Operating at the Edge: What Extreme Environments Teach Us About AI Systems

You are shipping faster than you have ever shipped before, and something feels off. Not broken, not wrong, just off. Like the floor shifted slightly and nobody mentioned it. I'm going to name that feeling today, and I'm going to offer you a framework for understanding it. Not from software, not from productivity theory, but from the most unforgiving operating environments humans have ever designed for. My name is Michal Zizo. I'm an earth and space architect and the founder of the Zizo, a creative disruption platform that works at the intersection of design, systems thinking, and the future of human performance. I'm a licensed architect. I studied at the Technion and the Politecnico di Milano. I did my space studies at the International Space University, and right now I'm serving as space architecture lead for Project Orbital Hospital, designing healthcare environments for humans in orbit by OSMED. I've had the privilege of presenting at NASA, the World Design Organization, YPO, the Dubai Future Foundation. I'm a two-time TEDx speaker, and the thread through it all, every project, every stage, is this one question. What does a human need from their environment in order to perform at their best? That question, which I've spent my career asking about buildings and spacecraft and extreme habitats, is the same question I want to ask with all of you today about the environments you all work at.

Now, this is what we're going to talk about. Five stops. Let's go. The 2024 and 2025 Stack Overview Developer Survey revealed a growing paradox. While AI tool adoption is surging, developer trust is plummeting due to frustrations with code quality and a desire for deeper comprehension. In survey after survey of software engineers working with AI tools today, a pattern keeps emerging. Faster outputs. Every developer reports that first, of course. Then, less clarity. More output, but less understanding of what's actually being produced. Then, more responsibility, which is striking because the whole promise of these tools is that they carry more of the load. And then the one that stops me every time. More anxiety around mistakes. Not less. More. Even as speed increases, the fear of something going wrong gets bigger. Does that match your experience? The dominant story being told right now is you just need to learn to use the tools better.

Better prompting, better workflows, better integration. And maybe that's partly true. But that story doesn't explain why it feels like this. Because this isn't a skill gap. This is an environmental shift, and those require a completely different response.

Tim Urban drew this. If you've read Wait But Why, you would know him. This is the exponential curve. And the critical insight isn't the curve itself. It's the gap. The gap between where you think you'll end up in the future and where you actually end up. We are all living inside that gap right now.

The gap isn't just about technology. It's about the conditions of work. The environment. Which brings me to the great misdiagnosis. We are treating AI as a productivity upgrade. But what it actually is is an environmental change. And that distinction matters enormously. Because a productivity upgrade, a faster framework, a better library, a faster pipeline, those require you to adapt skills.

But it ends up being a skill gap. A better library, a faster pipeline, those require you to adapt skills. But an environmental change requires you to adapt how you operate. I want to borrow something from my field of space architecture. Architects who design for extreme environments, Antarctic research stations, deep sea habitats, space crafts. They don't ask, why are the people inside struggling? Or why are they performing poorly? They ask, what kind of environment have I put them in? Space architects don't optimize for speed. They optimize for human sustainability inside that environment. The question is never how fast can this go. It's can a human being actually function safely inside the system, make decisions clearly, and not break down over time. That framework or reframe from examining the person to examining the environment is exactly what's missing from how we're talking today about AI and engineering teams. This doesn't feel hard because you're bad at your job. It feels hard because the environment changed. Instead of fixing the human, let's fix the environment. And I'll say it again, instead of fixing the human, let's fix the environment.

As an architect, when I say environment, I mean something very specific, not the tools, not the software. The condition a human being is placed inside. I know that in your world, environment has a different connotation. Production environment, runtime environment, local environment. And that's actually closer than you would think to what I mean. It's the full set of conditions that determine what a system can and cannot do.

An environment, in my definition, is the full set of design conditions that determine how humans perceive, decide, and act within a system. Now five properties shape it. One, visibility. What can the people inside actually see? What is legible and what is hidden? Speed. How fast does the environment move relative to human processing speed? Is the pace set by the humans or by the system? Number three is feedback proximity. How close in time is the consequence to the action? Does the environment tell you quickly when something is wrong? Number four, tolerance for error. How much margin does the environment allow before a mistake becomes critical, catastrophic? Is there redundancy or is every action load-bearing?

The next one is clarity of agency. Who does what? Decided where? Communicated how? Is the decision architecture legible or just implicit? These are all design properties. An architect made a choice, consciously or not, about each single one. Now I'm going to ask you to apply each of those to your own work. You are reading, reviewing, and shipping code you didn't really write. Code that is coherent, confident, and not fully traceable. Your visibility into the system has changed. Something is always happening below the line of sight.

Now in terms of speed, models update, patterns shift. The best practice from six months ago may be irrelevant today. The environment moves faster than any individual can track and it is not allowing to slow down. It doesn't allow you to catch up. For many teams, production is still the real feedback loop. By the time you know something is wrong, the AI-generated error has been compounded three layers deep. The distance between action and consequence has quietly grown. AI doesn't make surgical isolated mistakes. It makes systemic ones, confident, patterned, coherent, all the way down. And responsibility is still entirely yours. It was never transferred but the environment was redesigned around you as if it was. And that gap between where accountability lives and where legibility lives is where most things go wrong.

AI changed all five of these things simultaneously and we are still behaving as if nothing fundamental shifted. Well, in extreme environments, you don't just build differently, you operate differently. Let me tell you a little bit about extreme environments and see what we can learn from them. They are defined by the same exact five conditions. Visibility. The system is simply too large and too complex for any one person to hold. On the International Space Station, it's 14 countries worth of independent systems surrounding you at all times. No single operator sees the whole picture. You navigate by design clarity, instrumentations, protocol, role boundaries.

In terms of speed in a spacewalk outside the station, conditions outside shift in seconds. Temperature swings of 270 degrees Celsius as you pass in and out of sunlight every 90 minutes. The environment doesn't wait for you to catch up and adapt. Between Earth and the Moon, there is a three-second communication delay. Between Earth and Mars, there is a 20-minute each way, 40-minute round trip. By the time you know something is wrong, the window to respond in space may already be closing. The Columbia disaster in 2003 was initiated by a piece of foam insulation roughly the size of a briefcase. That singular anomaly, dismissed in pre-launch review, cascaded into the loss of a human life. That singular anomaly, dismissed in pre-launch review, cascaded into the loss of seven lives.

Small errors in complex interdependent systems don't stay small. For all the automation in modern spaceflight, a human being still holds the mission. The crew signs off. The flight director signs off. Accountability does not transfer to the system. It never does. What changes is whether the environment makes that accountability legible.

Sounds familiar now? That is exactly the AI environment you are operating in. The people who've thought most rigorously about operating in these conditions are not in software. They are in aerospace. They are on submarines. They are on Antarctic expeditions. They are in intensive care units. Extreme environments require you to prioritize operability over optimization. They prioritize clear responsibility over speed.

When something goes wrong, and it will, it always does, who owns what needs to be legible before the crisis, not during it. Extreme environments treat human sustainability as a system requirement, not a nice to have, and non-negotiable constraint on the whole architecture. When a human doesn't perform how you expect them to, you don't fix the human. You redesign the operating system around them. Space is extreme, volatile, chaotic. It does not pause while you catch up. It does not forgive a missed detail. Your systems have the same properties.

This is not a metaphor. Space is extreme, volatile, chaotic. It does not pause while you catch up. It does not forgive a missed detail. It is beautiful, but it does not care that you're moving fast. And here's the thing. Your systems have the same properties. Unpredictable, interdependent, unforgiving.

The chaos is not coming. It's not a phase that you're temporarily in while AI matures. It is the environment. The question is not how to make it less chaotic. The question is how to design clearly enough that humans can operate inside that chaos safely, sustainably, and with precision. And that's what space taught me.

Now, astronauts are not superhuman. They are humans in a super-designed environment. Let that sink in. If you feel like you're struggling right now with clarity, with accountability, with the sheer cognitive weight of working in an AI environment, it is not because you're behind. It is because the environment you're in was not designed for what it's now asking of you. And that is a design problem, which means it has a design solution. I got you.

Instead of asking, how do we move faster with AI? Let's ask, what kind of an environment are we creating? And can humans safely operate inside it? That question reframes everything. It reframes leadership. In space, the flight director and the capcom, they don't manage tasks. They run a mission. They know exactly who is doing what, in what order, with what authority, and under what conditions. Every communication chain is designed. Every decision pathway is legible before the crisis begins, not negotiated during it. That same clarity is what the AI era demands from engineering leaders.

That same clarity is what the AI era demands from engineering leaders. Not are my people keeping up with the tools, but have I designed the environment with enough clarity that they can make good decisions under pressure? It reframes responsibility. Instead of who wrote this code, ask, who owns this decision? Because in this new environment you're operating in, the author and the decision maker are no longer the same person. Extreme environments solved this decades ago. They call it command authority.

It also reframes velocity. Instead of, how do we ship faster? How do we move faster? Ask, what speed can we sustain without accumulating invisible risk? The International Space Station doesn't go as fast as it could. It goes as fast as it safely can with humans aboard. In mission control they say, slow is smooth. Smooth is fast. That is not a productivity hack. That is a mission design principle. Slow is smooth. Smooth is fast.

It also reframes how you operate. And I call it a co-pilot model. AI is extraordinarily good at a specific set of things. Scale, speed, pattern recognition, prediction, optimization, repetition, data compression, search, calculation. Humans, on the other hand, are extraordinarily good at a different set of skills. Emotional intelligence, moral reasoning, imagination, cross-context thinking, creativity, ingenuity, innovation, curiosity, vision. These are not competing lists. They are complementary operating modes. The question is not, will AI replace me or replace engineers? The question is, have you designed your environment so that AI handles what is predictable?

Let AI handle the computation. Your value is everything it cannot replicate. That is what mission control does. Autonomous systems handle the mechanics of flight. Humans hold the judgment calls. The relationship to the mission, the anomaly that doesn't match any prior system or pattern.

And it also reframes failure. Instead of how do we avoid failure, ask, how do we make failure survivable? Every extreme environment operates on the assumption that things will go wrong. They always do. The design question is, when they do, does the error stay local or does it cascade? Do we have the visibility to catch it? The protocol to respond?

In space, anomalies are not failures. They are data. I want to leave you with one image. Imagine an astronaut. Suited up. Preparing for a spacewalk. They're not exceptional because they're fearless. They're not operating well because they're smarter than everyone else. They perform under extreme conditions because someone thought very carefully about every layer of their environment. The suit. The checklist. The communication protocol. The rest cycle. The role clarity. Their board criteria. Every element designed in service of their ability to function. You are building in an extreme environment. You're working there. You're not broken. You're not behind. You are operating in conditions that have never existed before.

You are operating in conditions that have never existed before. And the first step is the same as it always is in space architecture. See the environment clearly. Because once you see it clearly, once you stop treating it like a temporary disruption and start treating it like the actual terrain, you can design for it.

Astronauts are not superhuman. Remember, they're humans in super designed environments. You are not managing tasks. You are running a mission.

With that, I will wish you Ad Astra to the stars. If you want to dive deeper, get your organization mission ready or just connect, please scan this QR code and feel free to reach out. Thank you so much.