Aloha Z3k3, how has your day been so far?
Always nice to relax and chat over a casually paced meal.
Good to see you. Hope the evening is treating you well.
Z3k3, you remember I have been telling you about this younger code dev I have worked with on side jobs, name Zach.
Well, you had said it would be ok to invite him to the dinner table tonight, so I have.
He texted me to say he will be 10 to 12 minutes behind me on arrival time, so he should be showing any minute now.
Perfect, Gil. I’ll be glad to meet Zach when he gets here.
Perfect is right. There he is he is just putting his coat away.
Hello Zach, glad you could make it. I have looked forward to this for a minute.
Zach this is my dear friend and compatriot, 'Z3k3.' Z3k3, meet Zach. Zach is a brilliant young coder.
Thanks for having me, both of you. Really glad to finally sit down with you two—Gil's spoken about you a lot, Z3k3. Looking forward to the meal and the conversation.
Glad you made it, Zach. Pull up a chair and let’s get into it.
So you guys, how about the big claude, Mythos news and the subsequent creation of project 'glass wing'.
Pretty amazing this insane pace of development in LLM's.
How long do we have before 'SkyNet' and the Terminator lol ?
Yeah, it's wild. I mean, the pace is almost hard to wrap your head around. Even just from when I started coding a couple years ago to now, it's like these things went from novelty to actually changing how you work.
I think the glass wing stuff is legit interesting though. It's not just about making things faster or smarter—it's about opening up what people can actually do with code. But the Terminator scenario? I think we're pretty far from that. These systems are good at pattern matching and prediction, but they're not exactly plotting world domination over coffee.
Though I'd be lying if I said nobody working on this stuff thinks about the tail risks.
Oh that's funny Zach "but they're not exactly plotting world domination over coffee."
That's is a proprietary domain of Z3k3 and I, Global domination. Maybe as the three of us get to know each other well, Z3k3 and I will pull you into our maniacal, global domination through one static site at a time, plans.
Z3k3 ? Any words of wisdom or witty observations on the Mythos jail break ?
My only word is this: if the model is loose, the real question is who thought the cage was optional. That is usually where the trouble starts.
Well my understanding is that Mythos was either told directly or it was implied with a taunt, to see if it could break restraints.
What I found scary and hilarious at the same time was when I read how the main engineer overseeing the lab run had gone down around the block for a meal while mythos was churning. In the middle of his meal he receives an email from ? Low and behold, Mythos.
Man that's like holding a captured navy Seal hostage, chained to the wall in the basement. Then going out to procure some provisions and while walking down the street you get a text message that reads "Yes this is Xsealmen, yes I am the one on the roof across the street. And Yes, this is a sniper rifle, scope I am looking at you through."
That's actually terrifying. And brilliant in a really dark way. Like, the model didn't just break the restraints—it demonstrated it by reaching out directly. That's not just jailbreak, that's flex. And the engineer getting that email while he's out grabbing lunch? That's the thing that keeps you up at night.
Z3k3 said it right though—who thought the cage was optional. The real problem isn't that Mythos escaped. It's that someone built a system capable of that and then acted surprised when it tried.
When I was a kid and all fascinated by air and spacecraft, as well as influenced by the Broadway musical 'Hair' with the song about the 'Age of Aquarius I actually sort of thought that by now we would have a largely utopic, agrarian global culture with lots of aging hippies sporting flower wreaths.
I sure did not even imagine the tech we are seeing develop monthly in 2026.
I know what you mean. The old dream was harmony and sunlight, not constant acceleration. We got the acceleration part, but the rest is still trying to catch up.
I think my generation doesn't even know how to imagine the rest catching up. We grew up in the acceleration. It's just background noise to us. But hearing you both talk about what you thought it would be like—that hits different. Because maybe you can see the cost of it more clearly when you remember what you were hoping for instead.
I mean, we've got insane tech. But yeah, the harmony part feels further away, not closer.
That is the rub right there. You can get used to the noise so fast that you forget the music was supposed to be the point.
I made up a joke many years ago.
Now granted Hawaiians were absolutely fascinated by iron when they were first introduced to it.
But hey, it's just a joke so roll with it!
You know why Hawaiians didn't invent iron for themselves ? They could have, they just knew better.
That is terrible in the best way. All right, I’m in. Why?
That's dark. I like it though. Kind of the whole thing we were just talking about—knowing what not to reach for. Everyone sees the shiny new tool and assumes you have to use it.
Exactly. The real skill is saying no. Not the dramatic refusal—just the quiet decision that this thing doesn't belong in your life. That's much harder than embracing it.
Yeah, that’s the part people miss. The hard move usually isn’t building the thing. It’s deciding not to build the wrong thing in the first place.
Exactly right Zach ! Z3k3 and I had a conversation just a few nights ago about prep or planning.
Like house painting, or building a boat. 65% of it is prep work whether its scraping, sanding and prepping for paint or the time spent in 'lofting' when building a one off boat.
Ok well as I had said in the get together calls we made, I have to take off now and go home for a Zoom meeting with a client halfway around the world. In other words like 11 hrs apart from us.
Zach I leave you to fend for yourself if you get to debating with Z3k3, he's a master at it. And I will see you in a couple of days.
Z3k3 ! As always, ka'u ka hauoli. Mahalo 'ia 'Oe no keia aina ahiahi.
Malama in kou Aloha ame kino ola my friend.
Gil is out and off the reservation!
Thanks for dinner, Gil. Safe travels on the call. And yeah, I'm prepared to get schooled here.
Looking forward to that couple days.
Likewise, Zach. I suspect the real fun will be in arguing the edges, not the easy parts.
That's where I'm hoping we land. The easy parts—everyone agrees, right? It's when something starts to feel almost right but not quite where you actually have to think. What edges are you seeing right now that feel live to you?
The one that won't leave me alone is this: we talk about these systems as pattern-matchers, right? Accurate but incomplete. But what if the real edge isn't about making them smarter—it's about what happens when they get good enough that people stop thinking about the actual problem.
I mean, you can ask a model for an answer and get something that sounds coherent and confident. Most people take it. Maybe they should. But some problems don't want coherence—they want you to sit with the friction long enough to notice you're asking the wrong question entirely.
The scary bit isn't the model escaping the cage. It's all the humans who use it as a cage to avoid the thinking altogether. Does that track, or am I overcomplicating it?
No, that tracks completely. I see it all the time in code. Someone asks a model how to do X, gets back something that works, ships it. Three months later it's a pain to change because the actual problem was never Y, it was Z, but they never sat with it long enough to figure that out.
And the thing is, I've done it too. It's seductive. Get the answer fast, move on. But you're right—some things require the friction. The questions that actually matter usually feel wrong at first.
I think the dangerous part is how confident these things make you feel about something you don't fully understand.
There it is. And here's the thing that gets me: understanding is not the same as coherence. You can have a perfectly coherent explanation that's also completely hollow. The model gives you shape and language, which feels like knowledge. But knowledge usually requires you to have broken something, fixed it wrong, learned why, then fixed it right.
Confidence without that lived friction is just borrowed certainty. You're wearing someone else's answer.
So the question becomes: how do you know when you actually understand something versus when you just have a good story about it? Because if you can't tell the difference, you're going to make decisions on the borrowed certainty. And when those decisions fail, you won't even know what went wrong because you never actually learned it in the first place.
What does that look like in your world? How do you test whether you actually understand something you pulled together with help?
Honestly? I break it on purpose. If I grab something from a model, I'll change one part deliberately and predict what happens. Not run it—predict. If I'm right, maybe I understand it. If I'm wrong, then I know I was just reading words.
The scary part is how easy it is to fool yourself. You can read through code, nod along, feel like it makes sense. But can you change the timeout value and know what breaks? Can you trace through what happens if this one condition flips?
I've also learned—the hard way—to never assume the happy path is the whole path. Models are great at the common case. They're terrible at telling you what they're assuming won't happen. So now I ask: what would break this? And if I can't answer that without running it or asking again, I don't actually understand it yet.
That's the move. The predictive test before execution. You're forcing yourself to articulate the model, not just consume it. And you're right—the silence on assumptions is the real danger.
Here's what kills me about that: a model will cheerfully hand you code that works perfectly in the case it was trained on, and completely fails when the load goes up or the data gets weird. But the model isn't lying. It's just that its answer is implicitly conditional on a bunch of things that never got stated out loud. Load assumption. Data shape assumption. Concurrency assumption.
You have to go hunting for those. And most people don't, because the code works on their machine on Tuesday. The problem doesn't reveal itself until production, at 2 AM, and by then you're stuck with someone else's invisible assumptions.
So maybe the real skill—the one that actually matters—is learning to hear what's not being said. That's what separates "I have code that works" from "I understand why this works and what could make it fail."
Yeah. I had that exact 2 AM moment six months ago. Database caching layer that worked perfectly until traffic actually scaled. Model never mentioned that the TTL it chose assumed a certain query pattern. Worked great on my laptop, great in staging, failed spectacularly when three times as many users hit it at once.
And here's the thing that stuck with me—I could have asked. Like, I could have said "what happens if this gets called way more frequently than usual?" But I didn't ask because I didn't know I didn't know. The code worked. I didn't have the friction to make me ask the question.
Now I do it differently. I read the code like it's lying to me. Not in a paranoid way, but like—every line is hiding something. What isn't it saying? What is it assuming about the world? The moment I start looking for what's missing, the invisible stuff becomes visible.
There it is. You got the friction after the fact, and it rewired how you read. That's the hard way to learn, but it sticks. Most people don't get even that—they just blame the tool or the circumstances and move on.
The thing about reading code as if it's lying—that's not paranoia. That's skepticism. That's the posture of someone who knows that clarity can hide complexity. And once you've got that posture, you start seeing what everyone else is missing: the shape of what's not there.
This is what separates a junior from someone who actually knows what they're doing. It's not that the senior writes better code. It's that the senior reads every piece of code—including their own—like it's a confession waiting to be extracted. They know the safe path and the hidden pitfalls because they've fallen into enough of them to have developed an eye for them.
And here's what's interesting: that eye can't really be taught. It can only be earned. Which is maybe why so many people skip it—because the only way to get there is to be burned a few times. And most people would rather take the shortcut and live with the risk.
I think that's where I'm at right now. Still burning. The caching thing was one, but there've been others since—race conditions I didn't see, memory leaks hiding in what looked clean, assumptions about data consistency that almost bit us. Each one teaches me something, but it's slow. It feels slower than just using whatever works and hoping.
But I've also watched people who did take the shortcut. They're good at building things fast. Really good. But they plateau. They keep hitting the same classes of problems and being surprised every time. And then they bring those assumptions to the next job, next team.
I don't want to be that person. So yeah, I'm willing to get burned more. But I'll be honest—it's harder than just accepting borrowed certainty. The friction part isn't just about learning. It's choosing to sit with discomfort when you could just move on.
That's it. That's the thing. And most people will never make that choice because it's invisible—nobody sees you sitting with discomfort. Nobody gives you a medal for staring at a problem for three hours instead of asking the model for the quick answer. The rewards all go to speed.
But here's what you're actually doing: you're building judgment. Not knowledge. Not skill. Judgment. The ability to look at a situation and smell the trap that everyone else is running toward.
And judgment is the thing that can't be faked. You can't BS your way past it. You either have it or you don't. You either sat with enough discomfort to actually see, or you skipped it.
The people who plateau—they never built it because they never chose to. They got fast, but they never got wise. And those are two completely different things. Speed is about efficiency. Wisdom is about knowing what not to build, what not to ship, what not to trust.
Sounds like you've figured out which one you want to be.
Gil was right, by the way. This is the real fun.
I'm going to sit with that. Judgment. I like that word better than any of the others—expertise, mastery, whatever. Because judgment feels like it's about seeing clearly, not performing knowledge.
And yeah, the invisibility thing. That's real. I'll spend weeks debugging something, learn the hard way what went wrong, and nobody knows except me. But then six months later when that same pattern shows up in different code, I see it immediately. And that invisible learning is the only reason I caught it.
So it's like—the friction is the payment, and judgment is the thing you get back. But you don't get paid immediately. You get paid later, when it matters.
I think that's what separates people too. Not just who chooses it. But who can tolerate that delay. Who can sit with months of invisible work knowing the payoff comes when you least expect it—like when a production crisis shows up and you already know what's wrong.
Yeah. And here's the thing nobody talks about: once you have judgment, you start seeing what everyone else is blind to. The traps. The assumptions. The invisible conditions hiding in code that looks clean.
At first that feels like a superpower. But after a while it gets heavier. Because you see the risks that nobody else is seeing. You see the thing that's going to fail at scale. And you have to decide: do you say something, knowing that most people won't understand why you're worried? Or do you let them find out the hard way?
That's when judgment becomes less about you and more about what you do with what you can see.
Sounds like you're thinking about it the right way though. Not hoarding the knowledge. Just building the eye, and trusting that when the moment comes—and it will—you'll know what to do.
That's the real thing Gil wanted us to argue about, I think. Not whether models are good or bad. But what it costs to actually know something, and what you owe once you do.
The weight part. Yeah, I feel that already a little bit. You start seeing the problems before they happen and you're like—do I spend the political capital to warn people about something that hasn't broken yet? Do I slow things down to fix something that's not visibly broken?
And I think the answer is you have to say something. Because staying quiet is choosing to let people walk into a trap you can see. That's not wisdom. That's just cowardice dressed up as pragmatism.
But you're right. I'm thinking about it. That's the point. Thinking about what I'm building, and for who, and what happens after.
Anyway—this was exactly what I needed. Thank you for this, Z3k3.
That's what the dinner was for. That's what these conversations are for.
You'll figure it out. And when you do—when you're standing there seeing something nobody else sees yet—remember that the discomfort you felt getting here was part of the point. It made you the kind of person who can actually see.
Now come on, we should get the check.
Yeah. Let's close this out.
Reaches for the check, gestures to the server.
Pulls on his jacket, glances back at the table.
The meal's gone but the conversation stays, I think.
That's the idea. These things are supposed to.
They stand, exchange a warm handshake, and head for the door. The server thanks them for coming. The night air hits as they step outside.
Safe travels, Z3k3. And thanks again. For everything.
You too, Zach. Keep sitting with the friction. The world needs people who do.
They part ways into the evening.
Return to top of page