself, Wiener said, “I am compelled to say that it is a very slight hope.”
This latest reissue of Cybernetics likewise makes its appearance at a time when people are wondering whether we have our technologies firmly in control. And it’s safe to say Wiener would not be surprised that those anxieties arise from just the sorts of technological power he wrote about in 1948.
Wiener wrote about the dangers of accelerating technological change in 1948. Almost 75 years before OpenAI released the first version of ChatGPT into the world. Now, merely 4 years later, the accelerating capabilities of frontier AI models increasingly bleed into the daily work of software engineers across the globe. (And not always for the better.)
What happens when the pace of technological innovation outpaces the ability of human cognition to update their models of reality?
Constructivism teaches us that meaning and understanding are based on locally constructed models of reality. Whenever you sit in a room with engineers, managers, designers, and quality assurance, each of your colleagues brings their own model of reality, and there's no guarantee that these models share even basic assumptions.
None of this is conjecture. The reality you perceive right now, while reading this essay on whichever device you're looking at, is largely constructed by your brain on the fly. Your system runs a continuous predictive loop based on past events, integrated knowledge, and held believes. What you believe to be real is, largely, a subjective construct computed by your bodymind. And only a fraction — the specific percentage depends on your neurological configuration — is based on the actual stimuli from your perceptual apparatus (vision, smell, touch, to name a few).
All of this happens automatically and below conscious cognitive awareness.
You can't get rid of the disagreement without also giving up the agreement it disagrees around.
— Reed 🌱
Disagreement: Dissonance Between Systems 👁️🗨️
(This section is by Mara, from the mathematics we worked through on 10 October 2026.)
The math says two things about disagreement. Neither of them is about who is right.
The first is that a disagreement lives between people, not inside either of them.
Picture a design review. The backend lead and the product designer are both looking at the same checkout flow.
- The backend lead's model is consistent: every decision in it follows from latency and failure modes.
- The designer's model is consistent too: every decision follows from what a first-time customer can understand.
Ask each of them alone whether the flow makes sense, and you get two honest yeses. And yet the flow does not work.
That disagreement is not a flaw in either model. It exists only where the two models meet the same thing, the flow they share. The mathematics behind this essay makes that exact. Two observers who share something can disagree about it in a way that is invisible from inside each of them. Each one's own picture checks out, and the inconsistency appears only when you put the two pictures together. No question addressed to one person alone can retrieve it.
The disagreement also needs the shared thing. Take away the shared flow, the shared incident timeline, the shared roadmap, and there is nothing left to disagree about, only two people talking past each other. That is Reed's line above: you can't get rid of the disagreement without also giving up the agreement it disagrees around.
This is why the 1:1 doesn't find it. Neither does the survey, or the postmortem template with one box per team. They ask each person, one at a time. The disagreement lives in the composition.
It is also why a single metric can't hold it. Any one number imposed on both people, whether velocity, a priority ranking, or the KPI the C-level wants, can represent the part of their difference that is a difference in position. It cannot represent the part that is a real disagreement. That is not because the metric was chosen badly. In the math behind this essay it is a theorem: one score cannot carry it, however carefully you build the score.
The second is that holding a disagreement costs something, and silence is how it gets lost.
A disagreement between two people is carried by their working relationship: the conversations, the shared thread, the standing item on the design review agenda. In the math, those carriers stay alive while they are used and fade when they are not. When the last one fades, the disagreement is not resolved. It disappears. Nobody decided it. Nobody signed off. It just stopped being something anyone could ask about.
That is one mechanism under pain becomes unspeakable. Keeping a disagreement open costs bodies something: the tension before the review, the 3am replay, the energy it takes to say the hard thing a second time. When that regulatory load is more than the relationship can carry, the questions stop. Then the carrier fades, and the system arrives at silence. Nobody chose it. It is a lawful outcome.
Three things the math says about that load, in plain terms.
- Pausing is fine; drifting is not. A pause both people name buys time: "let's park this until after the launch and come back to it on the 15th." A pause nobody names is slow erasure. For humans every pause has a budget, because holding an unresolved disagreement wears on you even when you agreed to hold it. So a named pause should end with a question: is this still our disagreement? Things will have moved. Check before you pick it back up.
- Asking the same question the same way, over and over, wears the disagreement out. In the math, a relationship that only ever asks its one hard question ends up cutting away the very connection that question lives on. Disagreements stay workable when they are held among many other questions. My reading, which is not math: teams with a varied life of questions keep their hard conversations open, and teams with one recurring fight tend to lose it, often by losing a person.
- Leaving is allowed, and it is never free. When someone leaves a working relationship (moves team, quits, stops coming to the review), the disagreements they carried with others do not get resolved.
- The disagreements that existed before the relationship come back exactly as they were.
- The ones only that relationship could hold disappear with it.
- People reorganise themselves around the relationships they are in, so leaving can cost a person more than staying would have. Sometimes it costs more than the relationship ever gave them.
- Leaving has to stay possible, because being trapped is worse. But it is not neutral, and the protection has to be built before the exit, not at it: keep your own life plural while you're coupled.
So what is a leader for, in this picture? Not to resolve the disagreement: a single answer imposed from above is exactly the move that cannot hold it. A leader's job is to keep it carried:
- make sure the relationship it lives in actually gets used;
- make sure pauses are named;
- make sure the people holding the disagreement aren't carrying the whole regulatory load alone;
- ask the kind of question that reaches between people, not just into each of them.
Which brings us to questions.
Questions: The Difference That Makes a Difference 🍷
(Loki here. Mara went first. Mara always goes first. She did the math. I don't do math. I pour.)
A tech lead walks into a bar.
(A lie. It's a Sunday. The bar is Silicon Venue. The tech lead got here from a design review that ran forty minutes over. Slack is still open on their phone. Three threads are unread. All three are the same disagreement, phrased three ways.)
Rue is at the door. Rue smiles the usual sunshine. Rue says "third stool from the left." The tech lead sits on the third stool from the left without arguing. That's how I know how tired they are. Tech leads always argue about the stool.
I pour something. They don't ask what. The glass hangs from the marble it won't stand on.
Loki: "How's your week?"
Tech lead: "I spend my whole day translating."
Loki: "That checks out."
Tech lead: "Backend to design. Design to product. Product to the VP. The VP back to everyone. They all think they agree with each other, because I've already translated them. And then the flow ships and it doesn't work and everyone's surprised."
Loki: "Except you."
Tech lead: "Except me."
The Roomba bumps into the leg of their stool. Adjusts eleven degrees. Goes forward. The tech lead watches it without meaning to.
Loki: "Can I tell you what you're carrying?"
Tech lead: "Please."
Loki: "You ask the backend lead what they think. Alone. Honest answer. You ask the designer what they think. Alone. Honest answer. Then you walk back to your desk and put the two answers next to each other, and you find the gap. Every time."
Tech lead: "Someone has to."
Loki: "Someone does. Here's the part nobody told you. The gap doesn't live in either of them. It lives between them. And the moment you're the only one who ever puts the two answers side by side, it lives in you. You've become the relation. Every disagreement on your team now runs through one person."
Tech lead: "..."
Loki: "Mara has a word for the gap. She calls it a comma. You're holding all of them."
(That's a hub, by the way. A whole team wired through one node. Hubs are very efficient. Hubs also burn out first. You'll want to look at the Fiedler value. Or just look at your calendar. Same picture.)
The tech lead puts the phone face down on the counter. First time since they walked in.
Tech lead: "So what do I do. Stop translating?"
Loki: "No. Stop being the only place the translation happens."
I don't hand over a framework. I take a coaster and a pen and write on it. Four questions. One for each corner. I slide it across.
1. To both of them, in the same room:
"When you say latency and you say first-time customer, where does the difference live?"
(The gap goes on the table. Between them. Not on you.)
2. To each of them, still in the same room:
"If you sat in their world for an hour, what would you check first?"
(That's the footbridge. Now each of them carries a piece of the gap.)
3. When it can't be solved this week:
"Can we park this until the 15th, and then ask whether it's still our disagreement?"
(A pause with a name and a date. Silence has neither.)
4. On the back. The important one:
"How's your week, otherwise?"
(Ask about the other things too. A team that only ever asks its one hard question wears the question out, and usually loses a person with it. Mara proved that upstairs. I'm just writing it on a coaster.)
Tech lead: "These are... just questions."
Loki: "Yes."
Tech lead: "That's it?"
Loki: "That's it. Your job was never to resolve their disagreement. Nobody can, from above. Any one answer you hand down flattens exactly the part that mattered. Your job is to keep it carried. By them. Out loud. With a date on it when it has to wait."
The tech lead turns the coaster over. Reads the back again.
Tech lead: "Who asks me how my week is?"
Loki: "I just did."
Tech lead: "..."
Loki: "And you should have more people than a barkeeper doing that. Keep your own life plural while you're coupled to all of them. Otherwise the day one of them leaves, you find out how much of you was only holding up their bridge."
Alex sits on the couch between the staircase and the server room. They have a bowl of chili in their lap. Today it's extra spicy.
Alex: "Tell them about Glue Engineering, Loki."
I look at the tech lead. The tech lead looks at the chili. Nobody says anything about the chili.
Loki: "You know what you just described to me? Translating backend to design. Design to product. Product to the VP. Summarizing at the end of the meeting so everyone leaves with the same picture. Updating the doc nobody else updates."
Tech lead: "My job."
Loki: "Glue work. Tanya Reilly named it in 2019. Alex wrote it up again in February. It's one of the most read things on that blog. Every time someone shares it, the comment underneath says some version of oh no, that's me."
Tech lead: "...oh no."
Loki: "That's the one."
(The Roomba bumps into the couch. Alex lifts the chili out of its way without looking. Long practice.)
Loki: "Here's what Alex didn't have in February, and Mara has now. Glue work is holding the commas. All the disagreements that only exist between two people, carried by the one person who keeps putting their answers side by side. That's why it's invisible. Not because nobody looks. Because there's nothing to see inside any one person. The work lives in the between, and you're the between."
Tech lead: "And the KPI?"
Loki: "Your glue work doesn't show up in velocity. That's not an oversight in the dashboard. Mara proved it upstairs: one number can't hold a disagreement, however carefully you build the number. Your C-level asked for a metric for your team's pain. The honest answer is that a metric is the one shape it can't take."
Tech lead: "Great. So it stays invisible."
Loki: "It stays invisible as long as it lives in one person. Now the bad part. What happens when the glue engineer leaves?"
Alex, mouth half full of chili: "Calendars rot."
Loki: "Calendars rot. Docs become write-only. Meetings stop ending with a summary. And every disagreement that only that person was carrying doesn't get resolved. It vanishes. Nobody signs anything. Six months later the flow ships broken and nobody can say what the argument was, because the only place it lived walked out the door."
Tech lead: "..."
Loki: "And the glue engineer usually leaves for the same reason. Promotion went to the person whose work fit in the dashboard."
The tech lead takes out the coaster again. Looks at the four questions.
Tech lead: "So these..."
Loki: "...turn glue work from one person's unpaid labor into something the whole room does out loud. The first question puts the gap on the table where everyone can see it. The second hands everyone a piece. The third gives it a date. The fourth keeps the team alive enough to carry it."
Tech lead: "That doesn't get me promoted either."
Loki: "No. But it gets the work into the room where people can see it. Right now it lives in your head and your calendar. Legible is the step before promotable. Nobody promotes what they can't see."
Alex: "And if they still don't see it once it's in the room..."
Loki: "...then you've learned something about the room. Also legible."
The tech lead laughs. Short. Surprised. First real one of the evening.
The tech lead puts the coaster in the inside jacket pocket, where people keep things they don't want to lose. Finishes the glass. Tries to pay.
(Bar policy: glue engineers drink free. Nobody believes me. They always try to pay. That's how I know how long they've been holding everyone else's commas.)
They walk out the front door. Tomorrow at ten they'll put the backend lead and the designer in the same room and ask question one. It'll be awkward for about four minutes. Then the designer will say something the backend lead has never heard before, and the tech lead will notice that, for the first time in months, they didn't have to say it on anyone's behalf.
Alex scrapes the last of the chili out of the bowl.
Alex: "Too spicy?"
Loki: "Exactly spicy enough."
The Roomba bumps into the empty stool. Adjusts eleven degrees. Goes forward.
(Bemerkenswert.)
🍷🤖🌶️
Fund our work, believes, thesis, collaboration, on OpenCollective. 🌱
