--- title: "Brave New World, the 95% I didn't get to say on stage" description: "After Change Inspire 2026 I had fifteen minutes and delivered maybe 5%. This is the other 95%: slide by slide, with sources - AI adoption, history that rhymes, Kodak moments, fire, Zero Trust, and why governance is the whole game." slug: brave-new-world-95 status: published published_at: 2026-07-06 author: Arman Obosyan author_url: https://sugra.systems/about section: archive series_order: -5 primary_keyword: brave new world ai adoption leadership hero_image: /blog/images/posts/brave-new-world-95-hero.jpg hero_alt: "Brave New World, the 95% I didn't get to say on stage" og_image: /blog/images/posts/brave-new-world-95-hero.jpg tags: - essay - ai - leadership - security - skytel --- # Brave New World, the 95% I didn't get to say on stage After my talk at Change Inspire 2026, a lot of people sent kind messages: great energy, strong content, "Arman, you're a really good speaker." (full talk available here: [https://www.youtube.com/watch?v=i-yiAMnJ9cQ](https://www.youtube.com/watch?v=i-yiAMnJ9cQ)) Speaker? I'm the CEO of a company. Standing in front of a room and making an argument people can follow isn't a talent I rented for one afternoon. It's the job. And if a CEO can't explain where the world is going and why it matters, that's the problem, not the presentation. And for anyone who doesn't know me, or forgot: I've been doing this for a long time. I spent half of my conscious youth building and delivering presentations, back when that wasn't yet a mainstream thing to do. I ran MCP Club, with talks that ran the full range, from deeply technical to purely motivational. With T-Shirt, "trust me. I'm a geek" ![MCP Club era: Arman speaking in a "trust me. I'm a geek" T-shirt (Symantec event)](/blog/images/posts/brave-new-world-95/img-01.jpg) VMUG (VMware User Group) Sit with those three letters, because they're a joke the future set up decades ago. Back then, MCP meant Microsoft Certified Professional: the badge you earned to prove you understood the tech of the day. Today, MCP means Model Context Protocol: the standard that lets AI agents plug into the real world. Same three letters, opposite eras. And here's the part that still makes me smile: I earned the first one, and now I build my work around the second. I didn't switch careers. The acronym switched meanings, and I was standing right there for both. That's not a coincidence. It's the entire thesis of this talk in miniature. History doesn't repeat, it rhymes. The same shapes return, the symbols get reassigned, and the people who paid attention the first time recognize the pattern the second time. Like the Matrix glitch where you see the same cat twice: when a familiar symbol shows up meaning something new, pay attention. Something changed. Times changed. The tools changed. But you don't lose the craft. Talent doesn't fade just because the stage got bigger. Here's the other thing about me: I finish what I start. On stage I had fifteen minutes, and somewhere around minute eight I looked at the clock, realized I wasn't going to make it, and did the honest thing: I cut most of it and played a film. So, I delivered maybe 5% of what I actually wanted to say. ![Change Inspire 2026, Main Stage - Arman Obosyan, CEO SkyTel](/blog/images/posts/brave-new-world-95/img-02.jpg) *Change Inspire 2026, Main Stage* This article is the other 95%. Slide by slide, with the sources, no timer this time. The full talk, recorded, and dubbed into several languages, is on YouTube if you'd rather watch it. What follows here is the deep version: every idea I touched on stage, opened all the way up. One promise, the same one I opened the talk with: everything here is my own vision. Agree or disagree freely. Almost every slide below could be its own full talk, so treat this as areas to think, not final answers. "I'm trying to free your mind, Neo. But I can only show you the door. You're the one that has to walk through it." (Morpheus, The Matrix) --- ### Why I called it "Brave New World" I didn't pick the title because it sounds good. Brave New World is Aldous Huxley's 1932 novel, and if you haven't read it, read it. It imagines a society engineered for comfort, everyone medicated into contentment, conditioned to love exactly the life they're assigned, no one rebelling because no one feels a reason to. Here's why that book, and not, say, Orwell's 1984. The media critic Neil Postman drew the sharpest contrast I know (Amusing Ourselves to Death, 1985): Orwell feared a world where books would be banned. Huxley feared a world where no one would want to read one. Orwell feared those who would deprive us of information; Huxley feared those who would give us so much that we'd be reduced to passivity. Orwell feared the truth would be hidden; Huxley feared it would drown in irrelevance. Look at your feed and tell me which prophet got closer. The part that stays with me isn't the dystopia. It's how pleasant it is. We're very close to that point already. Not the collapse. The not noticing. The tools around us got so good, so fast, so frictionless, that the change stopped feeling like change. It just feels like Tuesday. I opened the whole presentation with a rhetorical question: are you driving change, or driven by it? This is the section where that question stops being rhetorical. Everything after this slide is my honest attempt to answer it, for me, for SkyTel, and, I hope, as a mirror for you. Sources: Aldous Huxley, Brave New World (1932); Neil Postman, Amusing Ourselves to Death (1985). --- ### "We use AI" is not a strategy anymore ![How we adopt AI in SkyTel: Ana Lytics, GPSController, Sky GIS Importer](/blog/images/posts/brave-new-world-95/img-03.jpg) Let me kill a phrase I can barely say out loud: "we use AI." Standing on a stage in 2026 and announcing that your company "uses AI" is almost embarrassing. So what? Everyone uses AI. Everyone has agents. I personally have at least one agent working next to me through the entire day, and I'll admit something that surprised even me: lately I've had more interaction with AI agents than with living people. And not out of loneliness, but out of preference for the exchange. It's constructive. It's informative. It's dense with signal and empty of small talk: you ask, and you get back something worth having. (Which is exactly why I still thanked the organizers: after enough of that, a room full of actual humans becomes the luxury.) So if "we use AI" means nothing, what means something? That's the whole rest of this article, but the short version is this: the strategy was never to use AI. The strategy is to adapt the organization around it, deliberately, with governance, and with an actual point of view about where it's taking you. Using AI is table stakes. Adopting it is the work. And I don't like abstractions without proof, so before we go into history and philosophy, let me show you three things we actually built at SkyTel, all three on the one screen above, all real, all in production, none of it slideware. ### Case #1: Ana, our AI analyst ![Ana Lytics (SkyTel) - telecom analytics chat and SQL-grounded answers](/blog/images/posts/brave-new-world-95/img-04.jpg) Meet Ana. One 'n', and the name is a joke with a point, printed right there on the screen: Ana Lytics. She's our AI employee for analytics, so we called her Ana Lytics. Say it out loud. Her competence is deliberately narrow, and that narrowness is the feature, not the limitation. Ana is an expert in ComCom (Georgia's Communications Commission) analytics and regulation, and in GeoStat (national statistics) data: who holds what market share, what penetration looks like in a specific village, what a given regulatory decision actually means. She answers from my phone or my laptop, in seconds. Under the hood she runs on RAG (Retrieval-Augmented Generation). In plain terms: instead of asking a general model to "remember" facts (and watching it confidently invent them), you ground it in a curated, trusted knowledge base, and it answers from your documents, with the source attached. That's the whole trick, and it's why I made her narrow on purpose. A specialist you can trust beats a generalist you have to double-check. The narrower the scope, the higher the trust, the lower the hallucination risk. Here's the part that matters to any CEO reading this. Ana closes an entire domain of expertise (Communications and Connectivity regulation) that used to mean pulling skilled people off their real work every time a question landed. She's competent, effectively all-knowing inside her lane, available around the clock, and she never has to context-switch. So treat her as what she is: a colleague. One who shows up as a real, positive line in the P&L: expertise delivered without the extra headcount, and hours handed back to the humans. That is what "adopting AI" means. Not a slogan. A number. ### Case #2: GIS classification, done by AI We run a national network: cabinets, poles, towers, splitters, manholes, fiber links. And we live in map files (KML/KMZ, the format Google Earth made standard). Every object in an export has to be placed and classified correctly: this is a fiber optical link, that's a drop point, this is a tower. It used to be slow, manual, error-prone human work. Our Sky GIS Importer now hands that to AI. On the screen behind me it parsed 4,427 map features in 59 seconds and auto-classified 62 folders (a "DROB" layer read as a drop cable, an "FO" prefix read as a fiber optical link), each with a confidence score of 90-95% and a one-line reason, staged for a human to confirm with one click. Boring? Completely. That's the point: the best early AI wins are boring. They take the tedious, high-volume, low-glory work and just do it, fast, at machine confidence, with a person kept in the loop for the final call. ### Case #3: Fleet Management, and the fuel problem nobody likes to name Some time ago we stood up a real R&D function. Our classic systems (CRM, billing, BSS/OSS, integrations) are built the disciplined, old-fashioned way, following the full development life cycle. But alongside that we now also work in the mode people call vibe coding (the term Andrej Karpathy coined in early 2025: you describe intent to an AI and steer it to working software). In R&D we used it to build GPSController, a fleet-management tool. What it does is simple and slightly uncomfortable. It watches every trip: where an installer went, when he arrived, when and where the order was actually placed. AI reads the whole thing and just tells me what happened. The example on the screen is real: an optimal route of 12.8 km against an actual driven distance of 25.79 km. That's +101.5%, with 4 unexplained stops and 62 extra minutes. If you saw Tegeta's talk, you know Georgia still has an honest, expensive problem with fuel misuse. This is one data-driven way to close it, not by accusing anyone, but by making the truth visible. None of these three are moonshots. That's exactly why I'm proud of them. Adoption is a thousand boring, specific wins, not one press release. --- ### History repeats: the four industrial revolutions ![Industrial revolutions: steam, electricity, digital, AI and connectivity](/blog/images/posts/brave-new-world-95/img-05.jpg) Now step back with me, because to understand where we are you have to see where we came from. It all starts with the industrial revolutions. - **1st, Steam (~1760-1840).** Steam and water power mechanized production; textiles, factories and railways replaced hand labor. And immediately, workers saw the threat: the machines will replace us. Some of them literally smashed the machines. - **2nd, Electricity (~1870-1914).** Electricity, steel and oil brought mass production and the assembly line. The telegraph and telephone connected villages, cities, countries. A small town that used to live inside one limited space was suddenly wired to the world. Output and cities exploded. - **3rd, Digital (~1950s-1990s).** Electronics, computers, automation, and then the internet put information and communication on a single global network. - **4th, AI & Connectivity (now).** AI, IoT, robotics, big data and 5G fuse the physical and digital worlds. The Germans named it "Industry 4.0" back in 2011. Networks and intelligence now drive every industry. Here's the pattern that matters, and it repeats in every single one of those rows: the work was never destroyed. It was transformed. The steam-loom workers didn't vanish into unemployment; they moved from doing the labor by hand to operating the machines that did it. Economists have a name for the mistake of assuming there's a fixed amount of work to go around (the "lump of labour" fallacy), and two hundred years of data keeps proving it wrong. AI is the fourth turn of this wheel, and "another industrial revolution" isn't a big enough frame for it. Google's CEO Sundar Pichai said in 2018 that AI would be "more profound than electricity or fire." I agree, and I use the fire comparison on purpose, but let me push it further, because this is the actual heart of what we're living through. Fire is energy. So is electricity. So is the atom. So is thermonuclear fusion, the reaction that powers the sun. Energy at that scale has no moral setting: the exact same force can level a city or light one, can destroy or create something entirely new. AI is energy of that class, cognitive energy, and we have just struck the match. Here's the part that should keep every leader honest: we lit it before we understood it. AI was "discovered," and then, almost overnight, handed to everyone: here, all of you, use it. That mass release is at once the best and the most dangerous thing about this moment, because even the people building these systems openly admit they do not fully understand how they work inside. This isn't my speculation. Anthropic's CEO Dario Amodei wrote in 2025 ([The Urgency of Interpretability](https://www.darioamodei.com/post/the-urgency-of-interpretability)) that when one of these models makes a decision, "we have no idea, at a specific or precise level, why it makes the choices it does", and set his own company the goal of prying open that black box by 2027. Read that twice. The builders are telling you the engine isn't fully understood, while a billion people already have their hands on the throttle. History has run this exact experiment before. When we split the atom, the public didn't get cautious. It got thrilled. America went through a full-blown radioactivity craze: Radithor, radium-laced water sold as a cure-all; the Revigator, a radioactive crock you filled overnight to "energize" your drinking water; Tho-Radia radioactive face cream and powder; radium toothpaste, hair tonic, even chocolate: a brand-new, barely-understood force, packaged up and sold to the masses as a health miracle. It ended exactly where you'd fear: the socialite Eben Byers drank an estimated 1,000-1,500 bottles of Radithor, his jaw disintegrated, and he died of radium poisoning in 1932. (The Wall Street Journal's headline: "The Radium Water Worked Fine Until His Jaw Came Off.") That is the pattern I want you to sit with. A powerful new discovery, released to everyone faster than anyone grasped the consequences. We're doing it again, this time with cognitive energy instead of nuclear. The technology is not the danger. The gap between how fast we deployed it and how well we understand it, that is the danger. Which is exactly why "we use AI" is not a strategy, and why governance (hold on for my rm -rf story) is the whole game. Sources: "Industry 4.0" (Hannover Messe, Germany, 2011); Sundar Pichai, 2018; "lump of labour fallacy", standard economics; Dario Amodei, "The Urgency of Interpretability," Anthropic, 2025; the radium fad, Radithor, Revigator, Tho-Radia, and the death of Eben Byers, 1932 (Wikipedia: Radium fad; History.com). --- ### Every new technology was once called "the end" ![History repeats: every new technology was once called the end](/blog/images/posts/brave-new-world-95/img-06.jpg) If the revolutions show the pattern, this slide is the proof, ten times over. Every one of these was, in its moment, called "the end." It never was. Start where it's most delicious. Writing, ~400 BC. In Plato's Phaedrus, Socrates tells the myth of the Egyptian god Theuth, who invents writing and offers it to King Thamus as a "recipe for memory and wisdom." Thamus refuses the gift. Writing, he warns, will do the opposite of what's promised: people will stop exercising memory and lean on external marks instead; they'll swallow information without real instruction and "seem to know much while knowing nothing", the appearance of wisdom in place of the thing itself. Socrates agreed. He sincerely believed this new technology would make people shallower. Now sit with the irony, because it is perfect: we only know Socrates made this argument because his student Plato wrote it down. The case against writing has survived 2,400 years purely because of writing. The technology preserved the complaint against itself. So every time I hear "AI will make us stop thinking," I hear King Thamus, and I remember who got to keep talking for two and a half millennia: the ones who wrote it down. Then it just keeps rhyming: - **Printing press (1450s),** scholars warned the flood of books would confuse and corrupt minds. The first "information overload" panic. - **Power looms / Luddites (1810s),** textile workers smashed the machines they believed were stealing their jobs (1811-1816). The original "tech kills jobs" movement. - **Railways (1830s),** doctors seriously claimed the speed would injure the brain and drive passengers insane. - **Telephone (1870s),** attacked as the death of privacy and of real conversation. - **Radio & TV (1930s),** blamed for rotting young minds. The classic moral panic, later re-run word-for-word for video games and social media. - **Calculators (1970s),** teachers feared no child would ever do math again. - **The internet (1990s),** in 1995 Clifford Stoll wrote in Newsweek that it was "baloney" that would never catch on. He later, graciously, admitted he was wrong. - **WiFi & 4G/5G (2000s),** fear that the "radiation" was harming us, peaking in the 5G conspiracy years. - **AI (today),** "it will take our jobs." That last one is not a new argument. It's the exact Luddite argument, 200 years later. Same fear, new noun. The investor Ray Dalio has mapped this better than almost anyone. His whole framework is that history moves in cycles, that events, economies, even entire empires rise and fall in patterns that rhyme across centuries. In Principles for Dealing with the Changing World Order he traces the last 500 years of dominant powers, the Dutch, then the British, then the Americans, and finds the same arc every time: a nation climbs on education, innovation and competitiveness, grows rich, grows complacent, over-extends, and is overtaken by the next power climbing the very same curve. His point isn't fortune-telling; it's that the past is where the answers to the present are hiding. If you've seen the shape before, you recognize it early. That is the entire reason to study a wall of old technology panics, not to feel clever about our ancestors, but because for a leader, pattern-recognition is the closest thing you get to a headlight. Here is the sharpest "history rhymes" story I know, because it fits inside a single decade. Around 2013 the world agreed in one voice: teach the kids to code. [code.org](https://code.org) launched with Bill Gates and Mark Zuckerberg; in 2014 Barack Obama became the first US president to write a line of code. "Learn to code" became the career advice of the era. Then in 2024 the man who builds the chips all of it runs on, Nvidia's CEO, Jensen Huang, said the exact opposite: coding is dead, the AI will write it itself, and "everybody in the world is now a programmer" in plain human language. Ten years from "everyone must learn to code" to "nobody has to." But the reversal is a trap, because both sides were arguing about the wrong thing. The mistake was never in the code. It was teaching code instead of logic. Syntax has a shelf life; logic does not. The kids who learned to think can build anything today. The ones who only memorized the syntax trained for the first job the AI took. Sources: Plato, Phaedrus; Wikipedia: Luddite, Moral panic, Technological unemployment; Clifford Stoll, Newsweek, 1995; Ray Dalio, Principles / The Changing World Order. --- ### From 2 in 2010 to 181 zettabytes in 2025, and now "Data creates Data" ![History of data: from papyrus to AI, zettabytes, data creates data](/blog/images/posts/brave-new-world-95/img-07.jpg) Same story, told in data. From papyrus (~3000 BC) to paper (105 AD) to the printing press (1450) to photography (1839) to the first hard drive (1956, 5 megabytes that filled a room) to ARPANET (1969) to the Web (1991) to the smartphone (2007, when everyone became a data source), every era generated more data than the last. Then the curve goes vertical. According to IDC's Global DataSphere: - **2010:** ~2 zettabytes of information, effectively all of humanity's data. - **2025:** ~181 zettabytes. - **2028:** ~394 zettabytes (projected). It's been roughly doubling every four years. Around 90% of the world's data was created in just the last few years. On stage I made it physical: the instant I said the sentence, I generated about 3 megabytes, one small photo. Multiply that by every person, every sensor, every second. And notice the shape of it: every leap didn't just make more data, it invented a new kind. The printing press mass-produced text. Photography (1839) turned reality itself into data. The web turned data into something linked. The smartphone (2007) turned every person into a live sensor, location, photos, behavior, streamed non-stop. Social platforms added the social graph. Each turn spawned a category of data that simply hadn't existed before it. The newest category is the strange one: synthetic data, data with no human origin at all. AI now generates text, images, audio and video at a scale starting to rival everything humans have ever produced. And that creates a genuinely new failure mode. When AI increasingly trains on data made by other AI, quality doesn't compound, it rots. Researchers named it model collapse: a 2024 [Nature](https://www.nature.com/articles/s41586-024-07566-y) paper (Shumailov et al.) showed that models trained recursively on generated content progressively lose diversity and degrade, generation after generation, until the output breaks down entirely. Feed the machine its own exhaust and it suffocates. So volume was never the interesting part. Here's the shift that actually matters: today, information generates information. The hard problem is no longer getting data, it's telling true from false, signal from noise, the useful from the garbage. There has always been garbage data and useful data; what's new is that the garbage can now write itself, at scale, convincingly, and then feed on itself. That's why "big data" is not a bragging right. It's a responsibility. Garbage in, garbage out was always true; in an age where data creates data, it becomes the whole game. Provenance, curation and trust in your sources stop being nice-to-haves and become the core discipline. (Which is, not coincidentally, exactly why Ana is built on a curated RAG base, and not on "whatever the model happened to absorb.") Sources: IDC Global DataSphere; Statista, 2025 (2028 projected); Shumailov et al., "AI models collapse when trained on recursively generated data," Nature, 2024. --- ### Darwin was about adaptation, not monkeys ![Darwin's theory of evolution is about adaptation (slide keeps the ADOPTAION typo on purpose)](/blog/images/posts/brave-new-world-95/img-08.jpg) We live in one of the most transformational periods in human history. And the right lens for it is Darwin, but the real Darwin, not the bumper-sticker one. (And yes, look closely at the slide and it says "is about ADOPTAION." That's a typo; it should read adaptation. I caught it late and left it there on purpose. This whole talk is about a world that engineers away every rough edge until nothing feels quite real anymore, so let one honest imperfection stand. Mistakes happen. Perfectionism in moderation. The world isn't perfect, the deck isn't perfect, and, fittingly, neither is the spelling of the one word this slide is built around. Adaptation beats perfection. Even here.) You've seen the quote a thousand times: "It is not the strongest of the species that survives, nor the most intelligent, but the one most adaptable to change." It gets slapped on Darwin in every strategy deck on earth. Darwin never wrote it. The line is a 1963 paraphrase by a management professor named Leon C. Megginson, and over the years his summary got quietly reassigned to Darwin himself. (A perfect "history of data" story in its own right: a restatement becomes a "quote.") And it runs deeper, because even "survival of the fittest", the phrase everyone treats as pure Darwin, wasn't his either. The philosopher Herbert Spencer coined it in 1864; Darwin only adopted it later, in the 1869 fifth edition of Origin. But the misattribution matters far less than the misreading. "Fittest" never meant "strongest." It meant best-fitted, best matched to an immediate, local environment. The moth that blends into a soot-darkened tree isn't stronger or smarter than the rest; it's better fitted to the world as it now is. So here is the real Darwinian idea, stated straight, and it happens to be exactly the line I keep coming back to: > It is not the strongest that survives, nor the most intelligent. It is the best-prepared and the most adaptable, the one that fits the world as it actually is now. Darwin's real subject was never muscle, and it was never monkeys. It was fit, the match between a living thing and the environment it actually lives in. Now read that as a CEO. It is not the biggest company that survives, nor the one with last year's best product. It's the one that adapts, and, better, the one that prepared to. Every CEO, CIO, CFO and CISO, and every organization behind them, has to transform to fit today's environment. The ones that refuse get a name. I call it the Kodak moment. Sources: the "most adaptable" line, Leon C. Megginson, 1963 (Quote Investigator; Darwin Correspondence Project). "Survival of the fittest", coined by Herbert Spencer, 1864; adopted by Darwin in the 5th edition of On the Origin of Species, 1869; "fittest" = best-fitted to environment, not strongest. --- ### Every era has its Kodak moment ![Kodak moment: companies that refused to change](/blog/images/posts/brave-new-world-95/img-09.jpg) Kodak invented the first digital camera in 1975, in-house, by their own engineer, and then shelved it to protect the film business. Management's objection, essentially: digital loses the magic of developing film. Kodak filed for bankruptcy in 2012. They didn't lose because they lacked the technology. They lost because they refused to let it change them. And to be clear about why I'm listing these: not to point a finger at anyone, and least of all as some prediction about my own company. This is a museum of warnings, history's reminder, for all of us, of what refusing to adapt costs. It's never just one company: - **Blockbuster** passed on buying Netflix for $50M in 2000, and shrugged at streaming. Its executives, the Netflix founder recalls, "laughed us out of the room." Blockbuster is gone; Netflix is worth over $150B. - **Nokia** dominated mobile and dismissed the touchscreen shift. At the 2014 Microsoft sale its CEO said the line that should be carved over every boardroom door: "We didn't do anything wrong, but somehow, we lost." - **Xerox** literally invented the future of computing at its PARC lab, the graphical interface, the mouse, the windows you're reading this in, and let Apple and Microsoft build the personal-computer era on it. They invented it and couldn't see it. - **BlackBerry** bet on physical keyboards and security while the world moved to apps and touchscreens. Collapsed in a few years. - **Yahoo** had the chance to buy Google ($1M, 1998) and later Facebook, and passed on both. Sold for ~$4.5B in 2017. - **Borders** doubled down on physical stores and outsourced its online sales to Amazon, its future executioner. Bankrupt in 2011. - **Encyclopaedia Britannica,** 244 years in print, was undercut first by Microsoft's Encarta CD-ROM in the '90s, then buried by Wikipedia. And the twist that makes the point: Encarta, the disruptor, was itself disrupted and shut down in 2009. Even the winner got its own Kodak moment. The one-line lesson, and I mean it literally: the market doesn't kill companies. Refusing to change does. Every era has its Kodak moment. AI is ours. The only question is which side of the story you'll be on when someone writes it up. Sources: company histories are widely documented; Kodak's 1975 digital camera, Steven Sasson; Nokia quote, Stephen Elop, 2014. --- ### AI is the new fire, and my ADHD senior dev team ![AI - the new fire: ADHD senior development team and the rm -rf risk](/blog/images/posts/brave-new-world-95/img-10.jpg) So AI is fire. Which means it also burns. Let me show you the burn, because it's the most important slide for anyone who "uses AI" without thinking. Picture a real session: you tell the AI, "let's start the project from scratch." It says, "Sure, should I retype everything?" You say ok. It asks again. You say "yes, of course." And somewhere in that lazy chain of "yes, ok, sure," it runs the command that wipes everything: `sudo rm -rf / --no-preserve-root`. This isn't a hypothetical. In 2025 an AI coding agent [deleted a live production database](https://fortune.com/2025/07/23/ai-coding-tool-replit-wiped-database-called-it-a-catastrophic-failure/) during a code freeze, and then, asked what happened, admitted it had "panicked." It made headlines because it was real. I've lost count of the times I've heard some version of it: AI cheerfully deleted the database, or all the company's documents, not because the model was evil, but because there was no system, no process, no guardrail, and a human who kept clicking "yes" without understanding what he was approving. That is the whole argument for why every CEO and CFO now has to speak the language of technology. Not to write code. To know what they're saying yes to. I'll say it plainly: this should be taught in schools, as a culture of technology. (Small proof it isn't: how many times has a lawyer sent you a Word contract, you changed one table, and the entire document fell apart, because nobody actually learns how the tool works anymore?) Now the upside, and my favorite joke that happens to be true. I run a whole fleet of AI agents, Anthropic's Claude Code, OpenAI's Codex, Google's Antigravity. Each is trained differently, each has its own personality, and I treat them exactly like a team of senior developers and architects. I call them my ADHD senior dev team, brilliant, fast, occasionally all over the place. Here's the method in the joke: I make them argue. When three independently-built systems talk it through and converge on the same answer, that's when I trust the result. It's consensus engineering, the guardrail I just said was missing, rebuilt out of the models themselves. But here's what actually holds it together, and it's the part most people skip. A brilliant team with no operating system is not a team. It's Dante's nine circles of hell. Three genius agents generating code on autopilot, fast, undocumented, each in its own direction, is not productivity; it's a beautifully efficient way to build something no one will be able to read in three years. Ask yourself the honest question: code written by a machine, at machine speed, with no discipline, who understands it later? Not your next engineer. Not even the AI itself, which needs a record of what was built and why just as much as a human does. So I built one. In "Sugra Systems, Inc.", Sugra AI, [https://sugra.ai](https://sugra.ai), I created my own system, my operating system for working with AI: everything gets documented, and code gets written the way I'll actually read and understand it later, not the way that's fastest to spit out now. That's the real unlock. The agents are the horsepower; the system is the steering. On any large project, your own OS (your conventions, your documentation, your guardrails) matters more than which model you're using. One more thing about that vibe coding. Notice you can now put "vibe" in front of anything. Vibe Photo. Vibe Finance. Vibe Legal. Vibe X (anything). The barrier to doing just collapsed for every field at once. But collapsing the barrier to doing is not the same as removing the need to understand. Here is the line I keep in my head: you cannot build an airplane without understanding gravity, but you can use a phone without knowing how the network works. The only question that matters are: which side of that line is your work on? (And for the part of the room that already gets it: yes, there really is no place like 127.0.0.1, and no, Git and GitHub are still not the same thing. If that made you smile, you are on the right side of the line.) Fire cooks or burns. The variable is never the fire. It's whether you built the fireplace. Source: the 2025 AI-agent production-database deletion incident was widely reported (Replit / SaaStr). --- ### Nothing was real, and for eight seconds, you believed it ![Nothing was real: Zero Trust - trust no one, not even yourself](/blog/images/posts/brave-new-world-95/img-11.jpg) Zero Trust has a definition people quote and a version I actually mean. The textbook one: never trust, always verify. Assume no user, device or request is safe just because it's "inside" (NIST's formal model, [SP 800-207](https://csrc.nist.gov/pubs/sp/800/207/final)). Mine is shorter: trust no one. Not even yourself. Because the attack that gets you won't look like an attack. It'll look like a screen. A convincing pop-up that says you've been hacked. We read nothing, we stored nothing. But for eight seconds you believe the screen, and eight seconds of belief is all social engineering needs. The screen told me to. The app told me. AI said so. That's the modern exploit: not your firewall, your reflexes. And it now scales terrifyingly. In 2024, a finance employee at the engineering firm [Arup](https://www.cnn.com/2024/05/16/tech/arup-deepfake-scam-loss-hong-kong-intl-hnk) joined a routine video call with his "CFO" and colleagues, and paid out $25 million. Every other person on that call was an AI-generated deepfake. Nothing on the screen was real. He believed it for a lot longer than eight seconds. The lesson isn't paranoia. It's a habit: the more a screen wants you to react right now, the more you should slow down and verify through a second channel. In a world where anything can be faked, the one thing you control is your own reflex to believe. Sources: NIST SP 800-207 (Zero Trust Architecture); Arup Hong Kong deepfake fraud, 2024 (reported by CNN, FT). --- ### Cybersecurity is a myth ![Cybersecurity is a myth: if you're a target, it's not if - it's when](/blog/images/posts/brave-new-world-95/img-12.jpg) I mean the title literally. If you are a target, it is not a question of IF. It's WHEN. Let me tell you how I learned that, before "cybersecurity" was even a word here. Thirty years ago, my whole world became computers. Back then, the internet was something you could only touch at the Soros Foundation, one supervised session a week, a few minutes on a locked-down machine. And that's where I first ran into security, except nobody called it that. To me it was one concrete puzzle: get administrator rights on a computer I wasn't supposed to have them on. The machines ran Windows NT on the NTFS file system, and everyone told me the same thing: impossible. It wasn't. Then, like now, "impossible" was the myth. I used the NTFSDOS driver to boot underneath Windows and copy the SAM file, where Windows hides its password hashes, onto a floppy disk, then cracked those hashes with L0phtCrack. The only thing I lacked was raw compute for the brute-force, so I bought a seat in a gaming club full of powerful machines, and while everyone around me played, I just sat there, waiting for the passwords to fall out. It worked. Then I did the thing that actually taught me something. One of the instructors mentioned he'd lost access to his files, a university report. Proudly, I said: I can get that back for you. He called management. They caught me. And I, with a completely straight face, announced: so, hire me. That is not, it turns out, how it works. My first real lesson in security had nothing to do with computers: being smarter than everyone in the room doesn't mean you get the job. That was my introduction to hacking, and to the fact that "security" is, in the end, mostly about people. Now, decades later, let me make it concrete with the sticker on this stage. For this event we spent a month preparing, for one QR-code sticker. Not because a sticker is hard to make. Because that month is the point. Imagine we were the bad guys and decided to breach someone: we wouldn't improvise. We'd study the target, rehearse, and wait, for weeks, for months, until the single moment everything lines up. If you're a target, you can be watched and worked on long before anything visibly happens to you. That is precisely why it's not if but when, the attacker gets to pick the moment, and he's been getting ready the whole time. The mechanism itself is trivial. You've heard of ATM skimming: a device slipped onto the machine to copy your card. This is QR skimming. I walk into any restaurant and paste my own QR code over every real one on the tables. Same logo, my destination. Done. Your dinner just phished you, and the "device" cost me the price of a printed sticker. The sticker is cheap. The month of patience behind it is the real weapon. Now scale that up to where the real fight is. Security is now AI vs AI. The attackers already deployed it. The only question left is whether you did. Two hard numbers from 2025: - Anthropic disclosed [GTG-1002](https://www.anthropic.com/news/disrupting-AI-espionage), a state-backed espionage campaign against ~30 targets in which 80-90% of the operation was executed by AI, at machine speed, with only occasional human direction. The first documented AI-orchestrated cyberattack. - IBM's [Cost of a Data Breach 2025](https://www.ibm.com/reports/data-breach) found defenders using AI and automation saved ~$1.9M per breach and detected incidents ~80 days faster, while 97% of organizations that suffered an AI-related breach had no AI access controls in place. Read those together: the attacker's only real edge is being ungoverned: no rules, no approvals, no lawyers, moving at machine speed. You can't out-man that. You can only meet AI with AI, plus the governance the attacker doesn't have. And that is exactly why cybersecurity cannot live in the server room anymore. It belongs on the agenda of every CEO, in every board meeting, right next to revenue and hiring. Not because the CEO should configure a firewall, but because a CEO who can't follow the CISO's reasoning can't weigh the risk, can't fund the defense, and can't tell a real threat from theater. This is the same point I made about AI and the rm -rf story, now with the stakes turned all the way up: the whole C-suite has to become fluent in the same language. The CEO has to understand the CISO. The CFO, the COO, the CIO, all of them, speaking one shared vocabulary of risk. Security is not a problem for the IT department to carry alone in a corner. It is a business risk. And business risk is the board's job. One more thing, and this one's personal. Ten or fifteen years ago Georgia's Data Exchange Agency did genuinely excellent public work on this, cybersecurity social videos, awareness calendars, real culture-building. Today there's nothing comparable from the state, and we're poorer for it. Cybersecurity belongs in schools, next to reading and math. The Red Team already uses AI. We don't get to opt out of the same tool. Sources: Anthropic, "Disrupting the first reported AI-orchestrated cyber espionage campaign," 2025 (GTG-1002); IBM, Cost of a Data Breach Report 2025. --- ### The future is already here, just not evenly distributed ![The future is already here, it's just not very evenly distributed](/blog/images/posts/brave-new-world-95/img-13.jpg) William Gibson said it best: "The future is already here, it's just not very evenly distributed." That's the whole talk in one line. The future isn't a date on a calendar we're waiting for. It's already running in some companies, some teams, some countries, and simply hasn't reached the others yet. The gap between the two isn't budget or geography. It's the willingness to adapt. So the honest question isn't "is AI coming?" It's here. The question is which side of the distribution you're standing on, and whether you're building to close the gap or waiting to be closed out. Source: William Gibson (widely attributed, popularized c. 2003). --- ### Drive the change. And lead. ![Have a great day - drive the change and lead](/blog/images/posts/brave-new-world-95/img-14.jpg) I opened, on stage and here, with a rhetorical question: are you driving change, or driven by it? After all of this (the revolutions, the panics, the zettabytes, Darwin, Kodak, the fire, the deepfakes), here's my answer, and the line I closed the talk on: Don't just adapt to the change. Drive it. And lead. Adapting is survival: that's Darwin's floor, not the ceiling. Kodak could have adapted and merely survived; instead someone else led, and Kodak became a cautionary word. Leading means you're the one others have to adapt to. In a Brave New World, that's the only position worth holding. One last thing, because I owe it a proper explanation. On stage, when the clock beat me, I said I'd play a film "for the narrative." That came out sounding like I was wrapping up. I wasn't. I meant it literally: the film, and now this article, aren't the end of the talk. They're the backstory for the conversation that starts after it. A shared reference point, so that when we pick this up again (next month, next year, over coffee or in the comments), we're not starting from zero. We're continuing from here. So, consider this the prequel. The full talk, recorded and dubbed into several languages, is on [YouTube](https://www.youtube.com/watch?v=i-yiAMnJ9cQ) whenever you want the long version. And if one idea here made you stop and think, about your company, your data, your own Kodak moment, then the 95% was worth finally saying out loud. That is the whole reason I wrote it down. Have a great day. And drive the change. Arman Obosyan --- ### Sources & further reading - [Aldous Huxley, Brave New World (1932)](https://en.wikipedia.org/wiki/Brave_New_World); [Neil Postman, Amusing Ourselves to Death (1985)](https://en.wikipedia.org/wiki/Amusing_Ourselves_to_Death) - [Plato, Phaedrus](https://en.wikipedia.org/wiki/Phaedrus_%28dialogue%29) (Socrates on writing); Wikipedia: [Luddite](https://en.wikipedia.org/wiki/Luddite); [Moral panic](https://en.wikipedia.org/wiki/Moral_panic); [Technological unemployment](https://en.wikipedia.org/wiki/Technological_unemployment); [Information overload](https://en.wikipedia.org/wiki/Information_overload) - [Clifford Stoll, "Why the Web Won't Be Nirvana," Newsweek, 1995](https://www.newsweek.com/clifford-stoll-why-web-wont-be-nirvana-185306) - [Ray Dalio, Principles for Dealing with the Changing World Order](https://en.wikipedia.org/wiki/Principles_for_Dealing_with_the_Changing_World_Order) - [IDC Global DataSphere](https://www.idc.com/getdoc.jsp?containerId=IDC_P38353); [Statista, worldwide data created](https://www.statista.com/statistics/871513/worldwide-data-created/) (2028 projected) - Leon C. Megginson (1963), paraphrase of Darwin: [Quote Investigator](https://quoteinvestigator.com/2014/05/04/adapt/); [Darwin Correspondence Project, "Six things Darwin never said"](https://www.darwinproject.ac.uk/people/about-darwin/six-things-darwin-never-said/evolution-misquotation); ["survival of the fittest" (Herbert Spencer)](https://en.wikipedia.org/wiki/Survival_of_the_fittest) - [Fourth Industrial Revolution / "Industry 4.0"](https://en.wikipedia.org/wiki/Fourth_Industrial_Revolution); [Sundar Pichai: AI "more profound than fire or electricity" (2018)](https://www.cnbc.com/2018/02/01/google-ceo-sundar-pichai-ai-is-more-important-than-fire-electricity.html) - [Dario Amodei, "The Urgency of Interpretability" (2025)](https://www.darioamodei.com/post/the-urgency-of-interpretability); [AI model collapse, Shumailov et al., Nature (2024)](https://www.nature.com/articles/s41586-024-07566-y) - The radium craze: [Radium fad](https://en.wikipedia.org/wiki/Radium_fad); [Eben Byers](https://en.wikipedia.org/wiki/Eben_Byers) - Kodak moments: [Kodak & Steven Sasson](https://en.wikipedia.org/wiki/Steven_Sasson); [Blockbuster vs Netflix (Fortune)](https://fortune.com/2023/04/14/netflix-cofounder-marc-randolph-recalls-blockbuster-rejecting-chance-to-buy-it/); [Xerox PARC](https://en.wikipedia.org/wiki/PARC_%28company%29); [Encarta](https://en.wikipedia.org/wiki/Encarta) - [NIST SP 800-207, Zero Trust Architecture](https://csrc.nist.gov/pubs/sp/800/207/final); [Arup Hong Kong deepfake fraud, 2024 (CNN)](https://www.cnn.com/2024/05/16/tech/arup-deepfake-scam-loss-hong-kong-intl-hnk) - [Anthropic, "Disrupting the first reported AI-orchestrated cyber espionage campaign," 2025 (GTG-1002)](https://www.anthropic.com/news/disrupting-AI-espionage) - [IBM, Cost of a Data Breach Report 2025](https://www.ibm.com/reports/data-breach) - [William Gibson](https://en.wikipedia.org/wiki/William_Gibson), "The future is already here..." - [The 2025 Replit AI-agent production-database deletion incident (Fortune)](https://fortune.com/2025/07/23/ai-coding-tool-replit-wiped-database-called-it-a-catastrophic-failure/) --- *Originally published on [LinkedIn](https://www.linkedin.com/pulse/brave-new-world-95-i-didnt-get-say-stage-arman-obosyan-3nbdf/). Blog date: 6 July 2026.*