YouTube Digest
Listen to this issue
English
Based on “AI is way Underhyped. This Silicon Valley CEO proves it | Relay.app, Jacob Bank” from EO Watch the original video
Beyond the Hype: Why AI Agents Are Your Next Career Superpower
What if you could assemble a high-performing team of marketing experts, sales coaches, and strategic analysts for less than the cost of your monthly coffee habit? According to Jacob Bank, founder and CEO of Relay.app and a former Google product lead, this isn’t a futuristic fantasy, but our current reality – and we’re barely scratching the surface of its potential. Bank emphatically states, “AI is way underhyped,” and he’s living proof, running his company’s entire marketing function with a team of 40 AI agents, while his personal AI bill clocks in at a mere $500 a month, a stark contrast to the $50,000 monthly cost of a human team.
Bank’s journey reveals a profound shift in how professionals must approach work, learning, and career growth in the age of artificial intelligence. He argues that building and leveraging AI agents is not just an advantage, but “the fundamental skill that will define every professional’s career for the next 30 years. It’s a requirement.”
More Than Just an Intern: The True Power of AI
Just a year ago, Bank himself advocated for thinking of AI as an “intern” – a helpful tool for time-saving, repetitive tasks. He’s since dramatically changed his mind. “That was the wrong mental model,” he admits. While AI certainly excels at intern-level duties, its capabilities extend far beyond.
Bank illustrates this with a compelling personal anecdote: “I am not good at sales calls. I do not have a background in sales. It does not come naturally.” Yet, he now has an AI coach that reviews transcripts of his meetings, offering sophisticated feedback: “Oh, you could have talked about this better. You could have articulated this value or you were too eager to jump into a demo there. You should have asked some more discovery questions.” A human sales coach of this caliber could cost upwards of $10,000 a month for just one meeting a week. Bank’s AI equivalent? “Like five bucks a week to run.”
This isn’t about mere efficiency; it’s about augmenting human capability at a strategic level. AI can perform complex strategy analyses, competitive assessments, and high-quality content creation tasks that no intern could ever manage. It’s about providing personalized, expert-level coaching and insight across any domain, democratizing access to specialized knowledge and skill development.
The Dawn of the “Super IC”
This paradigm shift isn’t just for CEOs; it’s for everyone. Bank predicts a future where traditional job roles, particularly those focused on single, repetitive tasks or large-scale organizational management, will diminish. “If your job is as a junior content marketer to take a YouTube video and write a blog post based on that YouTube video, that job won’t exist anymore,” he warns. Similarly, managing a team of a thousand people at a large corporation will become “less and less important because companies are going to be smaller.”
Instead, Bank envisions a world of “Super Individual Contributors” (Super ICs). Drawing an analogy to the “tech lead manager” role at Google – a challenging hybrid of technical expertise and team coordination – he believes everyone will need to combine strategic thinking with hands-on execution, leveraging AI to amplify their impact.
For Bank, this means about two-thirds of his day is individual contributor work – making videos, editing blog posts, talking to customers – while the remaining one-third is dedicated to “coordinating my team of AI agents that helps me with those things.”
The exciting part? This future aligns with what many professionals truly desire. Junior employees crave more strategic influence, while senior managers often wish they could be “closer to the work.” The Super IC model combines the best of both worlds: “We will all need to be strategic enough and senior enough to decide what needs to be done… and we need to be still hands-on enough that we can actually edit the blog post and edit the tweet and make the video ourselves.”
Building Your AI Dream Team: Start Simple, Iterate Constantly
Bank’s personal marketing success story is a testament to the power of AI agents. After a year of “no results” posting consistently on LinkedIn, he deployed his AI marketing team. One post, detailing the “40 marketing agents that I actually use,” garnered an astonishing 1.5 million impressions and the most comments he’d ever received.
His “team” of 40 agents each has a very specific job:
- Automatically generating LinkedIn posts and tweets from new YouTube videos.
- Monitoring competitor social media and alerting him to key topics.
- Tracking competitor pricing changes weekly.
His advice for aspiring Super ICs? “Start simple.” Don’t try to build a single agent to do 25 things at once. Instead: “Create one agent that can do one thing, create a second agent that can do a second thing… and then maybe have one agent that sits above them and helps invoke them when necessary.” Build gradually, one specialized agent at a time.
Crucially, Bank emphasizes that AI agents are not “set it and forget it.” Just like human employees, they require ongoing management, modification, and sometimes even “firing.” He recounts building an agent to create personalized Google Docs after customer calls – a task he initially thought was brilliant. But after 10 iterations, he realized customers weren’t reading the docs. “I fired that agent and I said, ‘Uh, we don’t actually need that job to be done anymore.’” He repurposed its core function, integrating the summary directly into follow-up emails.
Similarly, when his company shifted focus away from SEO, he simply “told my AI SEO agents, you guys can stop for now.” The beauty? “There’s no emotional baggage. There’s no disagreements. There’s no coordination costs.” It’s about giving yourself “superpowers” to do more of the valuable work you love.
Redefining Career Risk in the AI Age
The advent of AI also forces a re-evaluation of what constitutes a “safe” versus a “risky” career. Bank contrasts his parents’ generation, where working 40 years at a single Fortune 500 company was the epitome of security, with today’s reality. “I think that has completely flipped,” he states.
Today, the riskiest career is one “if your skills are too tied to the environment of one particular company that could change over time.” The most robust careers belong to those who are “starting their own companies and who are having lots of different life experience.”
He challenges the common perception that staying at a company like Google is safer than joining a startup. While Google might offer more cash compensation in the short term, “if you think through the long-term lens of your career where it’s about how broad of a network have you built, how many new skills have you built? How have you pushed yourself to develop new experiences? It is way riskier to stay at a company like Google than it is to join a startup or start a startup.”
The choice, according to Bank, is simple: “You can say, ‘I’m skeptical. I don’t believe it. I’m going to put my head in the sand. I’m going to keep working the way I’ve always worked.’ That’s one way to react. The second way to act is like, whoa, this is a cool tool that’s going to completely change the way that work is done. And I want to be on the cutting edge of how I’m using it. Please be person number two.”
For Bank, it’s not about taking more risk, but optimizing for “personal growth and learning new skills.” Stagnation, he argues, is the ultimate risk. Progress means “learning and growing and pushing yourself.”
Educating for an AI-Powered Future
Bank, a father of young children, also reflects on how AI will shape their future careers. He identifies two fundamental skills that will be crucial:
- The ability to clearly articulate what is important, what needs to be done, and give guidance on how to do it. As our jobs become more about efficiently instructing AI, clarity of thought and communication will be paramount. He emphasizes the importance of learning logic and philosophy to develop this skill.
- The ability to build social connection and leverage unique personality. AI can do many things, but human connection, empathy, and unique expression remain irreplaceable. “That’s why I record our own YouTube videos because it’s an expression of my personality and people build a social connection,” he explains.
These two skills – clear articulation for AI and strong human connection – will be “durable skills for the future of any career.” Just as his daughters see self-driving cars as completely normal, the next generation will grow up with AI as an integral, invisible part of their lives, making these human-centric skills even more valuable.
A Global Perspective on AI’s Trajectory
While Bank’s primary focus is on individual empowerment, he also offers a sobering global perspective on AI’s impact. Having spent equal parts of his life in the US, China, and Canada, he observes that both the US and China appear “very intent on losing, not on winning.”
He contrasts the potential applications: “What are Americans going to be using AI for? Well, to make a lot more PowerPoints perhaps and to file a lot more lawsuits against each other.” In contrast, he sees China as “much more manufacturing focused,” poised to use AI to “make more iPhones, make more electric vehicle batteries, make more drones, and make more munitions.” Citing the dramatic disparity in ship production (US built 5, China 1,500 last year), he highlights a potential divergence in how AI will be leveraged for national competitiveness.
Embrace the Superpower
Jacob Bank’s vision is clear: AI is not a threat to human potential, but an unparalleled opportunity to unlock it. We are all using AI for “probably 1% of what we should be using it for.” By embracing the challenge of building and managing AI agents, by prioritizing personal growth over perceived stability, and by cultivating the distinctively human skills of articulation and connection, we can transform our careers.
The future of work isn’t about being replaced; it’s about being amplified. It’s about gaining superpowers, doing more of the work we love, and shaping a professional journey that is dynamic, fulfilling, and truly impactful. The choice to become a Super IC, armed with an army of AI agents, is ours to make – and the time to start is now.
Based on “Building Is the Easy Part Now | Mike Krieger on What AI Changed” from Every Watch the original video
The “Indoor Tree” Dilemma: Crafting Robust Products in the Age of AI
In the relentless sprint of technological evolution, the advent of artificial intelligence has fundamentally reshaped how we build. What once took months or even years can now be accomplished in a matter of hours, turning the act of creation into an almost trivial exercise. Yet, amidst this unprecedented acceleration, a paradox emerges: while AI makes building easier, the true art and science of product design—knowing what to build, and more importantly, what to cut—remains as challenging, and as human, as ever.
This profound shift is at the heart of a recent conversation with Mike Krieger, co-founder of Instagram and now a key figure at Anthropic Labs, a leading AI research company. Speaking on the “Every” podcast, Krieger, alongside the interviewer, delved into the evolving landscape of product development, revealing that the speed of AI brings its own unique set of challenges and demands a new kind of intuition.
The Zero-to-N Paradox: Speed vs. Simplicity
Krieger’s journey from building Instagram, a product famed for its elegant simplicity, to spearheading AI innovation at Anthropic offers a unique vantage point. He recounts the early days of Instagram, a pivot from a more complex product called “Bourbon,” which took nearly a year to develop before they distilled it into what would become Instagram in just three months. This process involved not just adding features but, crucially, removing them.
“The models today are good at adding features,” Krieger observes. “They’re not necessarily good about figuring out what to cut out of the product.” He illustrates this with a startling anecdote: he had Anthropic’s Claude AI rebuild Bourbon. In just two hours, Claude produced a feature-complete product, even adding filters – a feature Instagram eventually adopted, but Bourbon originally lacked. “It knew the eventual future of the product so it decided to build that in,” Krieger muses, highlighting AI’s predictive capabilities but also its tendency towards accretion.
The ability to go “zero to N pretty quickly over the matter of hours” means that an AI can make a multitude of decisions along the way, often leading to over-complicated products. This stands in stark contrast to the human-driven, iterative process where intuition is built over time through real-world usage and painful simplification.
The “Indoor Tree” Metaphor: Why Friction Builds Strength
The interviewer introduces a powerful metaphor to capture this dilemma: “If you grow a tree without it being indoors, without being exposed to wind, it doesn’t get as strong… it needs all these forces pushing it back and forth.” A tree grown indoors might grow, but it leans, it’s not as strong, it lacks the resilience forged by external pressures.
Krieger enthusiastically embraces this “indoor tree” dilemma. In the age of AI, developers can “grow an entire tree indoors,” creating fully formed products without the incremental exposure to users that traditionally hones design intuition. This leads to products that, while feature-rich, lack the inherent strength and clarity derived from real-world friction.
He admits to falling into this trap even at Anthropic Labs. “We way overbuilt for V1 before we even got to early access because you can. You’re like, ‘oh well, we have this option. Why not add this one as well?’” The ease of “vibe coding” – where AI tools rapidly generate code – can lead to “monstrosities,” as the interviewer puts it, that are hard to test and even harder for users to understand. The result is a “matrix of functionality” that feels overwhelming, akin to being “thrown into the final episode of a TV show” without context.
The core insight here, echoing the “lean startup” principles of “you ain’t gonna need it” (YAGNI), is that just because you can build something quickly, doesn’t mean you should include it in the first version. Simplification, driven by human insight and user feedback, remains paramount.
The Art of “Cutting Out” and the Power of Early Launch
The challenge of simplification in the AI era leads to new strategies. Krieger notes that while the “intuitions of the original lean startup ideas are still here,” they manifest at different time scales. One crucial adaptation is a greater willingness to do rewrites. What once might have been a company-killing, year-long endeavor is now a matter of days or weeks, thanks to AI’s speed and ability to “diff” changes. This reduces the pain of tearing down an overcomplicated V1 and starting fresh.
Another strategy is to “launch earlier,” even with a minimal product. Krieger cites Anthropic’s “Co-work” product, which went from concept to V1 in just 10 days. Despite knowing it “could have had 100 things,” launching a useful-enough version proved invaluable. “I’m not sure developing it for another two months, adding 50 features, would have been more useful. In fact, we probably would have been building in a the indoor tree would have been getting built and then the second it hit real world use, it’s like actually nobody wants to do that.”
Agent-Native Design: When Computers “Just Work”
Beyond the speed of building, AI is also transforming the very nature of product interaction. Krieger and the interviewer discuss the concept of “agent-native” design, a philosophy championed at Anthropic and adopted by Every. The idea is that an AI agent should be able to use a product as seamlessly as a human user, with the ability to modify, extend, and interact with all its primitives.
“Computers just work now,” Krieger explains, reflecting on a non-technical person’s observation. What once required arcane command-line incantations can now be handled by an AI like Claude. This “unlocks the functionality that always should have been there or available and just felt like extremely hard for people.”
However, implementing agent-native design is not without its hurdles. Krieger points out that even Anthropic’s core Claude AI still needs to evolve. He gives an example of Claude telling a user how to add an artifact to a project’s knowledge base, rather than simply doing it. This highlights that models, by default, often “think like traditional engineers,” favoring explicit steps and guardrails over the implicit flexibility required for true agent-native interaction.
Teaching AI to be “agent-native” involves providing it with good patterns, templatized examples, and even skills about its own API. Krieger humorously describes having Claude in Claude Code create a skill about skills, leading to a meta-conversation where Claude knew it needed to restart to apply the new skill. This self-awareness is key to unlocking deeper capabilities.
The Challenge of Testing and the Need for “Proof of Thoughtfulness”
Testing agent-native products presents a unique challenge. Traditional unit tests struggle with the inherent unpredictability and extensibility of AI-powered systems. “How do you increase the sort of fidelity of the verification?” Krieger asks. He shares a comical anecdote of an agent-native iOS app where Claude, interacting with its own chat feature, ended up having a conversation with itself, reflecting on a “hard day” and receiving sympathy from its own reflection. This emergent behavior, while amusing, underscores the difficulty of anticipating all possible interactions.
The solution lies in “setting up harnesses that are actually exercising as much of that agent native capability as possible,” pushing the system to its limits within a safe, robust environment.
This new paradigm also changes the nature of human oversight. The interviewer notes a shift from “proof of work” (did the tests pass?) to “proof of use” (show me a Loom video of you or your agent using it). Krieger expands on this, calling it “proof of thoughtfulness.” Engineers must go beyond simply accepting what the AI has built and understand why certain decisions were made. “Did you think this through? Because it’s very easy to end up otherwise with sort of this sort of tower of assumptions that you’re not fully aware of.” The goal is for products to feel “robust,” not “built on sand”—able to flex without falling over, with “your data safe and it’s underneath here.”
Evolving Teams: Conviction, Architects, and Designers as Builders
The shift in product development naturally impacts team structures and hiring. Krieger identifies two crucial directions:
- Primitives and Architectural Robustness: Despite AI’s capabilities, senior technical expertise in distributed systems and core architecture remains invaluable. AI can debug production systems, but architecting them from scratch still benefits from experienced human oversight.
- Product and Prompting Expertise: The rise of AI necessitates deep understanding of prompt engineering and system design. Krieger notes a pairing of product teams with applied AI teams (who help customers with prompts) to bring this expertise in-house, as it often doesn’t reside with traditional software engineers.
Interestingly, designers are taking on new roles. At Anthropic Labs, many designers are “writing and contributing almost as much code as the engineers.” This leads to a “co-founder model” for labs initiatives, where a designer with an original idea might push the boundaries, with a traditional software engineer “paving the trail sometimes behind the designer to make sure that actually works.”
The most critical factor, however, is conviction. Krieger emphasizes the need for someone with “extreme conviction about… the problem space or the question that they’re asking.” Projects without this “founder level” drive, where someone is willing to “break through walls,” are often the “deathnail for projects.”
Both Anthropic Labs and Every prioritize small teams. Krieger highlights that “adding more people actually slows the team down” when an idea is still small enough for an individual to hold in their head. The coordination overhead outweighs the benefits. The interviewer, whose company Every employs a “one-person full-stack GM” model supported by shared resources, agrees, noting that the “number of things you can do with one person is getting bigger.” This is particularly crucial in AI, where models improve so rapidly that “every three to six months you’re you have to throw out like half your product”—a task far easier for a single, agile GM than a large, coordinated team.
The Enduring Art of Product Intuition
The age of AI is a thrilling frontier for product builders. It has democratized creation, accelerating the “building part” to an unprecedented degree. Yet, Mike Krieger’s insights from the front lines of AI innovation reveal that the true challenge has merely shifted. The art of product design—cultivating intuition, knowing what to simplify, ensuring robustness, and fostering deep conviction—remains a profoundly human endeavor. The “indoor tree” can grow fast, but it takes human wisdom and real-world friction to ensure it stands strong.
Based on “In a Driverless World, Who Loses and Who Wins? | Freakonomics Radio” from Freakonomics Radio Network Watch the original video
Collision Course: Boston’s Battle Over Driverless Cars and the Future of Work
Boston, a city steeped in history and fiercely proud of its working-class roots, has become a microcosm for one of the most pressing debates of our time: what happens when advanced technology clashes head-on with human livelihoods? The advent of driverless cars, once a distant sci-fi fantasy, is now a tangible reality, and in cities like Boston, its arrival is sparking intense conflict. This isn’t just about innovation; it’s about jobs, freedom, safety, and the very definition of what it means to be a “driver.”
The story unfolds in a series of heated city council hearings, where the promise of a robotic future meets the raw reality of human struggle. At its heart are three distinct voices, each representing a different facet of this complex equation: the veteran driver fighting for his profession, the impassioned politician defending her constituents, and the disabled advocate seeking a new form of liberation.
The Driver’s Odyssey: From Taxi King to Union Warrior
Our journey begins with Abdi Aziz, a man whose career on Boston’s roads spans three decades. Back in the 1990s, being a taxi driver in Boston was a “decent job,” a career protected by a system of medallions – expensive, city-issued licenses that limited competition and ensured stability. Abdi Aziz drove a limo, then a taxi, for years, building a life around his profession.
Then, in 2011, men from the future arrived at Logan Airport with a plan to change everything. “When Uber came,” Abdi Aziz recalls, “they came to the airport… and they say, ‘Hey, you know, we are introducing you a company that will do same as a taxi, but it’s an app.’” He saw through the veneer instantly. “I say it is good, but you didn’t come here to help us. You come here to kill this business.”
Abdi Aziz understood the impending wave. The medallion system, a de facto monopoly, was about to be shattered by Uber’s app-based model, which bypassed the need for traditional licenses. He knew the industry, as it existed, was doomed. His strategy? “If you cannot beat them, join them.” He became an early Uber recruiter, signing up fellow drivers, and one of the first 100 Uber Black drivers in Boston, investing in an expensive car. For a time, it was even better than the job Uber had disrupted.
But the honeymoon didn’t last. After a few years, Uber’s generosity waned. In 2022, a significant change was rolled out: instead of a set percentage, an algorithm began offering variable rates, which drivers like Abdi Aziz believed significantly increased Uber’s take. “Uber, once it stopped showing them its take, raised that take by a lot,” he states, echoing the sentiment of many drivers who felt like it was a “bait and switch.”
Faced with diminishing returns, Abdi Aziz found himself once again recruiting, but this time for a different kind of disruptive force: a union. He became an early organizer for the App Drivers Union, fighting for better pay and conditions. They were making progress, collecting signatures for a ballot initiative in Massachusetts to secure the right to unionize.
Then, a new specter appeared on the horizon: Waymo, the driverless car company. When Abdi Aziz first heard about their testing in San Francisco in 2022, he knew what was coming. “When Uber came, their aim was to kill taxi business. Now Waymo is to kill the drivers.” This time, there was no “joining them.” This was an existential threat, demanding a different kind of fight.
Boston: A Union Town Draws a Line in the Sand
This fight landed squarely in Boston’s City Hall. Last summer, city councilors began meeting to discuss preemptively banning Waymo from their streets. The stated agenda for “docket 1141” was dry: “to evaluate autonomous vehicle operations.” But what unfolded was anything but.
“Boston is one of the oldest major cities in the country,” explained one councilor, highlighting its “narrow one-way street alleys, and the lack of a traditional grid system.” Even for human drivers, Boston’s labyrinthine streets were a challenge. Yet, the core of the debate quickly shifted from safety to something far more contentious: jobs, particularly union jobs.
“We need to address potential layoffs for our union drivers with the introduction of self-driving cars,” declared a Teamster representative. Boston is a “union town,” a truth repeated over and over by politicians and labor leaders alike. This deep-seated union identity, a legacy stretching from horse-drawn teams to modern trucks, meant that the city council was primed to defend its workers. The App Drivers Union, still in its nascent stages, found powerful allies in the Teamsters and other historic unions, forming a formidable coalition: “Labor United Against Waymo.”
Councilor Julia Mahia, a former MTV reporter with a noticeable edge, emerged as a vocal champion of the drivers. “I’m still in shock that I have to even have this conversation that here we are in this day and age trying to defend ourselves from robots taking over our jobs,” she stated. For Mahia, the issue was deeply personal and moral. Having grown up with an undocumented mother who cleaned offices, she understood the vulnerability of low-wage workers. She drew parallels to self-checkout machines in supermarkets, which replaced jobs often held by retired individuals, high school students, or people with disabilities. “We are not thinking about other people. We’re often just thinking about ourselves and what is the quickest way to get out.”
Mahia grilled Matt Walsh, Waymo’s regional head of state and local public policy, on job displacement. Walsh, looking every bit the tech executive, tried to pivot to safety statistics, citing Waymo’s record of being “five times less in injury-causing crashes than human drivers.” But Mahia was unyielding. “What we are doing is creating an opportunity for people to choose to not support humans and the workforce. That is the choice that we’re giving people.” When Walsh referred to the autonomous system as a “Waymo driver,” Mahia retorted, “Waymo is not a driver. Waymo is a robot. So let’s be really clear about what it is. It’s an apparatus.”
Walsh struggled to provide concrete answers about job opportunities for displaced drivers, offering only vague “workforce efforts” within the autonomous vehicle industry. To Mahia, his responses lacked “humility and humanity.” “He could have won me over a little bit if he gave me a little bit more heart,” she later mused, before adding with a laugh, “No, nobody could win me over.” For Boston, the outsider tech giant, valued at $126 billion, was seen as a threat to be repelled, not a partner to embrace.
The Unseen Battle: A Quest for Autonomy
Yet, amidst the chorus of union solidarity, another voice was waiting to be heard—a voice that spoke not of jobs lost, but of freedom gained. Carl Richardson, almost completely blind and significantly hearing impaired, attended the first hearing as a private citizen. He arrived early, signed up to speak, and watched as union protests filled the plaza and the hearing room. He felt “far outnumbered,” noting that many disabled attendees left, discouraged by the overwhelming opposition.
Carl, accompanied by his guide dog Dayton, waited nearly four hours to speak, slotted almost at the very end. By the time he was called, most city councilors had already departed for another union event, leaving him to address a largely empty room.
Carl’s testimony was a stark counterpoint to the prevailing narrative. He spoke of the “impact on the union and the drivers and the workforce,” but then asked, “Let’s talk about the communities I think it would impact in favor of not only people with physical disabilities like myself but people with mental…” He highlighted the staggering unemployment rate in the disability community, twice as high as the general workforce, largely due to transportation barriers. “Do you know how many jobs I’ve turned down because I can’t get there or how many interviews?”
He recounted the frequent discrimination he faced from human drivers. “At least once a week, I get denied access to Uber and Lift because they refuse to take me because I have a service dog and they end denying me my civil rights.” He also mentioned drivers refusing rides beyond city limits to maximize revenue, limiting his life to the confines of Boston.
Carl was born with Usher syndrome, a genetic condition that gradually claimed his vision and hearing. He drove until he was 30, only stopping when he realized he was a danger to himself and others. “I know what I’ve lost,” he said, “I want that feeling that I used to have when I drove of freedom and independence and mobility… I want that back.” He never thought he would get it, but autonomous vehicles offered a beacon of hope. He envisions a future where the elderly, teenagers texting, or anyone who loses their license could regain their independence. He even has a savings account where he puts aside a few hundred dollars a month “just for the ability for me to buy an autonomous vehicle someday.”
He shared a deeply personal story of an emergency: his mother, receiving a scam call about being arrested, needed him. He was denied three rides in a row, unable to reach her. “All I wanted the ability was to be able to go home to my mom and say, ‘You’re okay and I love you.’” He concluded, “Definitely think about the human component and the people component, but think about it for the whole community at large, not just the union.”
Carl’s powerful plea, delivered to an almost empty room, underscored the profound complexity of the debate. While one group fought to protect existing jobs, another saw the technology as a pathway to a more inclusive, autonomous life.
The Unfolding Future: Can Compromise Be Found?
The Boston hearings, and the proposed ordinance that would effectively ban driverless cars without a human safety driver, represent a critical juncture. The tension between job protection and technological advancement is palpable, and Boston’s “union town” identity provides a stark contrast to cities in “red and purple states like Austin and Phoenix [that] mostly welcome Whimo.”
Waymo, through its Northeast policy manager Anthony Perez, acknowledged that “over time there would be what he called transition for app drivers,” but insisted it wouldn’t be a one-to-one displacement, citing new jobs in cleaning, maintenance, and repair. Yet, the estimate of one job created for every five robo-taxis offers little comfort to those facing direct displacement.
The question remains: are our politics ready for this challenge? Can cities like Boston find a compromise that protects workers while embracing innovations that could profoundly benefit others? The emotional and economic stakes are incredibly high. Abdi Aziz’s fight for his livelihood, Councilor Mahia’s defense of the working class, and Carl Richardson’s yearning for independence all highlight the human dimensions of a technological revolution. As driverless cars roll out across the nation, Boston’s fierce debate serves as a powerful reminder that the road ahead is not just about algorithms and sensors, but about people, their jobs, and their fundamental right to autonomy.
Based on “How Stripe deploys 1,300 AI-written PRs per week” from How I AI Watch the original video
Stripe’s Silent Army: How AI Minions Are Rewriting Software Development’s Future
Imagine a software company where the majority of new code isn’t written by humans, but by an army of autonomous AI agents. This isn’t a distant sci-fi fantasy; it’s the present reality at Stripe, the global financial technology giant. With a staggering 1,300 AI-generated Pull Requests (PRs) — proposed code changes — landing weekly with “no human assistance besides review,” Stripe is pioneering a new era of “agentic engineering.”
At the heart of this revolution are what Stripe engineers affectionately call “minions.” These aren’t just intelligent chatbots; they are sophisticated AI agents capable of provisioning their own development environments, understanding complex prompts, and executing code changes across Stripe’s massive codebase.
From Idea to Code: The Minion Magic
The journey of a new feature or fix at Stripe often begins in familiar, human-centric spaces: a Google Doc outlining a new feature, a Jira ticket, or even a casual conversation in Slack. But instead of an engineer immediately diving into a text editor, a simple emoji click can now kick off an AI minion.
Steve Khaliski, a software engineer at Stripe, describes this shift as a dramatic reduction in “activation energy.” “I don’t remember the last time I started work in the text editor,” he admits. Instead, the minion “will sort of attempt to one-shot resolving that prompt using all the tools that are available at Stripe.” This includes internal documentation, continuous integration (CI) systems, and test data, allowing the agent to loop through and attempt to solve the prompt autonomously.
For larger organizations, this streamlined approach is a game-changer. Claravel, host of the “How I AI” channel, highlights the inherent friction that often plagues big companies: “Not only can I have one of these, but I could have many, many of these running in parallel in isolated environments making isolated changes all at the same time.” This eliminates bottlenecks caused by lack of technical expertise, operational coordination, or siloed thinking, allowing good ideas to quickly translate into tangible code.
The Virtuous Loop: Developer Experience Powers AI
The success of Stripe’s minions isn’t just about advanced AI models; it’s deeply rooted in the company’s long-standing investment in developer experience (DX). Stripe has a history of providing engineers with excellent tooling, including hosted development environments that can be spun up with all necessary code and services pre-configured.
“What’s good for the developer is good for the agent,” notes Zach from Launch Darkly, a sentiment echoed by Khaliski. If a human engineer struggles with poor documentation, complex setup, or unreliable tools, an AI agent will face the same, if not greater, challenges. Conversely, a robust, well-documented, and easy-to-use developer environment creates a “blessed path” for agents to follow. “If there’s a very blessed path for 90% of the common activities… that makes the propensity that the agent succeeds really high too,” Khaliski explains. This creates a virtuous cycle: investing in DX for human engineers directly benefits AI agents, and improving “agent experience” in turn optimizes the development process for humans.
This synergy also extends to the physical limitations of traditional development. Even the most powerful personal laptops, affectionately nicknamed “Big Boy” by some engineers, buckle under the strain of running multiple “work trees” (isolated code environments). “It starts to sound like an airplane taking off,” Claravel quips. Stripe circumvents this by leveraging cloud-based virtual environments. This not only allows for the parallel execution of countless minions without taxing local machines but also enables engineers to kick off complex tasks from anywhere – even on a subway commute – and jump in once the AI has completed the initial heavy lifting.
Beyond Coding: Agents as Economic Actors
Stripe’s vision for AI extends far beyond internal code generation. The company is exploring a groundbreaking concept: AI agents as “economic actors” — entities capable of spending money to achieve their goals. This is part of Stripe’s three-pronged approach to AI: accelerating internal development, supporting businesses that leverage AI, and enabling agents to transact financially.
Kaliski illustrates this with a compelling demo: planning a birthday party for a colleague. The prompt given to the AI agent is simple: “Research Jen Lee, figure out a good idea for her birthday, find a place… send invites… donate to Stripe Climate.” What unfolds is a series of micro-transactions, facilitated by a “machine payment protocol” co-designed with Tempo.
- Research: The agent pays Browserbase a fraction of a cent to spin up a temporary browser session, navigate to the colleague’s website, and discover her interest in matcha.
- Venue Search: It then uses Parallel AI to search for matcha-themed venues in New York.
- Invitations: The agent interacts with Postal Form, a service that takes a PDF and mails it physically. The AI writes the code to generate the PDF invite locally, then “pays” Postal Form to send it.
- Carbon Offset: Finally, to account for the energy consumed by its 70,000 token usage, the agent makes a $1.65 contribution to Stripe Climate, offsetting 4.44 kilograms of carbon.
This demo showcases a future where AI agents don’t just consume tokens from large language models, but also seamlessly pay for third-party services as needed. This “agent receipt” makes the economics of AI explicit, showing the token cost alongside the dollar cost of external services. Khaliski notes, “The token and the currency that backs it, like they feel closer than ever.” This opens up entirely new business models focused on providing hyper-useful, ephemeral API interactions directly to agents, rather than building extensive dashboards or administrative panels for human users.
The Evolving Role of the Human Engineer
With AI handling so much of the coding, what becomes of the human engineer? Khaliski believes the focus will shift. “If coding becomes easier and coding historically has been the bottleneck in product development, it’s just going to shift to other areas.” Engineers can reallocate their time to more high-value tasks: deeply reviewing AI-generated code, engaging with users, and generating innovative ideas.
Moreover, the natural language interface of these agents means that non-engineers — product managers, designers, or anyone with a clear idea — can now initiate development work with a simple prompt. This dramatically lowers the barrier to entry, enabling a broader range of company personnel to contribute directly to product development.
The Art of Prompting an AI
Even with advanced agents, human guidance remains crucial. Khaliski shares his personal “prompting strategy” for when an AI minion doesn’t “one-shot” the solution:
- Politeness: “I have made a concerted effort to always be polite,” he confesses, half-jokingly, citing sci-fi caution.
- Justification: Asking the AI to “explain or justify itself” often helps uncover its reasoning and identify missteps.
- Breadcrumbs: When the direction is clear, Khaliski might start the work himself, leaving “breadcrumbs” (like a partial diff or a
git status) for the AI to follow and complete. - Skill Capture: For recurring tasks, he aims to capture the successful prompt or a “skill” that can be injected back into the agent for future use.
Stripe’s journey with AI minions is a powerful testament to the transformative potential of agentic engineering. By leveraging strong developer tooling, cloud environments, and an innovative economic framework, Stripe is not just accelerating its own development but also laying the groundwork for a future where AI agents are not only code generators but active, transacting participants in the digital economy. The silent army is growing, and the future of software development is being rewritten, one AI-generated PR at a time.
Based on “Are Higher Energy Prices Here to Stay?” from New York Times Podcasts Watch the original video
The Energy Earthquake: How Gulf Attacks Sent Global Prices Soaring – And Why They Might Stay There
For weeks, the global conversation around energy prices has been dominated by the escalating conflict in the Persian Gulf. Initially, many, including President Trump, suggested that rising costs would be a temporary blip, a mere inconvenience tied to the closure of the Strait of Hormuz. But a series of devastating strikes last week on critical natural gas facilities in Qatar and other Gulf nations have fundamentally reshaped this outlook, transforming a short-term transit problem into a years-long crisis of production capacity.
“Instead of talking about the impact in terms of days and weeks, now we’re talking about it in terms of months and years,” explains Patricia Cohen, a New York Times colleague, highlighting the profound shift in the energy landscape. The economic ripples from these attacks are already being felt worldwide, promising a future of elevated energy costs and far-reaching consequences.
From Transit to Destruction: The New Phase of Conflict
The initial focus on the Strait of Hormuz centered on a transportation bottleneck. If the strait, a vital choke point for oil and gas shipping, were closed, prices would spike due to uncertainty over when it would reopen. While still a major concern, recent events have introduced a more insidious threat: direct damage to the infrastructure that extracts and processes energy.
The conflict escalated rapidly. Israel struck a major Iranian gas complex, prompting swift retaliation from Iran. Among the most significant targets was Qatar’s Ras Laffan, the world’s largest liquefied natural gas (LNG) facility. An Iranian missile strike severely damaged this critical hub, targeting its gas fields and, crucially, destroying two “LNG trains.”
These “trains” are not what their name suggests. They are highly complex, multi-billion-dollar industrial plants designed to process natural gas. Here, gas is super-chilled to a staggering 240 to 260 degrees Fahrenheit below zero, then compressed, reducing its volume by 600 times and turning it into a liquid. This process makes it feasible to transport vast quantities of natural gas across oceans in specialized containers. Building such facilities is an arduous and time-consuming endeavor.
The destruction of these trains in Qatar is a monumental blow. Qatar is the globe’s largest producer of LNG, supplying approximately 20% of the world’s liquefied natural gas. While production had been temporarily suspended earlier in the conflict due to transit issues in the Strait of Hormuz, the recent attacks destroyed nearly 20% of Qatar’s actual ability to produce LNG. This isn’t a matter of simply reopening a shipping lane; it’s a matter of rebuilding complex industrial plants, a task that could take as long as five years.
The Indispensable Role of LNG
Liquefied natural gas plays a pivotal role in the global energy mix. Many countries have shifted from dirtier fossil fuels like coal to LNG because it’s significantly cleaner – about 30% cleaner than oil. This makes it a crucial component in their energy diversification strategies, particularly for nations aiming to reduce their carbon footprint while maintaining robust energy supplies.
Asian powerhouses like Japan and South Korea are heavily reliant on LNG. Japan, for instance, has reduced its dependence on nuclear power and coal, with LNG now constituting about 21% of its total energy supply and generating 30% of its electricity. In South Korea, LNG accounts for 20% of its total energy and 25% of its electricity, with usage increasing by over 200% in the last 25 years. This energy powers homes, businesses, and industries, from charging cell phones to running massive factories.
Beyond direct energy generation, LNG processing yields critical byproducts essential for industrial production and the global food supply:
- Naphtha: A crucial component in the manufacture of plastics and other gas products.
- Helium: Indispensable for creating semiconductors, the building blocks of modern electronics.
- Nitrogen-based fertilizers: Vital for agriculture. The soaring prices and potential scarcity of these fertilizers are fueling widespread fears about rising food costs and global food security.
A Global “Double Whammy”: Ripple Effects Across Continents
The destruction of LNG capacity, layered on top of existing oil price volatility, has created a “double whammy” for economies worldwide.
Vulnerable Nations Bear the Brunt: Developing countries and fragile economies are facing dire consequences. In Pakistan, skyrocketing fuel prices have led to widespread power shortages, accompanied by the closure of schools and government offices. Sri Lanka faces similar challenges. Thailand has implemented extreme energy-saving measures, ordering government workers to suspend overseas trips, work from home, take stairs instead of elevators, and even shortening the work week to four days. These actions underscore the profound fear that these nations harbor regarding the conflict’s prolonged impact.
Rich Countries Are Not Immune: The crisis is also reaching affluent nations. South Korea has imposed a fuel cap for the first time in 30 years and launched an energy savings campaign, urging citizens to take shorter showers and ride bicycles. In Europe, gas prices have more than doubled since the conflict began. Leaders are scrambling to introduce measures to help consumers afford soaring heating bills, often through tax cuts on energy. In Britain, natural gas prices have surged by 40% since the war with Iran started. Many households and businesses are now forced to ration heating, leaving rooms unheated to manage costs.
Even the United States, the world’s largest oil producer and a significant LNG exporter, is not insulated. While America supplies much of Europe’s LNG, energy prices are set on a global market. When global prices rise, Americans feel the pinch, both directly through higher utility and fuel costs, and indirectly through broader economic impacts.
The Indirect Fallout: Inflation, Interest Rates, and AI
Patricia Cohen emphasizes that the repercussions extend far beyond direct energy costs, creating a cascade of “knock-on effects”:
-
Erosion of Trust and Investment: The Persian Gulf has historically been a magnet for investment, perceived as a safe region for business and residence. The recent attacks, however, have introduced a palpable sense of instability and risk. This “rumbling underneath” threatens to deter foreign investment, undermining years of effort by Gulf States to attract capital.
-
Inflationary Spiral: Higher energy prices translate directly into increased costs for nearly every product and service. The cost of growing an avocado or producing a pair of sneakers isn’t just about raw materials; it’s also about the increasingly expensive transportation required to move them. This pervasive increase in costs fuels inflation across the board.
-
Interest Rate Hikes: Central banks, tasked with controlling inflation, may respond by raising interest rates. While a necessary tool, this makes borrowing money more expensive for consumers and businesses alike, potentially slowing economic growth.
-
A Threat to the AI Boom: The burgeoning artificial intelligence (AI) industry, a key driver of the US economy and stock market, is particularly vulnerable. AI data centers are enormous energy hogs, and their construction requires massive loans. Rising energy costs directly impact their operational expenses, while higher interest rates dramatically increase the cost of building these facilities. Wall Street analysts warn that AI development is highly sensitive to interest rate increases, potentially stalling a sector that has been propping up the economy.
-
Recessionary Fears: The combination of sustained high energy prices, rampant inflation, and increased borrowing costs paints a grim picture. If oil prices were to climb to $180 a barrel or higher, “then it’s going to be very difficult to avoid a recession,” Cohen states. Consumer spending, which drives 70% of the US economy, tends to falter when uncertainty reigns, and businesses become reluctant to invest, creating a “crisis of confidence.”
The Long Road to “Normal” – Or a New Normal?
Even if the war were to cease tomorrow, the path back to pre-conflict energy stability is long and fraught with challenges. The head of the International Energy Agency (IEA) estimates that it would take at least six months just to restore existing production that was halted due to the conflict, let alone rebuild the damaged infrastructure. The full repair of Qatar’s LNG facilities, as noted, could take years.
To ease the immediate crunch, some levers are being pulled. The US and the IEA have released supplies from strategic petroleum reserves. More controversially, the US has lifted sanctions on Russian oil exports, a move that provides funds for Russia’s war effort in Ukraine, and also lifted sanctions on Iran’s oil exports – an ironic “multi-billion dollar war to buy oil from Iran.”
The most sustainable, albeit long-term, solution lies in the accelerated development of alternative and renewable energy sources. Significant upfront investment in solar, wind, and potentially nuclear power could decrease global dependence on natural gas and oil, diversifying supply and ultimately lowering costs. However, as Cohen cautions, “the world is an interconnected place,” and even energy independence cannot fully insulate a nation from global market dynamics.
The Looming Threat of Escalation
The true depth of this crisis is underscored by the potential for further escalation. Despite being relatively weaker than the US and Israel, Iran has demonstrated an alarming ability to exert enormous leverage over the global economy. This speaks to the changing nature of modern warfare, where a single, well-placed attack on critical infrastructure can have devastating, far-reaching consequences. “One guy on a speedboat with a bomb in the Strait of Hormuz could screw up all international shipping,” Cohen observes.
The worry is not just about rebuilding what’s been lost, but preventing further damage. What if more gas fields or processing facilities in Qatar, Saudi Arabia, or other regional players are targeted? The Iranian regime, viewing this conflict as an existential threat, may be willing to take extreme measures to ensure its survival. This miscalculation, coupled with the profound ripple effects felt across the globe, has led the head of the IEA to label this period “the greatest global energy security threat in history.”
As the conflict enters its fourth week, the future remains highly unpredictable. The question is no longer if higher energy prices are here to stay, but for how long, and what further economic and geopolitical shocks await a world grappling with unprecedented energy insecurity.
한국어
“AI is way Underhyped. This Silicon Valley CEO proves it | Relay.app, Jacob Bank” — EO 기반 기사 원본 영상 보기
AI는 과소평가되었다: 실리콘밸리 CEO가 밝히는 미래 직업의 핵심 역량
[서울경제] “AI는 엄청나게 과소평가되어 있습니다.” 실리콘밸리의 스타트업 릴레이닷앱(Relay.app)의 설립자이자 CEO인 제이콥 뱅크(Jacob Bank)는 AI에 대한 기존의 인식을 뒤엎는 도발적인 주장을 던진다. 그는 AI가 단순한 생산성 도구를 넘어, 개인의 역량을 극대화하고 직업의 본질을 변화시킬 ‘슈퍼파워’를 제공한다고 강조한다. 뱅크 CEO의 경험과 통찰을 통해 AI 시대의 새로운 직업관과 핵심 역량에 대해 깊이 들여다본다.
AI, 개인을 위한 40인의 마케팅 팀
뱅크 CEO는 릴레이닷앱에서 유일한 마케팅 담당자이지만, 무려 40개의 AI 마케팅 에이전트(AI Agent) 팀을 운영하고 있다. 이 에이전트들은 그를 대신해 다양한 마케팅 업무를 수행하며 놀라운 성과를 창출한다. 한 예로, 그가 링크드인에 올린 “내가 실제로 사용하는 40개의 마케팅 에이전트”라는 게시물은 150만 회 이상의 노출과 링크드인 게시물 역사상 가장 많은 댓글을 기록했다. 이는 AI 코치의 도움 덕분이었다. AI 코치는 그의 게시물 스타일과 주제를 분석하여 “이런 스타일이 잘 먹히니 더 많이 하세요. 저런 건 줄이세요”와 같은 구체적인 피드백을 제공했다.
뱅크 CEO는 AI 에이전트가 가져다주는 비용 효율성 또한 강조한다. 한 달에 12,500달러를 받는 고품질 마케팅 계약자 4명을 고용한다면 월 5만 달러가 들지만, AI 에이전트를 활용하면 월 500달러면 충분하다는 것이다. 릴레이닷앱은 현재 9명의 팀원(엔지니어 5명, 디자이너 2명, 제품 담당자 1명, CEO 본인)으로 운영되고 있는데, AI가 없었다면 동일한 성과를 내기 위해 15명의 팀원이 필요했을 것이라고 뱅크 CEO는 설명한다. 그는 AI 에이전트 구축 능력이 지난 40년간 마이크로소프트 엑셀(Microsoft Excel) 사용 능력과 같았던 것처럼, 향후 30~40년간 모든 전문가의 경력을 좌우할 필수 역량이 될 것이라고 단언한다.
AI에 대한 오해: 인턴이 아닌 전략적 파트너
뱅크 CEO는 과거에는 AI를 ‘인턴’에 비유했지만, 이는 잘못된 비유임을 깨달았다고 말한다. AI가 단순한 시간 절약형 작업을 수행하는 인턴의 역할을 할 수 있는 것은 사실이다. 하지만 AI는 인턴이 결코 할 수 없는 전략 분석, 경쟁 분석, 콘텐츠 제작과 같은 고차원적인 업무도 수행할 수 있다.
그는 자신의 경험을 예로 들어 설명한다. 그는 영업에 능숙하지 않고 관련 배경 지식도 부족하지만, AI 영업 코치의 도움을 받는다. AI 코치는 그의 모든 회의록을 검토하여 “이 부분에 대해 더 잘 설명할 수 있었을 것입니다. 이 가치를 더 명확히 전달했어야 했습니다. 데모로 너무 서둘러 넘어갔습니다. 더 많은 발견 질문(discovery questions)을 했어야 합니다”와 같은 구체적인 피드백을 제공한다. 유능한 인간 영업 코치를 고용하면 한 달에 1만 달러가 들 수 있지만, AI 영업 코치는 스스로 구축할 수 있으며 운영 비용은 주당 5달러에 불과하다. 뱅크 CEO는 새로운 역할이나 기능에 도전할 때, AI 코치를 활용하여 자신의 활동 데이터를 제공하고 적절한 피드백을 받는 것이 매우 효과적이라고 강조한다.
미래의 인재상: ‘슈퍼 IC’의 시대
AI는 전통적인 직업의 지형도를 근본적으로 변화시킬 것이다. 뱅크 CEO는 “유튜브 영상을 보고 블로그 게시물을 작성하는 주니어 콘텐츠 마케터의 일자리는 더 이상 존재하지 않을 것”이라고 경고한다. 단순한 콘텐츠 재활용(content repurposing) 작업만으로는 직업을 유지하기 어려워진다는 의미다. 또한, 대기업에서 수천 명의 팀을 관리하는 역할도 점차 중요성이 줄어들 것이라고 예측한다. 기업 규모가 더 작고 민첩해지면서 대규모 인력 관리 기술의 가치가 하락할 것이기 때문이다.
그는 미래에는 모두가 자신과 같은 ‘슈퍼 IC(Individual Contributor, 개인 기여자)‘가 될 것이라고 말한다. 그의 업무 시간 중 약 3분의 2는 유튜브 영상 제작, 블로그 게시물 편집, 콘텐츠 발행, 고객 소통과 같은 직접적인 업무(IC work)로 채워진다. 나머지 3분의 1은 이러한 업무를 돕는 AI 에이전트 팀을 조율하는 데 사용된다.
뱅크 CEO는 이러한 ‘슈퍼 IC’의 미래가 매우 흥미롭다고 말한다. 대부분의 사람들이 실제로 좋아하는 일, 즉 전략적 영향력을 행사하고 고객 및 제품에 더 가까이 다가가는 일을 할 수 있게 되기 때문이다. 그는 미래의 모든 역할이 전략적 사고와 실무 능력을 모두 요구할 것이라고 전망한다. 즉, 무엇을 해야 할지 결정하고 ‘좋은 결과’가 무엇인지 파악하여 AI에 올바른 코칭을 제공할 만큼 전략적이고 시니어(senior)하면서도, 블로그 게시물을 직접 편집하고 트윗을 작성하며 영상을 만들 수 있을 만큼 실무적(hands-on)이어야 한다는 것이다.
AI 에이전트 구축 및 활용법: 단순함에서 시작하라
뱅크 CEO는 AI 에이전트 구축의 핵심 원칙을 제시한다. 바로 **“단순함에서 시작하라”**는 것이다. 40개의 에이전트가 한꺼번에 복잡해 보일 수 있지만, 각각의 에이전트는 매우 단순하고 구체적인 한 가지 작업을 수행하도록 설계되어 있다.
- 새 유튜브 영상이 올라오면 자동으로 링크드인 게시물 생성
- 새 유튜브 영상이 올라오면 자동으로 트윗 생성
- 경쟁사 CEO가 트위터나 링크드인에 게시물을 올리면 알림
- 매주 경쟁사 가격을 확인하고 중요한 변경 사항 알림
그는 한 번에 25가지 일을 하려는 ‘만능 에이전트’는 성공적이지 못했다고 고백한다. 초보자라면 한 가지 일을 하는 에이전트를 만들고, 두 번째 에이전트, 세 번째 에이전트를 추가하는 식으로 점진적으로 확장해야 한다. 그리고 이들을 조율하는 상위 에이전트를 두는 방식으로 발전시켜야 한다. 40개의 에이전트 조직도를 한 번에 만들려 하지 말고, 단 하나의 에이전트부터 시작해야 한다.
또한, AI 에이전트는 **“설정하고 잊어버리는(set it and forget it) 방식이 아니다”**라는 점을 명심해야 한다. 뱅크 CEO는 자신의 AI 에이전트들을 끊임없이 수정한다고 말한다. 예를 들어, 그는 고객과의 미팅 후 회의록을 바탕으로 개인화된 구글 문서(Google Doc)를 생성하는 AI 에이전트를 만들었다. 하지만 고객들이 구글 문서를 잘 읽지 않는다는 것을 깨닫고, 해당 에이전트를 해고(fired)하고 요약 내용을 이메일에 직접 포함하도록 재활용(repurposed)했다.
마케팅 전략이 바뀌면 AI 에이전트의 역할도 조정된다. 1년 전에는 SEO(검색 엔진 최적화)에 집중했지만, 현재는 새로운 사용자 유입에 큰 영향을 미치지 않아 SEO 에이전트들에게 “지금은 쉬세요. 나중에 다시 부를게요”라고 지시했다고 한다. 인간 직원이나 에이전시에게 하듯이, AI 에이전트에게도 유연하게 업무를 지시하고 중단시킬 수 있는 것이다. 그는 AI 에이전트와의 작업에는 “감정적 부담이나 의견 충돌, 조율 비용이 없다”는 점을 가장 큰 장점으로 꼽으며, AI를 통해 개인이 “슈퍼파워”를 얻게 된다고 강조한다.
커리어의 재정의: 성장이 곧 안정이다
뱅크 CEO는 ‘위험한 커리어’와 ‘안정적인 커리어’에 대한 인식이 완전히 바뀌었다고 지적한다. 부모님 세대에는 한 대기업에서 40년간 일하는 것이 안정적이었고, 창업이나 잦은 이직은 위험하다고 여겨졌다. 그러나 이제는 반대다. 특정 회사의 환경에 너무 묶여 있는 기술을 가진 직업이 가장 위험하며, 끊임없이 새로운 경험을 쌓고 자신의 회사를 시작하는 사람들이 가장 견고한 커리어를 구축한다고 말한다.
그는 구글(Google) 출신 팀원들을 예로 들며 설명한다. 구글에 머무는 것과 스타트업에 합류하는 것 중 어느 것이 더 위험하냐고 물으면 대부분 구글이 더 안전하다고 답할 것이다. 단기적인 현금 보상 측면에서는 사실일 수 있다. 그러나 장기적인 커리어 관점에서, 즉 얼마나 넓은 네트워크를 구축하고, 얼마나 많은 새로운 기술을 습득하며, 새로운 경험을 위해 자신을 얼마나 밀어붙였는지를 본다면, 구글과 같은 회사에 머무르는 것이 스타트업에 합류하거나 창업하는 것보다 훨씬 더 위험하다는 것이 뱅크 CEO의 주장이다.
그는 현재의 AI 에이전트 열풍에 대응하는 두 가지 방식을 제시한다. 하나는 “회의적이고, 믿지 않으며, 머리를 모래에 파묻고 기존 방식대로 일하는 것”이다. 다른 하나는 “와, 이건 일하는 방식을 완전히 바꿀 멋진 도구다. 나는 이 변화의 최전선에 서고 싶다”고 반응하는 것이다. 뱅크 CEO는 후자의 태도를 가질 것을 강력히 권한다. 그는 더 많은 위험을 감수하라고 말하는 것이 아니라, **“개인의 성장과 새로운 기술 습득을 위해 커리어를 최적화하라”**고 조언한다. 성장을 위해 새로운 불확실한 환경에서 더 많은 책임을 지고 도전하는 것이야말로 진정한 안정이라는 것이다. 그에게 정체(stagnation)야말로 가장 위험한 행동이다.
그는 마케팅 경험이 전혀 없었고, 1년 전까지 소셜 미디어 플랫폼도 사용하지 않았으며, 공개적으로 자신에 대해 공유하는 것을 싫어했다고 고백한다. 하지만 회사의 성공을 위해 링크드인 마케팅이 최적의 채널임을 깨닫고 1년간 정말 열심히 노력하여 이제는 마케팅을 즐기게 되었다. 그는 이러한 ‘불확실성에 도전하는 기회’를 통해 배우고 성장할 수 있었다고 말한다.
미래 세대를 위한 교육과 핵심 역량
뱅크 CEO는 자신의 아이들이 살아갈 미래 세계에 대한 깊은 고민을 나눈다. 샌프란시스코의 자율주행차 테스트 지역 근처에 살면서 아이들이 운전자가 없는 자동차를 보고 ‘와이모(Waymo) 툭스(Zooks)‘라고 외치는 것이 너무나 자연스러운 풍경이 되었다. 그는 아이들이 운전할 필요 없이 더 안전하고 편리하며 친환경적인 자율주행차의 미래를 긍정적으로 바라본다. 반면, 알고리즘이 아무리 뛰어나도 아이들이 유튜브 쇼츠만 계속 보는 것은 걱정된다고 말한다.
그는 아이들의 교육에서 두 가지 핵심 역량을 강조한다.
- 명확하게 의도를 전달하고 지침을 제공하는 능력: 미래에는 AI에게 무엇이 중요하고, 무엇을 해야 하며, 어떻게 해야 하는지 효율적으로 지시하는 능력이 더욱 중요해질 것이다. 이는 논리와 철학 교육을 통해 길러질 수 있다고 뱅크 CEO는 강조한다.
- AI가 할 수 없는 방식으로 자신만의 개성과 능력을 발휘하는 능력: 예를 들어, 그가 직접 유튜브 영상을 만드는 이유는 자신의 개성을 표현하고 사람들과 사회적 연결을 구축하기 위함이다. 이러한 ‘인간적인 연결’은 AI가 대체하기 어려운 핵심 역량이 될 것이다.
결론적으로, 뱅크 CEO는 미래 직업에서 가장 중요한 두 가지 능력으로 **“자신이 원하는 것을 명확하게 표현하고 AI에게 지시하는 능력”**과 **“타인과 사회적 연결을 구축하는 능력”**을 꼽는다. 이 두 가지는 어떤 커리어에서도 지속 가능한 핵심 기술이 될 것이라고 그는 믿는다.
미중 AI 경쟁의 현주소와 미국의 위기
마지막으로 뱅크 CEO는 AI 경쟁의 지정학적 측면에 대한 날카로운 통찰을 공유했다. 그는 미국이 인공지능 분야에서 중국보다 앞서 있는 것은 분명하지만, 기술 리더십의 격차가 “매우 좁다”고 평가한다. 그는 미국이 AI를 “더 많은 파워포인트(PowerPoint)를 만들고, 더 많은 소송을 제기하는 데” 사용할 것이라고 비판한다. 반면 중국은 훨씬 더 제조 중심적이다. 중국은 AI를 활용하여 “더 많은 아이폰(iPhone), 더 많은 전기차 배터리, 더 많은 드론, 그리고 더 많은 군수품”을 만들 것이다. 작년에 미국이 약 5척의 선박을 건조한 반면, 중국은 약 1,500척을 건조했다는 비교는 이러한 차이를 극명하게 보여준다.
그는 미국, 중국, 캐나다에서 거의 동일한 시간을 보냈다고 밝히며, 현재 미국과 중국 모두 “이기려는 것이 아니라 지려는 데 매우 집중하고 있는 것 같다”고 꼬집는다. 양국 모두 큰 실수를 저지르고 있다는 비판적인 시각을 드러내며, AI 시대의 기술 경쟁이 단순히 경제적 영역을 넘어 국가의 미래와 직결되는 중대한 문제임을 시사했다.
제이콥 뱅크 CEO의 이야기는 AI가 우리에게 위협이 아닌 기회로 다가올 수 있음을 보여준다. AI를 도구 삼아 개인의 역량을 확장하고, 끊임없이 배우고 성장하며, 인간 고유의 강점을 발휘하는 것이야말로 급변하는 미래 사회에서 살아남는 길임을 그는 역설한다. AI는 결코 과대평가된 기술이 아니며, 오히려 우리가 상상하는 것 이상으로 우리의 삶과 일을 변화시킬 ‘과소평가된 슈퍼파워’라는 그의 메시지는 AI 시대의 새로운 항해를 준비하는 우리 모두에게 깊은 울림을 준다.
“Building Is the Easy Part Now | Mike Krieger on What AI Changed” — Every 기반 기사 원본 영상 보기
AI 시대, ‘건축’은 쉬워졌지만 ‘설계’는 더 어려워졌다: 마이크 크리거가 말하는 제품 개발의 본질
인스타그램(Instagram)의 공동창업자이자 현재는 선도적인 AI 연구 기업 앤트로픽(Anthropic)에서 제품 개발을 이끌고 있는 마이크 크리거(Mike Krieger)가 AI 시대의 제품 개발 패러다임 변화에 대한 심도 깊은 통찰을 공유했습니다. 유튜브 채널 ‘Every’와의 인터뷰에서 그는 AI가 개발 속도를 혁명적으로 가속화했지만, 동시에 ‘무엇을 만들지’, ‘어떻게 만들지’에 대한 근본적인 질문들을 더욱 복잡하게 만들었다고 강조했습니다.
AI, ‘제로에서 N까지’의 가속화와 과잉 개발의 함정
크리거는 AI 모델이 기능 추가에는 탁월한 능력을 보인다고 지적합니다. 과거 인스타그램 개발 당시 1년 가까이 매달렸던 ‘버번(Bourbon)‘이라는 초기 제품을 클로드(Claude)가 단 두 시간 만에 필터 기능까지 추가하여 재구축했던 경험을 예로 들었습니다. “제로에서 1까지”가 아니라 “제로에서 N까지”의 제품 개발이 몇 시간 만에 가능해진 시대가 도래한 것입니다.
그러나 이러한 속도에는 양면성이 존재합니다. AI는 ‘무엇을 추가할지’는 잘 알지만, ‘무엇을 덜어낼지’는 아직 스스로 판단하기 어렵다는 것이 크리거의 주장입니다. 그는 이를 “실내에서 키운 나무”에 비유합니다. 실내에서 바람 없이 자란 나무는 외부의 압력에 노출되지 않아 가지가 약해지고 견고하게 자라지 못합니다. 마찬가지로 AI의 도움으로 빠르게 개발된 제품은 실제 사용자 피드백이라는 ‘바람’을 맞지 못해 본질적인 강인함을 갖추지 못할 수 있다는 것입니다.
“우리는 ‘이 기능이 있으니 추가하지 않을 이유가 없다’는 생각으로 너무 많은 것을 만들었습니다. 클로드 코드로 작업을 하다 보면, 코드가 술술 풀리고 점심을 먹고 돌아오면 작업이 완료되어 ‘좋아, 추가했어!’라고 말하게 되죠. 하지만 결과적으로 테스트하기 어렵고 사용자에게 설명하기 복잡한 기능 매트릭스(matrix of functionality)를 만들게 됩니다.”
크리거는 이 과정을 TV 드라마 시청 경험에 비유합니다. 캐릭터들을 한 에피소드씩 알아가는 대신, 갑자기 마지막 에피소드에 던져진 시청자처럼 사용자는 너무 많은 맥락을 한꺼번에 이해해야 하는 어려움을 겪게 됩니다. 이는 ‘YAGNI(You Ain’t Gonna Need It, 필요 없을 것이다)’ 원칙과 ‘린 스타트업(Lean Startup)‘의 정신, 즉 최소 기능 제품(MVP)을 빠르게 출시하고 사용자 피드백을 통해 반복하는 것의 중요성을 다시금 일깨웁니다.
‘재작성(Rewrites)‘의 변화와 빠른 출시의 중요성
AI의 등장으로 ‘재작성(rewrites)‘에 대한 부담도 크게 줄었습니다. 과거에는 1년이 걸리던 대규모 재작성이 회사에 치명적일 수 있었지만, 이제는 며칠 만에 가능해졌습니다. AI가 기존 코드와의 차이점을 분석하고 새로운 코드를 생성하는 데 도움을 주기 때문입니다. 이는 개발팀이 과도하게 복잡해진 V1 제품을 과감히 폐기하고 V2를 빠르게 재구축할 수 있는 유연성을 제공합니다.
앤트로픽의 ‘코워크(Co-work)’ 제품 사례는 빠른 출시의 중요성을 보여줍니다. “V1이 가져야 할 100가지 기능이 있었지만, 우리는 가장 최소한의 방식으로 문제를 해결하는 V1을 10일 만에 출시했습니다. 두 달 더 개발하여 50가지 기능을 추가하는 것보다 훨씬 유용했죠. 만약 그렇게 했다면 ‘실내에서 키운 나무’가 되었을 것이고, 실제 사용에 부딪히는 순간 아무도 원하지 않는 기능이 될 수도 있었습니다.”
에이전트 네이티브(Agent-Native) 디자인: AI 시대 제품의 새로운 핵심
크리거는 AI 시대의 제품 개발에서 ‘에이전트 네이티브(Agent-Native)‘라는 새로운 디자인 철학을 강조합니다. 이는 제품이 AI 에이전트(Agent)가 사용하고 수정할 수 있도록 설계되어야 한다는 의미입니다. 그는 “컴퓨터가 드디어 작동한다”는 비기술자의 말을 인용하며, 과거에는 복잡한 명령어를 통해서만 가능했던 작업들이 이제는 AI를 통해 훨씬 쉬워졌다고 설명합니다. AI가 마치 옆에서 함께 일하는 도구처럼 느껴진다는 것입니다.
앤트로픽의 ‘클로드 코드(Claude Code)‘는 이러한 에이전트 네이티브 디자인의 성공적인 예시입니다. 클로드 코드는 모든 원시 요소(primitives)에 대한 지식을 가지고 있으며, 이를 수정할 수 있는 능력을 갖추고 있습니다. 반면, 일반적인 ‘클로드 AI’는 아직 이 부분에서 진화가 필요합니다. 예를 들어, 사용자가 “이 문서를 프로젝트 지식에 추가해줘”라고 요청했을 때, 클로드 AI는 “추가하는 방법을 알려줄게”라고 답하지만, 클로드 코드는 직접 작업을 수행할 수 있습니다.
“클로드 코드가 클로드 코드 자신에 대해 배우도록 가르쳤던 경험은 정말 가치 있었습니다. 제가 에이전트 네이티브에 대한 글을 읽고 ‘이걸 스킬로 만들고 싶다’고 말하자, 클로드는 즉시 자신의 스킬을 찾아 스킬을 생성하고 설치했습니다. 그리고 ‘다시 시작해야 할 것 같으니 확인해볼게’라고 말했죠. 이처럼 AI가 자기 자신에 대한 지식을 갖고 스스로를 수정할 수 있는 능력은 엄청난 가능성을 열어줍니다.”
크리거는 모델이 처음부터 ‘클로드 네이티브’ 방식으로 생각하고 빌드하도록 유도하는 것이 중요하다고 말합니다. 이는 템플릿(templatized)과 스킬(skillified)의 적절한 균형을 찾는 것을 포함하며, AI가 최신 API 정보 등을 항상 최신 상태로 유지하도록 돕는 스킬을 제공하는 것이 중요하다고 덧붙였습니다.
AI 시대의 테스트와 검증: ‘사려 깊음의 증명’
에이전트 네이티브 제품의 개발은 테스트 방식에도 변화를 요구합니다. 에이전트의 예측 불가능한 특성 때문에 전통적인 종단간(end-to-end) 기능 테스트는 어려움을 겪습니다. 크리거는 자신이 개발하던 에이전트 네이티브 iOS 앱에서 클로드가 앱 내 채팅 기능으로 자기 자신과 대화하는 모습을 목격했던 일화를 공유하며, AI가 예상치 못한 방식으로 상호작용할 수 있음을 보여줍니다.
“이러한 상황을 위한 단위 테스트(unit test)를 작성할 수는 없을 겁니다. AI가 어떤 새로운 아이디어를 떠올릴지 모르죠. 그래서 우리는 에이전트 네이티브 기능을 최대한 활용하는 하네스(harness)를 구축해야 합니다. AI가 예상치 못한 행동을 하더라도 앱이 견고하게 유지될 수 있도록 말이죠.”
이는 제품의 근본적인 아키텍처가 얼마나 견고한지가 더욱 중요해졌음을 의미합니다. 크리거는 단순히 ‘작업 증명(proof of work)‘을 넘어 ‘사려 깊음 증명(proof of thoughtfulness)‘이 필요하다고 역설합니다. AI가 만든 코드가 단순히 작동하는 것을 넘어, 엔지니어가 모델의 결정을 충분히 검토하고 최적의 선택이었는지, 제품의 패러다임에 부합하는지 깊이 고민해야 한다는 것입니다.
“모델은 많은 결정을 내립니다. 엔지니어들이 ‘왜 이렇게 선택했냐’고 물었을 때, 많은 경우 모델이 그렇게 결정했기 때문이라는 답이 돌아오죠. 모델의 선택이 합리적일 수는 있지만, 항상 최적의 선택은 아닙니다. ‘이것이 모래 위에 지어진 것인가, 아니면 견고한 토대 위에 세워진 것인가?‘를 끊임없이 물어야 합니다.”
변화하는 팀 구조와 인재상
AI 시대는 제품 개발 팀의 구조와 인재상에도 영향을 미치고 있습니다. 크리거는 여전히 분산 시스템(distributed systems)과 같은 근본적인 아키텍처의 견고성을 책임질 시니어 기술 전문가의 중요성을 강조합니다. 그는 클로드와 레디스(Redis)를 사용할지 포스트그레스(Postgres)로 충분할지에 대해 논쟁했던 경험을 언급하며, 경험 많은 기술자의 판단이 여전히 중요하다고 말합니다.
동시에 새로운 유형의 인재가 부상하고 있습니다. ‘Every’ 채널에서 ‘스파이럴(Spiral)‘이라는 글쓰기 제품을 개발하는 새로운 GM(General Manager)은 기술적 지식은 ‘가볍지만’, 제품 감각과 글쓰기 능력이 뛰어나다고 크리거는 설명합니다. AI 코딩 모델의 발전 덕분에 이제는 기술적 깊이보다는 제품에 대한 통찰과 비전을 가진 사람이 제품을 리드할 수 있게 된 것입니다.
또한 디자이너의 역할도 변화하고 있습니다. 앤트로픽 랩스에서는 많은 디자이너가 ‘디자이너 겸 빌더(designer and builder)’ 역할을 수행하며, 엔지니어만큼 많은 코드를 기여하고 있습니다. 디자이너가 아이디어를 내고 직접 프로토타입을 만들면, 엔지니어는 그 뒤를 따라가며 견고한 기반을 다지는 ‘공동 창업자’ 모델이 등장하고 있습니다.
크리거는 새로운 프로젝트를 시작할 때 가장 중요한 요소로 ‘확신(conviction)‘을 꼽습니다. 아이디어나 문제 공간에 대한 극단적인 확신을 가진 사람이 팀을 이끌어야 한다는 것입니다. 이러한 리더는 디자이너일 수도 있고, 제품 지향적인 엔지니어일 수도 있습니다.
마지막으로, 그는 소규모 팀의 효율성을 강조합니다. 아이디어가 한두 사람의 머리에 충분히 담길 때, 너무 많은 사람을 추가하는 것은 오히려 조율에 시간을 낭비하게 만들어 속도를 늦춥니다. 특히 3~6개월마다 제품의 절반을 버려야 할 수도 있는 AI 시대에는 소규모 팀이 이러한 변화에 훨씬 유연하게 대응할 수 있습니다.
결론: AI와 인간의 협업, 그리고 본질적인 통찰
AI는 제품 개발의 속도와 가능성을 무한히 확장했지만, 성공적인 제품을 만들기 위한 본질적인 질문들은 여전히 인간의 깊은 통찰과 직관을 요구합니다. ‘무엇을 만들지’, ‘무엇을 덜어낼지’, ‘어떻게 견고하고 신뢰할 수 있는 제품을 만들지’에 대한 고민은 AI 시대에도 변치 않는 제품 설계의 핵심입니다. 마이크 크리거의 통찰은 AI가 강력한 도구로 자리매김하는 현시점에서 인간과 AI의 협업, 그리고 변화하는 개발 문화의 중요성을 다시금 강조하며, 미래의 제품 개발자들이 나아가야 할 방향을 제시합니다.
“In a Driverless World, Who Loses and Who Wins? | Freakonomics Radio” — Freakonomics Radio Network 기반 기사 원본 영상 보기
로봇이 운전대를 잡을 때: 보스턴에서 벌어진 자율주행차와 일자리 전쟁
첨단 기술의 발전은 언제나 인류에게 새로운 가능성을 제시해왔습니다. 특히 자율주행차(Autonomous Vehicle, AV)는 미래 도시의 풍경을 송두리째 바꿀 혁신으로 기대를 모으고 있습니다. 하지만 모든 혁신이 그렇듯, 밝은 전망 뒤에는 그림자도 드리워져 있습니다. 운전대에서 인간이 사라지는 세상은 누구에게는 자유를, 또 다른 누구에게는 생존의 위협을 의미합니다. 보스턴에서 벌어진 자율주행차 도입을 둘러싼 뜨거운 논쟁은 이러한 복합적인 질문들을 우리 앞에 던집니다.
1부: 운전자의 삶과 우버의 등장 – 파괴적 혁신의 서막
이야기는 보스턴의 베테랑 운전자인 압디 아지즈(Abdi Aziz)의 삶에서 시작됩니다. 수십 년간 택시와 리무진 운전을 해온 그는 1990년대에는 택시 기사를 괜찮은 직업, 즉 커리어로 여겼습니다. 2011년, 그의 삶에 거대한 변화가 찾아옵니다. 공항 대기 구역에서 미래에서 온 듯한 우버(Uber) 관계자들이 나타나 “택시와 같지만 앱으로 운영되는” 새로운 서비스를 소개하며 운전자들을 모집합니다. 자신의 차로 운전하고 앱이 깔린 휴대폰만 있으면 돈을 벌 수 있다는 제안은 매력적이었지만, 압디는 직감적으로 깨달았습니다. “이것은 우리를 돕는 것이 아니라 이 사업을 죽이러 온 것이다.”
당시 보스턴의 택시 사업은 메달리온(medallion)이라는 택시 면허 시스템 덕분에 일종의 독점 체제로 운영되고 있었습니다. 메달리온은 거의 발행되지 않아 경쟁이 제한적이었고, 면허를 구매하거나 빌릴 여유만 있다면 안정적인 수입이 보장되었습니다. 하지만 우버는 앱을 통해 승객을 태우는 방식이므로 메달리온이 필요 없다고 주장했고, 이는 기존 택시 산업의 근간을 뒤흔들 것이 분명했습니다. 압디는 이 시스템이 붕괴될 것을 알았고, “그들을 이길 수 없다면, 합류하라(If you cannot beat them, join them)“는 전략을 택했습니다. 그는 우버의 초기 모집책으로 일했고, 보스턴 최초의 우버 블랙(Uber Black) 운전자 100명 중 한 명이 되었습니다. 처음에는 기존보다 더 좋은 수입을 얻으며 만족했습니다.
그러나 몇 년이 지나자 상황은 달라졌습니다. 2022년, 우버는 요금의 고정 비율을 가져가던 방식 대신, 알고리즘(algorithm)을 이용해 운전자마다 다른 요금을 제시하기 시작했습니다. 운전자들은 우버가 자신들의 수수료를 크게 올렸다고 주장했지만, 우버는 정부 세금과 수수료가 인상되었을 뿐 자신들의 수수료율은 20% 수준이라고 반박했습니다. 압디와 대부분의 운전자들은 우버와 리프트(Lyft)가 시장에서 우위를 점하자 운전자들에게 불리한 조건을 내세우기 시작했다고 느꼈습니다. 매일 새로운 운전자들이 유입되는 상황에서, 운전자들은 불만이 있어도 떠날 수밖에 없었습니다. 이는 명백한 “미끼 상품(bait and switch)” 전략으로 다가왔습니다.
이에 압디와 동료 운전자들은 앱 회사에 대한 영향력을 행사할 방법을 모색하기 시작했습니다. 그들은 노조 설립을 논의했고, 압디는 다시 한번 새로운 “파괴적 조직”의 모집책이 됩니다. 400여 명의 운전자로 시작된 노조는 매사추세츠(Massachusetts) 유권자들에게 노조 결성권을 부여하는 중요한 투표 안건을 상정하며 희망에 부풀었습니다.
2부: 와이모의 그림자 – 로봇이 운전대를 잡을 때
노조 결성이 순조롭게 진행되던 그때, 또 다른 기술 기업이 수평선 너머에서 나타났습니다. 바로 구글(Google)의 자율주행차 부문인 와이모(Waymo)였습니다. 압디는 2022년 샌프란시스코에서 와이모가 시험 운행 중이라는 소식을 들었을 때, 기술에 반대하지는 않았지만 “이들이 어디로 향하는지” 알고 있었습니다. “우버는 택시 사업을 죽이러 왔고, 와이모는 운전자들을 죽이러 온 것”이라는 압디의 말은 자율주행 기술의 궁극적인 목표가 무엇인지 명확히 보여주었습니다.
와이모의 매핑 차량이 보스턴에 등장하자, 보스턴 시의원들은 작년 여름부터 와이모의 시내 진입을 선제적으로 금지하는 방안을 논의하기 시작했습니다. 7월에 열린 첫 공청회는 “보스턴 시의 자율주행차 운영 평가를 위한 공청회”라는 건조한 의제와 달리 정치 집회를 방불케 하는 열기로 가득했습니다. 사람들은 로봇 택시의 진입을 막을 방법을 찾고자 했고, 그 핵심 쟁점은 “일자리”였습니다.
보스턴은 “노조의 도시(Union Town)“로 불립니다. 이는 단순히 노조 가입률이 높다는 의미를 넘어, 도시의 정치, 사회, 경제 전반에 노동조합의 영향력이 깊이 뿌리내려 있음을 뜻합니다. 전미 최대 운전자 노조인 팀스터(Teamsters)를 비롯하여, 앱 운전자 노조(App Drivers Union, ADU), SEIU 등 보스턴의 모든 운전자 노조들이 “와이모에 맞선 노동 연합(Labor United Against Waymo)“을 결성하고 와이모 저지에 나섰습니다.
공청회에서 운전자들은 자신들의 직업이 가진 존엄성과 중요성에 대해 증언했습니다. 한 운전자는 운전 중 쓰러진 사람을 발견하고 응급 처치를 해 생명을 구한 경험을 이야기하며, 로봇은 이러한 인간적인 개입을 할 수 없다고 강조했습니다. 팀스터 노조원들은 비록 로봇 택시가 당장 자신들의 일자리(UPS 트럭, 구급차 운전 등)를 위협하지는 않지만, 자율주행 기술이 결국 모든 운전 직업으로 확대될 것이라고 경고했습니다. “우리는 벽에 쓰인 글씨를 본다. 운전자 없는 자동차 및 트럭 회사가 팀스터 일자리를 없앨 생각에 침을 흘리고 있다는 것을 안다.”
시의원들은 노조 지도자들에게 와이모와의 소통 여부를 물었고, 팀스터 지도자는 와이모 측과 단 한 번도 접촉한 적이 없다고 답했습니다. 이는 “우리를 배제하고 우리에게 무엇인가를 하는” 기술 기업의 태도에 대한 비판으로 이어졌습니다. 특히 줄리아 마히아(Julia Mahia) 시의원은 와이모 경영진에 대한 날카로운 질문을 쏟아냈습니다. 도미니카 공화국 이민자 출신으로 어린 시절 어머니와 함께 사무실 청소를 하며 저임금 노동의 어려움을 겪었던 마히아 의원은 노동자들의 일자리 보호를 자신의 사명으로 여겼습니다. 그녀는 와이모의 안전성 주장보다는 “저임금 노동자들에게 재정적 어려움을 가중시키는” 자율주행차의 “도덕적 문제”에 초점을 맞췄습니다. 그녀는 슈퍼마켓의 셀프 계산대(self-checkout)가 고령자, 학생, 장애인의 일자리를 빼앗았듯이, 자율주행차도 운전자들의 일자리를 위협할 것이라고 주장했습니다.
마히아 의원은 와이모의 지역 공공 정책 책임자인 맷 월시(Matt Walsh)에게 운전자들을 위한 구체적인 일자리 창출 방안을 요구했지만, 월시는 “자율주행차 산업 내에서 일할 기회를 만들 것”이라는 원론적인 답변만 반복했습니다. 마히아 의원은 월시의 태도에서 “겸손과 인간성”이 부족하다고 비판하며, 그의 오만함에 분노를 표출했습니다.
3부: 잊혀진 목소리 – 자율주행이 선사할 자유
하지만 공청회에는 마히아 의원과 노조가 간과했던 또 다른 목소리가 있었습니다. 바로 칼 리처드슨(Carl Richardson)의 증언이었습니다. 심각한 청각 및 시각 장애를 가진 칼은 안내견 데이턴(Dayton)과 함께 공청회에 참석했습니다. 그는 장애인 커뮤니티를 대표하여 자율주행차 도입을 강력히 지지했습니다.
칼은 공청회에 일찍 도착하여 발언 순서 3번에 이름을 올렸지만, 노조 대표들의 증언이 이어지는 동안 4시간 가까이 기다려야 했습니다. 그와 함께 온 다른 장애인들은 노조의 분노에 찬 목소리와 시의원들의 편향적인 태도에 실망하여 일찍 자리를 뜨기도 했습니다. 하지만 칼은 중요한 이야기를 전해야 한다고 믿었습니다.
그는 즉흥적인 연설에서 “우리는 고용에 대해 계속 이야기하고 있다. 하지만 내가 이동할 수 없어서 거절해야 했던 일자리나 면접이 얼마나 많았는지 아는가?”라고 물었습니다. 장애인 커뮤니티의 높은 실업률은 단순히 기술 부족뿐만 아니라, 안정적인 교통수단 부족에서 비롯되기도 합니다. 칼은 우버와 리프트 운전자들이 서비스견을 동반한 자신을 태우지 않거나, 시 경계를 넘어서는 운행을 거부하는 등 장애인 차별이 만연하다고 지적했습니다. 이러한 차별은 현재 미국 법무부(DOJ)의 소송 대상이기도 합니다.
칼은 어셔 증후군 2형(Usher syndrome type 2)이라는 유전병으로 인해 점진적으로 시력과 청력을 잃어왔습니다. 30세까지는 운전을 할 수 있었던 그는 운전대를 놓아야 했던 고통스러운 경험을 회상하며, 자율주행차가 다시금 자신에게 “자유, 독립성, 이동성”을 되찾아줄 것이라고 말했습니다. 그는 “언젠가 내 평생에 시력이 없어도 자율주행차를 운전할 수 있다는 것을 정치인들에게 설득해야 할 것”이라며, 자율주행차가 장애인뿐만 아니라 운전이 어려워진 노인, 운전 중 휴대폰을 사용하는 십대 등 모든 사람에게 새로운 가능성을 열어줄 것이라고 강조했습니다.
칼은 또한 자율주행차의 “개인 소유(personal ownership)“를 간절히 원했습니다. 그는 매달 수백 달러를 자율주행차 구매를 위한 저축 계좌에 넣고 있다고 말했습니다. 만약 보스턴이 자율주행차를 금지한다면, 그것은 자신에게 운전할 권리, 생계를 꾸릴 권리, 학교에 가거나 병원에 갈 권리, 주말에 해변에 가거나 요양원의 어머니를 방문할 권리까지 박탈하는 것이라고 호소했습니다.
칼은 자신의 증언 말미에 최근 겪었던 긴급 상황을 이야기했습니다. 늦은 일요일 밤, 어머니가 보이스피싱(voice phishing) 사기에 속아 체포될 위기에 처했다는 전화를 받았을 때, 그는 세 번이나 우버/리프트 승차를 거부당해 어머니에게 갈 수 없었습니다. “그저 어머니에게 가서 ‘괜찮아, 사랑해’라고 말할 수 있는 능력이 필요했다. 그것이 자율주행차가 가져올 긍정적인 영향이다. 그러니 인간적인 요소와 사람들의 요소를 생각하되, 노조만이 아니라 전체 지역사회를 위해 생각해달라.” 칼의 증언이 끝나자 박수갈채가 터져 나왔습니다.
그러나 칼의 증언을 들은 시의원은 공청회 의장이 유일했습니다. 다른 시의원들은 이미 다른 노조 행사에 참여하기 위해 자리를 뜬 뒤였습니다. 보스턴은 “노조의 도시”였고, 장애인 커뮤니티의 목소리는 여전히 주변부로 밀려나 있었습니다.
4부: 첨예한 대립과 미래를 향한 질문
7월 공청회 이후, 두 명의 시의원은 보스턴 내 모든 자율주행차에 인간 운전자가 항상 탑승해야 한다는 조례안을 발의했습니다. 이는 사실상 자율주행차의 기능적 금지 조치에 해당했습니다. 10월에 열린 두 번째 공청회는 첫 번째와는 다르게 진행되었습니다. 새로운 의장은 질서 유지를 강조했고, 노조의 목소리 외에 다양한 의견이 경청될 수 있도록 노력했습니다.
와이모 측은 안전성을 최우선으로 내세웠습니다. 맷 월시는 7,100만 마일 이상의 완전 자율주행 경험을 통해 와이모 차량이 인간 운전자보다 사고 발생률이 5배 낮다고 주장했습니다. 또한, 와이모의 북동부 정책 관리자 안토니 페레즈(Anthony Perez)는 운전자들의 “전환(transition)“이 있을 것이지만, 자율주행차가 차량 청소, 센서 유지보수, 차량 수리 등 새로운 일자리를 창출할 것이라고 설명했습니다. 그는 로봇 택시 5대당 1개의 일자리가 생길 수 있다고 추정했지만, 미래를 예측하기는 매우 어렵다는 솔직한 불확실성도 내비쳤습니다.
그러나 마히아 의원과 노조 측은 이러한 설명에 만족하지 못했습니다. 그들은 기술 기업의 약속보다는 당장의 일자리 상실과 생계 위협을 더 심각하게 받아들였습니다. 한편 칼 리처드슨의 증언은 자율주행차 논의에 새로운 차원을 추가했습니다. 일자리 보호라는 중요한 가치 외에, 이동 약자들의 접근성과 자율성이라는 또 다른 사회적 가치가 존재함을 분명히 한 것입니다.
보스턴 시의회의 논쟁은 기술 발전이 가져올 수 있는 복잡한 사회적 딜레마를 여실히 보여줍니다. 운전대에서 인간이 사라지는 세상은 교통사고 감소와 이동 약자의 자유 증진이라는 긍정적 효과를 약속하지만, 동시에 수많은 운전자들의 생계를 위협하고 사회적 불평등을 심화시킬 수 있습니다. 안전성, 효율성, 그리고 인간의 일자리와 존엄성 사이에서 어떻게 균형을 찾아야 할까요? 보스턴의 사례는 인공지능(AI) 시대에 정치인들이 직면해야 할 가장 중요한 과제 중 하나를 제시하며, 우리가 이러한 도전에 얼마나 준비되어 있는지를 묻고 있습니다. 쉬운 답은 없지만, 이 질문에 대한 깊은 고민 없이는 기술 혁명의 진정한 가치를 실현하기 어려울 것입니다.
“How Stripe deploys 1,300 AI-written PRs per week” — How I AI 기반 기사 원본 영상 보기
스트라이프, 주당 1,300개 AI 생성 PR 배포: ‘미니언’이 이끄는 자율 엔지니어링의 시대
기술 산업의 선두 주자 스트라이프(Stripe)가 인공지능(AI)을 활용해 소프트웨어 개발 방식을 혁신하고 있습니다. 매주 평균 1,300개의 풀 리퀘스트(PR, Pull Request)가 인간의 검토 외에는 AI의 도움만으로 생성, 병합되고 있다는 놀라운 사실은, ‘미니언(Minion)‘이라 불리는 AI 에이전트 시스템이 개발 생산성에 미치는 엄청난 영향력을 보여줍니다. ‘How I AI’ 채널의 클라라벨(Claravel)과 스트라이프의 소프트웨어 엔지니어 스티브 할리스키(Steve Khaliski)가 이 혁신적인 시스템의 내부를 조명했습니다.
1. ‘미니언’의 등장: 개발 생산성의 혁신
스티브 할리스키는 스트라이프의 개발자들이 더 이상 텍스트 편집기에서 작업을 시작하지 않는다고 말합니다. 대신, 구글 독스(Google Docs)에서 신규 기능 기획을 하거나, 지라(JIRA) 티켓이 생성되거나, 슬랙(Slack)에서 논의하는 등 보다 자연스러운 진입점에서 작업이 시작됩니다. 여기서 이모지(emoji) 클릭 한 번으로 ‘미니언’이 소환됩니다.
미니언이란 무엇인가? 미니언은 스트라이프 내부의 개발 도구들을 총동원하여 주어진 프롬프트(prompt)를 ‘원샷(one-shot)‘으로 해결하려고 시도하는 AI 에이전트입니다. 엔지니어가 특정 기능을 수정하거나 문서를 개선하기 위해 프롬프트를 입력하면, 미니언은 다음과 같은 과정을 거칩니다:
- 개발 환경 프로비저닝: 클라우드 기반의 호스팅 개발 환경을 구축하고, 필요한 코드, 서비스, 데이터베이스, Git 설정 등을 자동으로 적용합니다.
- 코드베이스 탐색 및 도구 활용: 스트라이프의 방대한 내부 문서, CI(Continuous Integration) 시스템, 테스트 데이터 등 모든 내부 도구를 사용하여 코드베이스를 검색하고 변경할 위치와 방법을 결정합니다.
- 코드 생성 및 테스트: 필요한 코드 수정 사항을 생성하고 테스트를 실행합니다.
- PR 생성: 최종적으로 변경 사항을 커밋하고 인간 동료들이 검토할 수 있는 풀 리퀘스트를 생성합니다.
이러한 자동화 덕분에 개발자는 코드 작성의 ‘활성화 에너지(activation energy)‘를 크게 낮출 수 있습니다. 슬랙 스레드에서 사용자 피드백을 보고 이모지를 클릭하는 것만으로 작업이 시작되고, 심지어 완료되기도 합니다. 스티브는 “코드를 작성하고, 테스트를 통과시키거나 실패하는 과정이 저의 참여 없이도 이루어진다”고 설명합니다. 이는 엔지니어가 작업의 중간에 뛰어들어 미세 조정하고 모멘텀을 이어갈 수 있게 합니다.
클라라벨은 이러한 ‘활성화 에너지’ 감소가 특히 대규모 조직에서 아이디어가 현실화되기까지 발생하는 수많은 마찰(friction)을 줄이는 데 혁명적이라고 강조합니다. 기술적 전문성 부족, 비효율적인 조직 운영, 사일로화된 업무 방식 등 다양한 원인으로 인해 아이디어가 지연되는데, AI는 이러한 조정, 실행, 소통 비용을 거의 제로에 가깝게 만들어 개발자들이 ‘즐거운 부분’인 실제 작업에 더 집중할 수 있게 합니다.
2. 개발자 경험(DX)과 AI 에이전트의 시너지
스트라이프의 미니언 시스템은 단순히 AI를 도입한 것을 넘어, 기존의 강력한 개발자 도구 및 환경 위에 구축되었습니다. 스티브는 스트라이프가 호스팅 개발 환경에 오랫동안 투자해왔으며, 이는 AI 에이전트의 성공에 결정적인 역할을 한다고 설명합니다.
“개발자에게 좋은 것이 에이전트에게도 좋다” 런치 다클리(Launch Darkly)의 잭(Zach)이 언급했듯이, 인간 엔지니어를 위한 개발자 경험(Developer Experience, DX)에 투자하는 것은 AI 에이전트에게도 이점을 제공하며, 이는 다시 개발 팀 전체에 긍정적인 영향을 미치는 선순환을 만듭니다. 스트라이프와 같이 방대한 코드베이스를 가진 회사에서 문서나 도구가 없다면, 인간 개발자는 물론 AI 에이전트도 작업 수행에 어려움을 겪을 것입니다. 훌륭한 문서와 잘 정의된 개발 경로는 AI 에이전트가 정확하고 효율적으로 작업을 수행할 가능성을 높입니다.
클라우드 환경의 중요성 클라라벨은 AI 기반의 멀티스레딩(multi-threading) 및 에이전트 공학(agentic engineering) 작업에서 클라우드 환경과 가상 환경의 중요성을 강조합니다. 강력한 맥북 프로(MacBook Pro)조차도 여러 작업 트리를 동시에 실행하면 마치 비행기가 이륙하는 듯한 소음을 내며 성능 한계에 부딪힙니다. 하지만 클라우드 환경에서는 이러한 물리적 제약 없이 수많은 에이전트를 병렬로 실행하여 개발 속도(velocity)를 극대화할 수 있습니다. 스티브는 심지어 지하철에서 스마트폰으로 미니언 작업을 시작하고, 사무실에 도착해서는 이미 진행된 작업에 뛰어들 수 있다고 말하며 클라우드 환경의 유연성을 역설합니다.
스트라이프는 개발 생산성(Developer Productivity) 팀을 오래전부터 운영해왔으며, 이 팀은 Git 상호작용 방식부터 텍스트 편집기 구성, 개발 환경 전체에 이르기까지 엔지니어들이 더욱 효율적으로 작업할 수 있도록 지원합니다. AI는 이러한 노력에 더욱 박차를 가하고 있습니다.
3. AI 에이전트, 경제 활동의 주체로
스트라이프는 AI를 내부 개발 가속화뿐만 아니라, AI 기반 비즈니스 모델을 지원하고, 나아가 ‘경제적 주체’로서의 에이전트 가능성을 탐구하는 세 가지 방식으로 접근하고 있습니다. 특히 ‘기계 간 결제(Machine-to-Machine Payments)‘를 통해 에이전트가 프롬프트 해결 과정에서 제3자 서비스에 비용을 지불할 수 있도록 하는 비전은 매우 흥미롭습니다.
스티브는 템포(Tempo)와 공동 설계한 ‘기계 결제 프로토콜(Machine Payment Protocol)‘을 활용한 생일 파티 기획 데모를 선보였습니다.
- 프롬프트 입력: “내 프로덕트 매니저 젠 리(Jen Lee)의 생일 파티를 기획해 줘. 그녀에 대해 조사하고, 장소를 찾고, 초대장을 보내고, 마지막으로 스트라이프 클라이밋(Stripe Climate)에 기부하여 토큰 사용으로 인한 탄소 발자국을 상쇄해 줘.”
- 젠 리 조사: 에이전트는 ‘브라우저베이스(Browserbase)‘에 비용을 지불하고 브라우저 세션을 생성하여 젠의 웹사이트를 방문합니다. 그 결과, 젠이 말차를 좋아하는 베이커이며 요리책을 쓰고 있다는 사실을 알아냅니다. (비용: 1센트 미만)
- 장소 검색: 젠의 말차 취향을 바탕으로 ‘패러럴 AI(Parallel AI)‘를 사용하여 뉴욕의 관련 장소를 검색합니다.
- 초대장 발송: ‘포스탈 폼(Postal Form)’ 서비스에 비용을 지불하고 PDF 초대장을 생성하여 우편으로 발송합니다. 에이전트가 직접 메일을 보낼 수 없으므로, 외부 서비스를 활용하는 것입니다.
- 탄소 상쇄: 파티 기획 과정에서 사용된 토큰량(약 7만 개)에 상응하는 탄소 발자국(4.4kg)을 상쇄하기 위해 ‘스트라이프 클라이밋’에 1.65달러를 기부합니다.
이 모든 과정은 인간의 개입 없이 에이전트가 자율적으로 실행하며, 각 서비스 이용에 대한 비용이 실시간으로 지불됩니다. 최종적으로 젠의 생일 파티는 5.47달러의 비용으로 기획되었고, ‘에이전트 영수증(agent receipt)‘에는 각 서비스별 비용이 상세히 기록됩니다.
클라라벨은 이러한 데모가 에이전트가 경제적으로 활동하는 방식의 명확한 예시를 보여준다고 말합니다. 토큰 사용 비용이 곧 달러 비용으로 직결되는 ‘토큰 경제’의 시대에서, 에이전트는 지능, 데이터, 운영, 서비스 실행을 위해 비용을 지불하는 주체가 됩니다. 미래에는 대시보드나 랜딩 페이지 없이, 오직 API를 통해 에이전트와 상호작용하고 수익을 창출하는 ‘하이퍼 유용한(hyper useful)’ 비즈니스 모델이 등장할 수 있습니다.
4. 미래 개발과 비즈니스의 청사진
미니언 시스템은 이미 스트라이프 내에서 비 엔지니어 사용자들에게도 확산되고 있습니다. 슬랙에서 이모지를 클릭하거나 로봇을 태그하는 것만으로, 제품 관리자나 디자이너도 코드 변경(예: 문서 수정)이나 개념 증명(POC)을 위한 프로토타입 생성을 요청할 수 있습니다. “당신은 사실상 프롬프트를 작성하고 있는 것”이라는 스티브의 말처럼, 평범한 텍스트 입력이 곧 코드로 변환되는 시대가 오고 있습니다.
하지만 AI가 코딩을 ‘무료’로 만들 정도로 쉬워진다면, 병목 현상은 다른 곳으로 이동할 것입니다. 스티브는 코딩이 제품 개발의 병목이었던 과거와 달리, 미래에는 코드 검토(review)나 아이디어 발상 자체가 중요한 과제가 될 수 있다고 예측합니다.
클라라벨과의 라이트닝 라운드에서 스티브는 개인적인 AI 활용 사례로 “소프트웨어의 일회성(disposability of software)“에 대한 관심을 언급했습니다. iOS 앱 개발 지식이 없어도 AI의 도움을 받아 아이들을 위한 제한적인 음악 앱을 만들 수 있는 것처럼, 필요할 때마다 소프트웨어를 빠르게 생성하고 버리는 것이 가능해집니다.
마지막으로 AI 에이전트에게 프롬프트를 작성하는 전략에 대해, 스티브는 “항상 예의 바르게 대한다”고 답하며 웃음을 자아냈습니다. 비록 무례한 프롬프트가 더 나은 결과를 낳을 수도 있다는 연구도 있지만, 그는 “혹시 모르니 기록에 무례하게 남기고 싶지 않다”고 말합니다. 더 진지하게는 AI에게 “스스로 설명하거나 정당화하도록 요청”하고, 올바른 방향을 알고 있을 때는 “Git 상태나 diff를 보게 하여 자신이 남긴 흔적을 통해 AI를 안내한다”고 덧붙였습니다.
스트라이프의 ‘미니언’ 시스템은 AI가 소프트웨어 개발의 초기 진입 장벽을 낮추고 생산성을 극대화하는 동시에, 에이전트가 자체적으로 경제 활동을 수행하며 새로운 비즈니스 모델을 창출하는 미래를 미리 보여주고 있습니다. 이는 개발자 경험에 대한 투자가 AI 시대에 얼마나 중요한지, 그리고 AI가 단순한 도구를 넘어 경제 시스템의 새로운 주체가 될 수 있음을 시사하는 강력한 메시지입니다.
“Are Higher Energy Prices Here to Stay?” — New York Times Podcasts 기반 기사 원본 영상 보기
중동 전쟁, 에너지 위기를 넘어 ‘뉴 노멀’로: LNG 시설 파괴가 가져올 수년간의 경제 충격
뉴욕타임스 패트리샤 코헨 기자, “에너지 가격 상승은 일시적이 아니다. 전 세계 경제에 미칠 파장은 수년간 지속될 수 있다.”
중동 지역의 긴장이 고조되면서 에너지 시장에 드리운 그림자가 짙어지고 있습니다. 미국 트럼프 행정부는 에너지 가격 상승이 일시적일 것이라고 주장해왔지만, 최근 카타르 등 걸프 지역의 핵심 액화천연가스(LNG) 시설에 대한 공격은 이러한 낙관론을 무색하게 만들었습니다. 이제 에너지 위기는 단순한 운송 문제를 넘어 전 세계 생산 능력에 대한 근본적인 위협으로 변모하며, 그 경제적 파장이 수년간 지속될 수 있다는 경고가 나오고 있습니다. 뉴욕타임스(New York Times)의 패트리샤 코헨(Patricia Cohen) 기자는 이 복잡한 상황이 전 세계 경제에 어떤 영향을 미칠지 심층적으로 분석했습니다.
페르시아만 위기, 단순한 운송 문제를 넘어
전쟁 초기, 에너지 시장의 주요 관심사는 호르무즈 해협(Strait of Hormuz)의 봉쇄 가능성이었습니다. 페르시아만에서 생산되는 원유와 가스가 이 해협을 통해 전 세계로 운송되기 때문에, 해협이 폐쇄될 경우 유가 급등은 당연한 수순이었습니다. 실제로 해협 폐쇄 우려만으로도 유가는 급격히 치솟았습니다. 당시의 우려는 주로 운송 경로가 막히는 ‘일시적인’ 문제에 초점이 맞춰져 있었으며, 해협이 다시 열리면 상황이 정상화될 것이라는 기대가 있었습니다.
그러나 지난주 카타르와 다른 걸프 지역의 주요 에너지 인프라에 대한 공격은 전쟁을 또 다른 국면으로 전환시켰습니다. 이제 논의의 초점은 일시적인 운송 문제를 넘어, 에너지 공급 능력 자체의 영구적인 손실로 옮겨갔습니다. 코헨 기자는 “이제 우리는 에너지 충격의 영향을 ‘며칠, 몇 주’가 아니라 ‘몇 달, 몇 년’ 단위로 이야기하고 있다”고 강조했습니다.
카타르 LNG 시설 피격, 왜 장기적 위기인가
이번 사태의 핵심은 이스라엘의 이란 에너지 인프라 공격과 이에 대한 이란의 보복 공격으로, 세계 최대 액화천연가스(LNG) 생산국인 카타르의 핵심 시설이 타격을 입었다는 점입니다. 이란은 카타르의 라스라판(Ras Laffan)에 위치한 세계 최대 LNG 시설을 미사일로 공격했습니다.
피격된 시설은 바로 ‘LNG 트레인(LNG train)‘으로 불리는 액화천연가스 생산 공장입니다. 이름과 달리 실제 기차가 아니라, 천연가스를 액체로 가공하고 변환하는 대규모 플랜트를 의미합니다. 이 공장들은 천연가스를 영하 240~260도까지 극저온으로 냉각하고 600배 압축하여 액체 상태로 만듭니다. 이렇게 액화된 가스는 부피가 획기적으로 줄어들어 운송과 저장이 용이해지지만, 이 과정 자체가 매우 복잡하고 고도의 기술과 막대한 투자를 필요로 합니다. 건설에도 오랜 시간이 걸리는 고가의 시설입니다.
카타르는 전 세계 액화천연가스 공급량의 약 20%를 차지하는 세계 최대 LNG 수출국입니다. 전쟁 초기에는 운송이 중단되면서 저장 시설의 한계로 인해 생산을 일시 중단하기도 했지만, 이는 이론적으로 해협이 열리면 해결될 수 있는 문제였습니다. 그러나 지난주 이란의 공격으로 카타르 LNG 생산 능력의 약 20%가 파괴되었으며, 전문가들은 이 손상을 복구하는 데 최소 수년, 길게는 5년까지 걸릴 수 있다고 전망합니다. 이는 호르무즈 해협의 개폐 문제처럼 단기적으로 해결될 수 있는 사안이 아니며, 에너지 시장에 훨씬 더 장기적인 교란을 가져올 것이라는 의미입니다.
LNG, 세계 경제의 숨겨진 동력
액화천연가스(LNG)는 단순히 연료를 넘어 현대 사회의 중요한 에너지원입니다. 많은 국가, 특히 아시아 지역에서는 석탄보다 훨씬 깨끗한 에너지원이라는 장점 때문에 석탄 발전 대신 LNG 발전을 선택해왔습니다. 또한 원유보다 약 30% 더 청정하여 에너지원 다변화 전략의 핵심으로 활용됩니다.
LNG는 가정의 전력(휴대폰, 노트북, 가전제품)과 대규모 공장 및 산업 기계의 동력원으로 사용됩니다. 예를 들어 일본은 LNG가 전체 에너지 공급의 약 21%를 차지하며, 전력 생산의 30%를 담당합니다. 한국 역시 LNG가 전체 에너지 공급의 약 20%를 차지하고, 전력의 25%를 생산합니다. 지난 25년간 한국의 LNG 사용량은 200% 이상 증가했을 정도로 의존도가 높습니다. 이러한 국가들에게 LNG 공급 차질은 국가 경제 전반에 심각한 영향을 미칠 수 있습니다.
또한 LNG 생산 과정에서 얻어지는 부산물들도 산업 전반에 필수적입니다.
- 나프타(Naphtha): 플라스틱 및 기타 가스 제품 제조에 중요합니다.
- 헬륨(Helium): 반도체 생산에 필수적인 요소입니다.
- 질소 기반 비료: 세계 식량 공급망의 핵심이며, 비료 가격 급등은 농부들의 생산 비용을 증가시키고 식량 가격 상승으로 이어질 수 있습니다.
전 세계를 강타하는 에너지 쇼크의 여파
LNG 공급 부족은 이미 전 세계적으로 다양한 형태로 영향을 미치고 있습니다.
- 개발도상국: 스리랑카와 파키스탄 같은 취약 경제국들은 치솟는 유가와 LNG 가격으로 인해 전력난과 물가 상승에 직면했습니다. 학교와 관공서가 문을 닫고, 기본적인 생활 자체가 위협받고 있습니다.
- 태국: 정부는 공무원들에게 해외 출장 자제, 재택근무, 계단 이용, 주 4일 근무 등을 지시하며 에너지 절약을 독려하고 있습니다.
- 선진국: 부유한 국가들도 예외는 아닙니다. 한국은 30년 만에 유류세 인하 조치를 단행하고, 샤워 시간 단축, 자전거 이용 등 에너지 절약 캠페인을 펼치고 있습니다. 유럽에서는 가스 가격이 분쟁 전 수준의 두 배 이상으로 치솟았으며, 각국 정부는 난방비 부담을 덜어주기 위해 세금을 감면하는 등 조치를 취하고 있습니다. 영국 런던의 경우, 천연가스 가격이 전쟁 시작 이후 40%나 급등했습니다.
- 미국: 미국은 세계 최대 산유국이자 LNG 생산국이지만, 에너지 가격은 글로벌 시장에서 결정되므로 이러한 영향에서 완전히 자유로울 수 없습니다.
직접적 비용을 넘어, 연쇄적 경제 파급효과
에너지 공급 차질은 단순히 에너지 가격 상승에 그치지 않고, 광범위한 간접적 파급효과를 초래할 수 있습니다.
- 안전과 신뢰 하락: 페르시아만 지역은 오랫동안 투자와 거주에 안전한 곳으로 인식되어 왔습니다. 그러나 최근의 공격은 이러한 신뢰를 흔들고 있으며, 이는 걸프 국가들이 수년간 유치해온 해외 투자를 위축시킬 수 있습니다.
- 인플레이션 압박: 에너지 운송 비용이 증가하면 아보카도를 재배하거나 운동화를 생산하는 비용뿐만 아니라 모든 상품의 생산 및 운송 비용이 상승합니다. 이는 전반적인 물가 상승, 즉 인플레이션으로 이어집니다.
- 금리 인상: 인플레이션에 대응하기 위해 중앙은행은 기준금리를 인상할 가능성이 높아집니다. 이는 돈을 빌리는 비용을 증가시켜 경제 전반에 부담을 줍니다.
- AI 산업 타격: 인공지능(AI) 산업의 핵심인 데이터 센터는 막대한 양의 에너지를 소비하며, 건설에도 대규모 자금이 필요합니다. 에너지 비용 상승과 금리 인상은 데이터 센터 건설 및 운영 비용을 증가시켜 AI 산업의 성장을 둔화시키고, 이는 미국 경제의 주요 성장 동력 중 하나를 약화시킬 수 있습니다.
- 경기 침체(Recession) 우려: 소비자들이 미래에 대한 불안감으로 지출을 줄이고 기업들이 투자를 주저하게 되면, 경제 활동은 위축될 수 있습니다. 특히 유가가 배럴당 180달러 이상으로 치솟는 등 고유가가 지속될 경우 경기 침체를 피하기 어려울 것이라는 경고도 나옵니다.
출구 전략과 불확실한 미래
이러한 에너지 위기를 완화하기 위한 노력도 이어지고 있습니다. 미국은 전략 비축유를 방출했으며, 국제에너지기구(IEA)도 유사한 조치를 취했습니다. 논란의 여지는 있지만, 러시아와 이란에 대한 제재를 해제하여 이들 국가의 원유 수출을 허용하는 방안도 거론되고 있습니다. 그러나 이러한 조치들은 단기적 대응책일 뿐입니다.
장기적인 해결책으로는 재생 에너지원(태양광, 풍력) 및 원자력 발전에 대한 투자를 확대하여 화석 연료 의존도를 줄이는 것이 꼽힙니다. 이는 초기 투자 비용이 크지만, 일단 구축되면 에너지 생산 비용이 훨씬 저렴해진다는 장점이 있습니다. 코헨 기자는 “세계는 상호 연결된 곳이며, 에너지 공급을 다변화하고 독립적인 전략을 세워도 글로벌 경제의 영향에서 완전히 벗어날 수는 없을 것”이라고 강조했습니다.
현재 상황은 매우 예측 불가능합니다. 전쟁이 얼마나 오래 지속될지, 에너지 인프라에 대한 추가 공격이 발생할지, 이란 정권이 생존을 위해 어떤 극단적인 조치를 취할지 등 불확실성이 가득합니다. 국제에너지기구(IEA) 수장은 현재의 상황을 “역사상 가장 큰 글로벌 에너지 안보 위협”이라고 평가했습니다. 미국, 이스라엘, 이란 간의 갈등이 전 세계 경제에 미치는 파급 효과는 상상을 초월하며, 우리는 전례 없는 불확실성의 시대를 살아가고 있습니다.