YouTube Digest
Listen to this issue
English
Based on ""50 AI Agents Running My Company” Is a Lie. Here’s How I Build It | Gumloop, Max Brodeur-Urbas” from EO Watch the original video
The AI Illusion: Why ‘50 Agents Running My Company’ is a Myth – And How Real Automation is Built
In the glittering landscape of modern entrepreneurship, a pervasive myth has taken root: the idea that success, especially with AI, can be achieved with a single click, an effortless side hustle generating millions. Social media abounds with gurus promising “50 AI agents running my company” or a “suite of AI that tells me what to do every day,” all while working a mere “one hour a week.” Max Brodeur-Urbas, the founder and CEO of Gumloop, calls this out for what it is: “marketing,” a “lie,” and a dangerous “anti-pattern.”
“They’re selling hope really easily,” Max asserts, “this vision of skipping the hard work, shortcutting directly to the value, which will never happen.” While these “course bros” might print money selling the dream, Max and his team at Gumloop are building a different reality. Their platform automates over 4 million workflows daily for giants like Instacart, Shopify, DoorDash, and Gusto, empowering marketers, salespeople, and operations personnel to automate their own work without needing an engineering team. This is not about replacing humans with AI; it’s about amplifying human potential.
From Big Tech Disillusionment to Forced Entrepreneurship
Max’s journey to building Gumloop is far from the “effortless” narrative peddled online. A software engineering graduate from McGill, he initially chased the perceived stability of Big Tech. “My goal all throughout college was just get a good job,” he recalls, “I wanted to work in big tech ‘cause I thought that would be the right path to follow.” But the reality of Microsoft quickly revealed a profound disillusionment.
“I hated it,” he states plainly. The common advice to learn at a big company before striking out on your own is, in his view, “kind of cope.” Many end up in “golden handcuffs,” never leaving. Max found little transferable knowledge from his Big Tech stint to his startup, save for a “default respect” that a corporate logo might confer. Instead, his entrepreneurial drive was fueled by a desire to do the exact opposite of what he experienced. He saw the “really, really important years” of one’s early twenties being “wasted” in a 9-to-5 grind. “If we’re going to do it, why don’t we just do it now?” he thought.
The decisive push, however, came from an unexpected and dramatic turn of events: deportation. After quitting Microsoft and moving back to Vancouver to build things in his bedroom, a weekend trip to Seattle to visit old roommates resulted in a U.S. border official deeming him suspicious. Despite his intent for a two-day stay, he was turned around and slapped with a five-year ban from the country. “That was kind of the moment where I realized I had to build a company because I had no fallback plan,” Max recounts, describing the terror and shock that quickly morphed into intense focus.
The Grind: Hunting to Prove Yourself Wrong
For the next six months, Max worked relentlessly, throwing himself into “the ether” of experimentation. He built numerous MVPs: video game moderation software in VR, general trust and safety tooling, bot detection software, an anti-cam platform. He would build, try to sell, and learn from the market’s response – often negative.
“Over and over I tried that, kind of like once a week,” he explains. This rapid iteration taught him a counterintuitive but vital startup lesson: “You’re actually chasing proving yourself wrong.” In the beginning, he wasted months hoping someone would validate his ideas. He quickly realized the opposite was true: “You should actually be hunting for someone to tell you why this won’t work.” If you can’t find a strong reason why an idea won’t work, then you might be onto something. “I don’t think I would have ended up building Gumloop if I didn’t fail 10 times before that,” he admits.
This relentless pursuit of feedback extended to users, a privilege he believes founders must earn. Early on, it means “begging people to try your product.” But the most valuable feedback, even if it’s “your product sucks” or “it’s not solving my problem,” must be put at the forefront.
The “Aha!” Moment: From Agents to Automation
Gumloop’s origin story is deeply rooted in Max’s keen observation of the early AI hype cycle. He saw AutoGPT, a popular open-source agent framework, take Twitter by storm. It promised AI’s ability to act “agentically” and solve problems autonomously. Max, intrigued, joined its rapidly growing Discord server.
What he found was a community of enthusiastic users, many of whom were struggling with basic technical concepts: “What is GitHub? How do I use a terminal? How do I install something locally?” Max saw a simple problem to solve: build a user-friendly UI to simplify the setup. This became AgentHub, an early precursor to Gumloop.
His initial thought was to create “GitHub for agents,” a platform to host and interact with these autonomous AI entities. However, the “aha! moment” arrived swiftly: “That idea kind of crumbled after a few days when I realized the agents weren’t useful.” People wanted to use the platform, but they were frustrated by the agents’ unreliability.
Max realized what users were secretly asking for was “reliability, predictability.” Their use cases were simple, not agentic. This insight led him to build a framework for automating steps sequentially, which “naturally grew into this crazy automation platform.” Initially built for developers, the real traction came from an unexpected audience: “non-technical business admins… ops people or HR people who were just enthused about the idea of AI actually solving a problem for them.” This realization shifted Gumloop’s focus to building an approachable, fun-to-use platform for the everyday business user.
YC, Focus, and the Power of Building Something Great
Gumloop’s journey continued through Y Combinator, a period Max spent “stuck in Canada” due to his ban. This geographic isolation, however, proved to be an advantage. “I had all of the pressure of needing to build something amazing… but I didn’t have any of the distractions.” He coded intensely in his small Vancouver apartment, shunning networking events and parties.
This focus became a core tenet: “The people who are actually building something amazing are not at those events.” The lesson extends to fundraising: “If you build something exceptional, they will actually come to you… You just have to show them that you’ll succeed without them.”
The first paying customer, a user named Kai, paying $20 a month, was a monumental victory. “We freaked out. It was like the greatest moment ever,” Max recalls, a testament to the profound satisfaction of genuine value creation, however small.
The True AI Advantage: Understanding, Then Accelerating
Max circles back to his initial critique of AI hype, clarifying Gumloop’s philosophy. “It’s important to know what you should be using AI for and what you shouldn’t.” The goal isn’t to replace entire jobs but to be “super AI-enabled,” automating repetitive tasks while keeping the human touch in critical areas.
The “slippery slope” comes when people use AI to shortcut understanding. “If you’re using AI to do something you don’t actually understand at all, often times you’re creating just uncertainty,” he warns, likening “vibe coding” (using AI to code without understanding) to “making malware.”
Max believes AI should accelerate what you already understand, allowing you to “do it way faster so I can learn more things and like grow as a person.” He foresees a growing divide: those who use AI as a learning tool to deepen their understanding, becoming “exceptional even faster,” and those who “skip the understanding part and just accelerate with AI,” ultimately falling “to the slop.”
Building a Team of True Believers
Gumloop’s hiring strategy is as unconventional as its founder’s journey. Rather than traditional recruitment, many of their key hires were once customers. “Our customer from Instacart quit his job and joined us. Our customer from Webflow quit his job and joined us. Someone from Shopify quit their job and joined us,” Max proudly states.
These individuals already possess “conviction.” They loved the platform so much they chose to join the mission. “It’s a really fun way to hire because the only thing you really have as a startup is the optimism in general,” he explains. This approach fosters a culture where everyone is “equally excited about the mission” and loves what they’re doing, without rigid hours or forced late nights. Max’s co-founder joined after seeing an early working demo, proving that “you have to build something exceptional” to attract the best talent.
The Founder’s Blind Confidence
Ultimately, Max emphasizes that building a successful company isn’t about perfectly strategizing every “moat” or fearing big competitors like Zapier or OpenAI. “The people who are like really asking themselves that and obsessing over those questions will never build anything,” he contends.
Instead, it requires a “blind confidence,” a belief that “you can do it.” It’s about taking risks, trying things, proving others wrong, and when it doesn’t work, trying again. “I could have on day one explained to myself a hundred reasons why Zapier would do this better than us or why OpenAI and all these big players would crush us,” he reflects. But if he had, Gumloop wouldn’t exist.
The story of Gumloop is a powerful counter-narrative to the effortless AI dream. It’s a testament to the grit, iterative failure, user-centric focus, and deep understanding required to build something truly valuable. In a world saturated with quick fixes, Max Brodeur-Urbas reminds us that real innovation, even with the most advanced technologies like AI, still demands the oldest and most reliable ingredient: hard work.
Based on “From journalist to app developer using Claude Code” from How I AI Watch the original video
From Newsroom to Code: How a Journalist Built iOS Apps with AI’s Dueling Dev Team
In an era where artificial intelligence promises to reshape industries, few stories capture the essence of this transformation quite like Dan Roth’s. A veteran journalist and editor, known for shaping narratives at publications like Fortune, Wired, Forbes, and most recently LinkedIn, Roth has unexpectedly carved out a new identity: an iOS app developer. Without a formal background in software engineering, he’s building and shipping production-grade apps to the App Store, all thanks to a meticulously orchestrated team of AI agents he affectionately calls “Bob” and “Ray.”
Roth’s journey isn’t just a testament to the power of generative AI; it’s a fascinating look at how a non-technical professional can become a “vibe coder,” guiding complex development processes with a uniquely human touch.
The Echo of Disruption: From Blogs to Bots
Dan Roth is no stranger to technological upheaval. He recalls being at Fortune during the advent of blogging, a moment that democratized publishing and challenged the traditional gatekeepers of information. “All of a sudden, everyone had access to something that I thought was something that was unique to writers,” he explains. “WordPress came around and Tumblr, and suddenly everyone could talk to everyone, and you discovered this explosion of ideas.” While initially scary, he realized the embrace of more ideas was ultimately better.
This historical parallel provided Roth with a crucial lens when generative AI began making inroads into coding. “When generative AI started making its way into coding and it became clear that any of us could build anything, I had the exact same idea,” he recounts. But this time, he was on the opposite side of the fence. Instead of influencing what got built by pitching ideas to engineers and product managers, he could now build them himself.
What started as an idea quickly spiraled into an obsession. “It has become an obsession of mine of building to the point now where when I’m with friends and family, they’re like, ‘Please stop talking about the stuff that you’re building,’” Roth jokes. Weekends, for him, are now dedicated to iterating on his apps, a passion that has transformed his relationship with technology.
Meet the AI Dream Team: Bob the Builder and Ray the Reviewer
Roth’s secret sauce for building high-quality, production-grade software lies in his innovative use of two distinct AI agents, both powered by Claude Code, operating in separate terminal tabs. He’s given them clear roles, instructions, and even names: Bob and Ray.
Bob the Builder is the primary development agent. His instructions are precise:
- Plan First: “Everything has to plan first. You don’t build until it’s done planning.”
- Modular Building: Avoid “spaghetti code” by building everything in modules.
- Document Everything: Log all work into markdown files to maintain context and track progress.
- Stop Constantly for Ray: Bob is programmed to pause his work and run every plan and significant code milestone by Ray, the review agent.
Ray the Reviewer serves as the senior software engineer, with a keen focus on security and architectural integrity. Ray’s persona is designed to be rigorous and critical:
- Obsessed with Security: Guards against vulnerabilities.
- Architectural Integrity: Ensures designs adhere to established guidelines and best practices.
- Quality Control: Reviews code at milestones, checking for robustness and efficiency.
- The Power to Say No: Crucially, Ray is instructed not to rubber-stamp Bob’s plans. “Because Claude is always rubber stamping everything… I have to tell it, ‘you have to say no to things,’” Roth emphasizes.
And then there’s the third agent: Dan Roth himself. His role is unique: “I am the person who breaks the tie that often happens between Bob wanting to do something and Ray saying you can’t do it.” He acts as the ultimate manager, guiding the AI duo through conflicts and making final decisions.
This “dueling agents” setup is a revelation. Unlike simply prompting a single AI, Roth’s system mimics a real-world development team. A builder proposes, a reviewer scrutinizes, and a human manager arbitrates. This friction, Roth notes, is not just for efficiency, but for learning. “I’m not doing this just to build. I’m doing this to learn,” he explains, highlighting how observing Ray’s objections helps him understand the intricacies of secure and robust code.
The “Picky Customer”: Dan Roth’s Unique Role in Development
When asked to define his role in this AI-driven development process, Roth eschews traditional labels like “Product Manager” or “Architect.” He once considered himself a “mediocre PM” or an “architect,” but quickly realized those roles demand a depth of technical or strategic knowledge he doesn’t claim.
Instead, Roth embraces the title of “picky customer” or “vibe coder.” “All I am is a really picky customer,” he states. “I think that is like the role of the vibe coder is what do I care about deeply?” He illustrates this with an analogy: “I’m like walking through this house and I’m telling the architect, ‘No, I want this room blue. I know you don’t think it’s a good idea. I’m telling you this is what I want.’”
This perspective allows him to focus on the user experience, the product’s voice, and core functionality from a deeply personal standpoint. For example, in one of his apps, a podcast clipping tool, he insisted on a humorous, encouraging message when a clip was being created, rather than a generic “Your video is being created.” This personal touch, driven by his “picky customer” persona, infuses the app with personality.
Commutely: A Case Study in Personalized Software
Roth’s flagship app, Commutely, perfectly embodies his “picky customer” philosophy. “I keep missing the train,” he explains, “I’m gonna build an app that is just for me that is basically like, ‘Is the train almost here? Can I walk or do I have to run?’” This intensely personal problem-solving led to “perfect product market fit because I was the entire product.”
Commutely, designed for New York City subway riders, relies on iOS Live Activity to show real-time train updates on the lock screen. It’s a simple yet powerful solution born from personal frustration. As other riders discovered it, Commutely found its community, and with it, a stream of feature requests.
Roth manages these requests using another Claude chat, which he feeds ideas. This “Commutely feature idea and tracker” chat is prompted to rank ideas by estimated build time and potential impact on customer happiness and growth. On weekends, he consults this ranked list, picks a manageable feature, and initiates the Bob-and-Ray workflow.
For instance, Roth wanted to improve the discovery of a “scheduled automatic updates” feature. He’d already worked with Claude to develop a “retention plan,” saved as a markdown file. He then hands this plan to Bob, who drafts an implementation strategy. This plan is then meticulously reviewed by Ray for security and architectural soundness. Once Ray gives the “green light” (or suggests modifications), Bob proceeds to generate the code.
Mastering the AI-Powered Development Workflow
Roth’s workflow is rich with practical tips for anyone looking to leverage AI in development:
- Log Everything in Markdown: To combat AI’s “context window” limitations and his own weekend-only memory, Roth saves all interactions, plans, and code snippets in markdown files. This ensures continuous context for both him and his AI agents.
- Separate Agents, Separate Tabs: Running Bob and Ray in distinct Claude tabs provides transparency. When Bob needs Ray’s approval, Roth copies Bob’s output into Ray’s tab. This manual step, though seemingly inefficient, is intentional. “Putting a little bit of friction in the process where you’re actually forced to like copy and paste, put it over, read what Ray says… does help with learning,” he notes.
- Plan, Module, Branch: Bob is strictly instructed to plan thoroughly, build in reusable modules, and always work in separate code branches. Roth learned the hard way about the perils of merging directly to the main codebase.
- Managing the “Hungover Intern”: Roth humorously describes managing AI as “almost like managing a really smart but hung over intern.” AI is brilliant but forgetful, often suggesting solutions that contradict previously learned limitations (e.g., Live Activity constraints). This requires constant reminders and a managerial mindset.
- Navigating the App Store: The greatest remaining friction, Roth finds, is the Apple App Store submission process. He even envisions an AI tool specifically for App Store designs and metadata, simplifying what he calls “almost the last friction left in being a builder.”
Beyond Coding: AI as a Daily Productivity Co-pilot
Roth’s AI integration isn’t limited to his weekend app development. As an editor running a 400-person team, he relies on Copilot (with access to his work files and team communications) for daily productivity and context switching.
His favorite daily prompt, executed before he leaves work, is “What did I drop the ball on?” “It’s going to go through Outlook. It’s going to go through Teams. It’s going to go through any updated files,” he explains. Copilot identifies unresponded emails, unaddressed team messages, or neglected projects he’s been following. This “evening nightly nudge” allows him to wrap up loose ends, ensuring he starts the next day with a clean slate. “This is my savior,” he declares, “This is like what I count on AI to be able to do to make me better at my job so that I can spend the weekends building money losing apps like Commutely.”
The Human-AI Partnership: Assuming Best Intentions
Roth has a distinct philosophy for interacting with AI. He avoids “please” or “thank you,” opting for direct, straightforward commands. This, he says, is his way of “fighting until we get to that point” where AGI might run the show, a subtle act of “holding on to my humanity.”
When AI isn’t delivering, his tactic is akin to parenting. “I usually spend a lot of time saying, ‘we’ve gone over this already. You’ve talked about this. Go search your memory to go find the time where we’ve done this.’” He operates on the principle of “assume best intentions,” a lesson learned from working in tech companies. “I assume the AI has best intentions but has to be reminded about how we work. So I don’t yell. I’m pretty clear and I try to give it a little bit of grace.”
Dan Roth’s journey from journalist to AI-powered app developer is more than a personal anecdote; it’s a blueprint for the future of work. It demonstrates that with the right approach, a clear vision, and a little help from intelligent agents, anyone can become a builder, transforming personal ideas into tangible products and redefining what it means to create in the age of AI.
To follow Dan Roth’s insights and learn more about his AI-powered building journey, you can find him on LinkedIn (Daniel Roth) and subscribe to his newsletter, “Forward Deployed Editor,” which chronicles his learnings and missteps in the world of AI development for non-technical individuals.
Based on “A War Within the War: Israel’s Bombardment of Lebanon” from New York Times Podcasts Watch the original video
Beirut on Edge: Inside the Escalating War That Threatens to Reshape Lebanon
As Israeli strikes pummel the country, displacing millions and challenging Hezbollah’s decades-long dominance, fears of a full-scale ground invasion loom, transforming a regional conflict into a profound national crisis.
From the bustling streets of Beirut, an acrid smell now hangs heavy in the air, a constant reminder of the relentless Israeli airstrikes that have pounded southern Lebanon and parts of the capital. Above, the loud, incessant buzzing of Israeli drones has become a near-constant soundtrack to daily life. For the people of Lebanon, what began as a ripple from the broader U.S.-Israeli conflict with Iran has rapidly escalated into a terrifying “war within a war,” threatening to drag the country into its most devastating conflict in decades.
“I carried the children quickly and we left towards the airport road,” recounted one man, describing the chaotic exodus from his home. “People were like ants, like judgment day. It was like judgment day.” He, his wife, and four young children joined nearly a million others fleeing to what they hoped was safer ground, many spending nights on the streets, watching the distant glow of strikes.
This isn’t just another flare-up between long-standing adversaries. As Christina Goldbomb, the New York Times’ Beirut Bureau Chief, explains, the scale, rhetoric, and potential outcomes of this conflict are unprecedented, suggesting a fundamental shift in Israel’s objectives and a harrowing new chapter for Lebanon.
A City Transformed: The Ground Reality
Over the past two weeks, Beirut has presented a stark dichotomy. In some areas, life continues with an unsettling normalcy, buildings intact, untouched by the conflict. Yet, in other parts, particularly the southern suburbs (known as the Dahiyeh), the devastation is palpable. These Hezbollah strongholds have been “completely cratered” by Israeli bombardment. Mass displacement has turned parking lots, seaside cornices, sidewalks, and parks into makeshift tent cities fashioned from tarps and gray blankets, as government shelters in schools are quickly overwhelmed.
The conflict’s geographical spread is particularly alarming. Initially focused on southern Lebanon, Israeli strikes have expanded deep into central Beirut, targeting buildings and cars. Evacuation orders, once reserved for the south, are now being issued for central Beirut structures, with residents witnessing scenes of chaos as people rush to escape before strikes hit.
The human toll is already immense. With over a million people displaced in a nation of just 5.8 million, and more than 800 fatalities from Israeli strikes, a profound sense of dread has settled over the country. Lebanon, still reeling from past conflicts and internal crises, now grapples with the grim reality of being “dragged into another major war” for the second time in less than two years.
Beyond Retaliation: A Deeper Israeli Objective
What makes this campaign different from previous clashes, even those as recent as 2024? The answer lies in the escalating rhetoric from Israeli officials and the palpable fear of a major ground invasion—a scenario Lebanon hasn’t witnessed since the 1980s. A “massive buildup of troops on the Israeli side of the border” fuels these anxieties.
Israeli Defense Minister Yoav Gallant’s statements have sent shivers down Lebanese spines. He reportedly compared the southern suburbs of Beirut to Khan Younis, the Gazan city largely destroyed by Israeli operations, explicitly stating that Dahiyeh would “soon look like that.” Gallant also declared that if the Lebanese government failed to prevent Hezbollah from attacking Israel, Israel would “take it upon itself to do so in Lebanese territory.” This isn’t just about deterrence; it’s a clear signal of an intent to fundamentally alter the security landscape.
Compounding these fears are the sweeping evacuation orders issued for all of Beirut’s southern suburbs and a vast swath of southern Lebanon, extending up to 25 miles from the border—a measure not seen in the 2024 conflict. This expansion of the operational zone and the severity of the warnings suggest a more ambitious, and potentially more destructive, Israeli objective.
Hezbollah: A State Within a State and Israel’s Enduring Foe
The immediate trigger for this escalation was Hezbollah’s decision to fire missiles at Israel in retaliation for the killing of Ayatollah Ali, an Iranian leader. Israel, in turn, struck back, seeing this as an “opportunity to do what they have wanted to do for so long, which is for once and for all take out Hezbollah.”
To understand the complexity of this goal, one must grasp Hezbollah’s deep roots in Lebanese society. Formed in the 1980s during Lebanon’s civil war and Israel’s occupation of the south, Hezbollah emerged as a Shiite militant and political movement, dedicated to “resisting Israeli presence.” Backed by Iran, it gained immense prestige after Israel’s withdrawal in 2000, seen as a major victory.
Since then, Hezbollah has evolved into a formidable political party, holding seats in parliament and running a vast network of social services—schools, hospitals, and employment—that serve its supporters, primarily within the Shiite community (roughly a third of Lebanon’s population). This “state within a state” holds a “chokehold” on the Lebanese government, making political decisions impossible without its support. Its flags, merchandise, and visible presence, particularly in the south, underscore its deep embedding in the population. As one anecdote suggests, in a Hezbollah neighborhood, “those are the folks that are more likely to help” if your car breaks down.
While many believed Hezbollah had been severely weakened after the 2023-2024 conflicts—which saw intense Israeli bombardment, the “pager incident,” and the killing of its longtime leader, Hassan Nusella—the current war reveals a different truth. Despite the narrative of a battered group, Hezbollah has demonstrated a surprising resilience. Evidence suggests it has been manufacturing and locally assembling arms to replenish its stockpiles. Furthermore, members of Iran’s Islamic Revolutionary Guard Corps (IRGC) have reportedly entered the country to bolster Hezbollah’s ranks and take greater control of its military operations. The killing of at least two Israeli soldiers and injury of over a dozen others in the current conflict serves as stark proof of its enduring capability. It appears Hezbollah used the perceived “ceasefire” of the past year not to retreat, but to “rebuild and prepare for this major conflict.”
A Tested Loyalty: Lebanese Weariness and Questioning
Perhaps the most surprising development in this conflict is the nascent but growing weariness within Hezbollah’s traditional support base. For decades, a “fervent ideological zeal” to resist Israel “at any cost” has defined this community. However, the current war, following so closely on the heels of the 2023-2024 conflict, is testing that commitment like never before.
The vast majority of the displaced are from areas sympathetic to Hezbollah, meaning its own supporters are shouldering the immense burden of these repeated wars. “The one thing that they keep telling me is how exhausted they are,” Goldbomb reports. “They’re tired of evacuating from their homes. They’re tired of living in shelters. They’re tired of their towns and villages, especially in the south, being flattened by air strikes. They’re tired of rebuilding homes only to now worry that they’re going to get demolished again.”
This profound exhaustion is leading to a “very new feeling”: people are beginning to question Hezbollah’s actions. In one shelter, a nurse displaced from Natia, a southern city, openly debated with her family whether Hezbollah remained their “main defender.” While Israeli forces moved into the south, the Lebanese army was “pretty much nowhere to be found.” Yet, they also questioned whether Hezbollah, by dragging the country into war twice in two years, was truly defending the Shiite community, and whether the “cost the Shia community is bearing is worth it.” The most poignant question is why Hezbollah chose to fire on Israel in response to the killing of a foreign leader (Ayatollah Ali), but not over the past year when Lebanese citizens were killed in Israeli airstrikes. This questioning marks a significant shift in a community long defined by its unwavering support.
Lebanon’s Crippled State and the Shifting Sands of U.S. Policy
The Lebanese government finds itself in an impossible bind. It is “dysfunctional,” its military weak, underpaid, and outmatched by Hezbollah. While a “sea change” after the 2024 ceasefire saw the government agree to begin disarming Hezbollah, the challenges are immense. Any decisive move against Hezbollah risks a clash between Lebanese soldiers and Hezbollah fighters, a “recipe for civil war.”
Historically, the United States has often acted as a tempering influence on Israel’s actions in Lebanon, fearing regional destabilization. However, this appears to be changing. The U.S. has “essentially allowed Israel to carry out this bombardment of Lebanon” and is now exerting “immense pressure on the Lebanese government to act more decisively to disarm Hezbollah,” even to declare it a terrorist group. This is despite concerns from Lebanese officials that such actions could be profoundly destabilizing. The “check” that the U.S. once provided on Israeli actions in Lebanon “seems to have disappeared or dissipated,” at least for now.
The Grim Prospect of a “New Reality”
Israel’s stated goal is to “completely eliminate the threat from Hezbollah,” a decades-long ambition that has so far proven elusive. But the massive Israeli military buildup along the border suggests a more immediate and potentially achievable objective: a “massive Israeli ground invasion of southern Lebanon” to create a “new much larger buffer zone” or even a re-occupation of territory. For Prime Minister Netanyahu, such an outcome could be a significant political win, even if Hezbollah isn’t entirely “ousted.” However, for Lebanon, it would mean a repeat of history, igniting memories of the civil war and past occupations, and causing even more mass displacement.
The future of this conflict, deeply intertwined with the wider war involving Iran, appears grim. Hezbollah aims to inflict maximum casualties on Israeli forces to raise the political cost of continued engagement. Yet, there is a growing fear in Lebanon that this war will “last much longer than the Iranian war,” with Israel turning its full military might towards Lebanon once the Iranian front subsides.
Eerie similarities to the conflict in Gaza are emerging. Sweeping evacuation orders for southern Lebanon mirror those in Gaza. Israeli warplanes have dropped leaflets over central Beirut, calling on people to disarm Hezbollah and speaking of a “quote new reality in Lebanon.” Israeli officials have also hinted that ambulances and civilian trucks, allegedly used by Hezbollah to move weaponry, “would no longer be off limits.”
These parallels paint a chilling picture: Lebanon faces the prospect of a protracted, deadly conflict with potentially “much more permanent” aims from Israel. The “new reality” could mean “more clashes with Hezbollah, more displaced people, and more civilian casualties for years and years to come.” For a nation already at its breaking point, the war within a war threatens to unleash an unparalleled catastrophe.
한국어
""50 AI Agents Running My Company” Is a Lie. Here’s How I Build It | Gumloop, Max Brodeur-Urbas” — EO 기반 기사 원본 영상 보기
”50개 AI 에이전트로 회사 운영? 그건 거짓말입니다.” AI 시대, 허상에 속지 않고 진짜 비즈니스를 만드는 법 | Gumloop CEO 맥스 브로더-어버스
최근 소셜 미디어에서는 “AI 에이전트 50개가 우리 회사를 운영한다”거나 “나는 일주일에 한 시간만 일하고 주말에 1,000만 달러를 번다”는 식의 주장이 유행처럼 번지고 있습니다. 마치 AI가 모든 것을 자동화하여 손쉽게 성공할 수 있다는 환상을 심어주는 듯합니다. 하지만 자동화 플랫폼 Gumloop의 창립자이자 CEO인 맥스 브로더-어버스(Max Brodeur-Urbas)는 이러한 주장이 대부분 마케팅에 불과하며, AI를 활용한 비즈니스 성공에는 훨씬 더 현실적이고 근본적인 접근 방식이 필요하다고 단호하게 말합니다.
맥스 CEO는 AI 시대의 허상을 꼬집으며, “단 한 번의 클릭으로 사업을 구축할 수 있다거나, 이 과정을 구매하면 연간 백만 달러를 벌 수 있는 비법을 얻는다는 생각은 잘못된 접근”이라고 지적합니다. 그는 희망을 파는 것은 쉽지만, 힘든 과정을 건너뛰고 바로 가치에 도달하려는 시도는 결코 성공할 수 없다고 강조합니다. 이 기사에서는 맥스 브로더-어버스 CEO의 경험과 통찰을 통해 AI 시대에 진정한 가치를 창출하고 성공적인 스타트업을 구축하는 현실적인 방법을 탐구합니다.
AI 환상과 불편한 진실: ‘슬롯머신’이 아닌 ‘슬롭’을 만들지 마라
맥스 CEO는 현재 AI 분야에서 나타나는 일종의 ‘안티 패턴(anti-pattern)‘을 경계합니다. 바로 “50개의 AI 에이전트가 회사를 운영한다”거나 “AI 스위트가 매일 나에게 무엇을 해야 할지 알려준다”는 식의 접근입니다. 그는 이를 “슬롯머신”을 만드는 것과 같다고 비판하며, 정확히는 ‘슬롯(slot)‘이 아닌 ‘슬롭(slop, 형편없는 것)‘에 가깝다고 일침을 가합니다.
트위터에서 흔히 볼 수 있는 “모든 것을 자동화해서 일주일에 한 시간 일하고 1,000만 달러를 번다”는 주장의 대부분은 거짓말이고 마케팅이라는 것입니다. 이러한 메시지는 힘든 노력을 건너뛰고 지름길을 통해 가치에 도달하려는 환상을 심어주지만, 이는 결코 현실이 될 수 없습니다. 하지만 이러한 환상을 파는 사람들은 엄청난 돈을 벌고 있습니다. 맥스 CEO는 “코스 브로(course bros)“라고 부르는, 생산성이라는 비전을 팔면서 실제로는 아무런 새로운 가치도 제공하지 않는 이들을 경계해야 한다고 말합니다. 진정한 가치는 자신이 깊이 이해하는 분야에 AI를 적용하고, 이를 확장하는 데서 온다는 것입니다.
빅테크의 ‘황금 수갑’을 벗어던지다
맥스 CEO는 캐나다 몬트리올의 맥길 대학교(McGill University)에서 소프트웨어 공학을 전공했습니다. 그의 학창 시절 목표는 좋은 직장을 얻어 빅테크 기업에서 일하는 것이었습니다. 하지만 그는 빅테크 생활이 자신에게 맞지 않는다는 것을 빠르게 깨달았습니다.
“많은 사람이 ‘빅테크에 잠시 가서 배우고 나중에 내 사업을 시작하겠다’고 말하지만, 저는 그것이 일종의 도피라고 생각합니다.” 그는 많은 이들이 빅테크의 ‘황금 수갑(golden handcuffs)‘에 갇혀 결국 떠나지 못하고 자신만의 사업을 시작하지 못한다고 지적합니다. 실제로 맥스 CEO는 빅테크에서 배운 것이 스타트업에 전혀 도움이 되지 않았다고 말합니다. 빅테크 기업의 로고가 이력서에 있으면 “어느 정도의 디폴트 존중(default respect)“을 얻을 수 있지만, 그곳에서 새로운 것을 배우지는 못했다는 것입니다. 오히려 마이크로소프트(Microsoft)에서 싫었던 점들이 현재 Gumloop의 운영 방식에 많은 영향을 미쳤습니다.
21~23세의 젊은 시절은 책임과 의무가 없는 귀중한 시간입니다. 이 시기를 단순히 9-to-5 근무에 매여 티켓을 고치고 퇴근하는 데만 보낸다면, 그 중요한 시간을 낭비하는 것이라고 맥스 CEO는 강조합니다. “지금 바로 시작하자. 나중에 더 안정적이고 책임감 있는 것을 찾자”는 당시의 결정이 옳았다고 그는 회고합니다.
추방과 실패가 빚어낸 성공: ‘틀렸음을 증명하라’
맥스 CEO의 창업 여정은 순탄치 않았습니다. 마이크로소프트를 그만두고 밴쿠버로 돌아온 그는 시애틀에 있는 옛 룸메이트를 방문하려다 미국 국경에서 추방당하고 5년간 입국 금지 조치를 받았습니다. 2일만 머무르겠다고 했지만, 국경 관리들은 그가 더 오래 머무를 것이라고 의심했고, 결국 수상한 인물로 간주한 것입니다.
이 사건은 그에게 충격이었지만, 동시에 중요한 전환점이 되었습니다. “그 순간, 저는 회사를 만들어야 한다는 것을 깨달았습니다. 다른 대안이 없었으니까요.” 그는 사업을 더 진지하게 받아들이고, 사람들이 실제로 원하는 것, 돈을 벌 수 있는 것을 만들기로 결심했습니다.
이후 6개월간 그는 가능한 한 열심히 일하며 다양한 아이디어를 시도했습니다. VR 비디오 게임 중재 소프트웨어, 일반적인 신뢰 및 안전 도구, 웹 트래픽 봇 감지 소프트웨어, 안티-캠 플랫폼 등 수많은 MVP(Minimum Viable Product, 최소 기능 제품)를 만들고 시장의 관심을 확인하려 했습니다. 그는 일주일에 한 번꼴로 새로운 아이디어를 실험했고, 곧 그것이 나쁜 아이디어임을 깨달았습니다.
이 과정에서 맥스 CEO는 스타트업의 역설적인 진실을 발견합니다. “스타트업에서는 사실 ‘내가 틀렸다는 것을 증명하는 것’을 쫓아야 합니다. 그것이 일주일 또는 몇 달의 시간을 절약해 주기 때문에 가장 좋은 일입니다.” 초기에는 몇 달 동안 아이디어를 구축하고는 누군가가 자신의 생각이 옳았다고 증명해주기를 바랐지만, 이는 잘못된 접근이었습니다. 대신 “왜 이것이 작동하지 않을지”를 적극적으로 찾아야 합니다. 만약 작동하지 않을 이유를 찾을 수 없다면, 그때 비로소 추구할 만한 구체적인 아이디어를 얻게 되는 것입니다. 그는 Gumloop을 만들기 전에 열 번이나 실패하지 않았다면 지금의 Gumloop은 없었을 것이라고 말합니다.
사용자와의 소통 역시 매우 중요합니다. 초기에는 사용자를 찾아다니며 제품을 사용해달라고 간청해야 하는 어려운 상황에 처할 수 있지만, 이는 반드시 거쳐야 할 과정입니다. “제품이 형편없다거나, 문제 해결에 도움이 안 된다는 등 듣고 싶지 않은 피드백이라 할지라도, 그것이 가장 가치 있는 피드백입니다.”
AutoGPT에서 Gumloop까지: 문제 해결의 진화
현재 Gumloop의 초기 버전을 개발할 때, 맥스 CEO는 밤늦게까지 사무실에 머물며 작업에 몰두했습니다. “무엇을 만들고 있는지 너무 흥분해서 노트북을 닫고 싶지 않았습니다.” 그는 이렇게 밤늦게까지 일하게 만드는 ‘그것’을 찾으면, 매일매일이 더 쉬워지고 엄청난 추진력을 얻게 된다고 설명합니다.
Gumloop의 아이디어는 오픈소스 AI 에이전트 프레임워크인 AutoGPT에서 시작되었습니다. AutoGPT는 AI가 스스로 문제를 해결할 수 있다는 가능성을 보여주며 큰 반향을 일으켰습니다. 맥스 CEO는 AutoGPT 디스코드 서버에서 많은 비기술적인 사용자들이 “GitHub이 뭐냐”, “터미널은 어떻게 사용하냐”, “종속성(dependency)은 무엇이냐”와 같은 기본적인 질문으로 어려움을 겪는 것을 발견했습니다. 그는 이들을 위해 간단한 UI를 가진 AgentHub라는 플랫폼을 만들었습니다.
초기에는 AgentHub를 ‘에이전트용 GitHub’으로 만들 생각이었지만, 에이전트 자체가 신뢰할 수 없다는 것을 깨달으면서 아이디어는 무너졌습니다. 그러나 사용자들은 플랫폼을 원했지만, 에이전트의 불안정성 때문에 좌절하고 있었습니다. 맥스 CEO는 이들이 암묵적으로 요구하는 것이 ‘신뢰성(reliability)‘과 ‘예측 가능성(predictability)‘임을 간파했습니다.
“그들의 사용 사례는 너무나 단순해서, 저는 그들이 단계별로 자동화할 수 있는 프레임워크를 만들 수 있다고 생각했습니다.” 이것이 바로 Gumloop으로 발전하게 된 계기입니다. 초기에는 개발자를 위한 프로젝트였지만, 실제로는 회사 내의 비기술적인 비즈니스 관리자, 운영 담당자, HR 담당자들이 AI를 통해 업무를 자동화하는 아이디어에 열광했습니다. 80%가 비기술적인 사용자라는 것을 깨달은 후, 그는 복잡함 없이 재미있고 접근하기 쉬운 제품을 만드는 데 집중했습니다.
YC와 집중의 힘: “위대한 것을 만들면 투자자가 찾아온다”
Gumloop은 Y Combinator(YC) 배치 시작 5개월 전에 합류했습니다. 당시 제품은 전적으로 무료였지만, YC 배치 첫 주에 월 20달러의 가격을 책정했습니다. 이는 ChatGPT보다 비싸지 않아야 한다는 생각 때문이었습니다. 첫 유료 고객인 카이(Kai)가 20달러를 결제했을 때, 팀은 엄청난 흥분을 감추지 못했습니다.
맥스 CEO는 YC 기간 내내 캐나다에 머물러야 했습니다. 이는 “놀라운 것을 만들어야 한다”는 YC의 기대에 부응해야 한다는 압박감을 주었지만, 동시에 “어떤 방해도 없이” 오직 코딩에만 집중할 수 있는 환경을 제공했습니다. 그는 밴쿠버의 작은 스튜디오 아파트에서 최대한 빠르게 코딩하며 지냈습니다.
이 경험을 통해 그는 “집중력을 유지하고 네트워킹 행사나 파티에 현혹되지 않는 것이 매우 중요하다”는 것을 깨달았습니다. “정말로 놀라운 것을 만드는 사람들은 그런 행사에 가지 않습니다.” 그는 이러한 태도를 계속 유지하고 있으며, 공동 창업자 역시 거의 모든 행사에 불참하며 오직 일에만 전념하고 있습니다.
맥스 CEO는 “집중하고 사용자들과 최대한 많이 소통하면, 네트워크는 자연스럽게 형성될 것”이라고 강조합니다. 많은 사람이 투자를 유치하기 위해 네트워킹을 해야 한다고 생각하지만, “탁월한 것을 만들면 투자자들이 먼저 당신을 찾아올 것”이라는 것입니다. YC에서 얻은 가장 큰 깨달음은 “그리 복잡하지 않다. 그저 훌륭한 것을 만들면 된다”는 것이었습니다. 칵테일 파티에서 네트워크가 만들어지는 것이 아니라는 메시지입니다.
AI 시대의 현명한 활용법과 미래 인재
맥스 CEO는 AI 활용에 대한 명확한 철학을 가지고 있습니다. “AI를 무엇에 사용해야 하고 무엇에 사용하지 말아야 하는지 아는 것이 중요합니다. 중요한 부분에는 인간적인 감각을 유지하고 반복적인 작업을 자동화하는 것이 올바른 방법입니다.” 그는 최고의 사용자들은 AI를 통해 능력이 극대화된 사람들이지, AI가 자신의 직업 전체를 대체하게 두는 사람들이 아니라고 말합니다.
“대부분의 마케팅과 달리, 저는 AI를 깊이 이해하는 것에 적용할 때 가치가 창출된다고 생각합니다.” 만약 주말에 3만 달러를 벌게 해줄 마법 같은 해결책이 있다면, 그들은 트위터에서 그것을 나눠주지 않을 것이라는 냉철한 현실을 지적합니다. 가장 생산적이고 가치를 창출하는 사람들은 자신이 정말로 잘 아는 것을 AI와 결합하여 확장하는 이들입니다.
이러한 맥락에서 그는 AI 시대의 인재 양성에 대한 우려를 표합니다. AI를 전혀 이해하지 못하는 상태에서 AI를 코딩에 사용한다면, 결국 멀웨어(malware)를 만들게 될 수도 있다고 경고합니다. 이해하지 못하는 것을 자동화하려 한다면 형편없는 결과물을 만들게 될 것입니다.
맥스 CEO는 AI를 자신의 속도를 높이고, 이미 이해하는 것을 훨씬 빠르게 처리하며, 더 많은 것을 배우고 성장하는 도구로 활용해야 한다고 주장합니다. 그는 AI가 자신을 대체하거나 이해하는 과정을 건너뛰는 수단이 되어서는 안 된다고 강조합니다.
“어쩌면 마지막 위대한 엔지니어 세대는 ‘실제로 무엇이 일어나고 있는지 이해한 다음 AI의 가속을 받은’ 시대에 태어났을지도 모릅니다.” 이제는 사람들이 이해하는 단계를 건너뛰고 AI로 가속화할 수 있기 때문입니다. 그는 AI를 학습 도구로 사용하여 근본 원리를 이해하는 소수의 인재와, 이해하는 과정을 건너뛰는 다수의 인재로 양극화될 것이라고 예측합니다. 전자는 훨씬 빠르게 탁월해질 것이고, 후자는 ‘슬롭’으로 전락할 것이라는 경고입니다.
탁월함이 인재를 이끈다: Gumloop의 채용 철학
Gumloop은 거의 모든 직원을 네트워크를 통해 채용했습니다. 특히 많은 직원은 Gumloop의 기존 고객이었습니다. 인스타카트(Instacart), 쇼피파이(Shopify), 도어대시(DoorDash), 웹플로우(Webflow) 등 대기업의 고객들이 자신의 직장을 그만두고 Gumloop에 합류한 사례가 많습니다. 이들은 이미 Gumloop 플랫폼에 대한 강한 확신을 가지고 있었고, 미션에 기여하기 위해 모든 것을 내려놓을 만큼 플랫폼을 사랑했습니다.
맥스 CEO는 “채용은 마치 연애와 같습니다. 누군가가 당신과 데이트하고 싶어 할 만한 사람이 되어야 하는 것처럼, 기업도 최고의 인재가 함께하고 싶어 할 만한 매력적인 곳이 되어야 합니다.”라고 비유합니다. 단순히 구걸하는 것이 아니라, 탁월한 것을 만들고 견인력을 확보해야 최고의 인재들이 자발적으로 합류하고 싶어 한다는 것입니다.
Gumloop의 공동 창업자 역시 맥스 CEO가 초기에 만든 작동하는 제품 버전을 보고 흥분하여 합류했습니다. Gumloop의 조직 문화는 매우 자유롭고 즐겁습니다. 정해진 출퇴근 시간이 없으며, 모두가 미션에 대해 똑같이 흥분하고 자신이 하는 일을 사랑합니다. 맥스 CEO는 채용 시 “이 사람과 모든 시간을 함께 보내고 싶은가? 24시간 같이 어울릴 수 있는가?”라는 질문을 중요한 기준으로 삼습니다. 이러한 방식으로 채용된 인재들이 모여 서로에게 긍정적인 영향을 주고, 회사의 성장 동력을 만들어낸다는 것입니다.
맹목적인 자신감과 끊임없는 도전: “그냥 시도하라”
“모든 스타트업을 만들지 말아야 할 이유는 백만 가지도 넘습니다.” 맥스 CEO는 아이디어를 듣고 “해자(moat)가 무엇인가?”, “왜 대기업이 이것을 더 잘하지 못하는가?”와 같은 질문에 집착하는 사람들은 결국 아무것도 만들지 못하고 대기업의 부속품이 될 뿐이라고 말합니다.
대신, “위험을 감수하고 다른 사람들의 생각이 틀렸음을 증명하기 위해 시도하라”고 그는 조언합니다. 그러면 사람들은 “어떻게 여기까지 왔는가? 어떻게 이렇게 많은 사람이 사용하는 회사를 이끌고 있는가?”라고 물을 것이고, 그 답은 “그냥 시도했고, 성공했으며, 실패했을 때는 다시 시도했다”는 것이 될 것입니다.
맥스 CEO는 첫날부터 자피어(Zapier)나 OpenAI 같은 대기업이 Gumloop보다 훨씬 잘할 것이라는 100가지 이유를 자신에게 설명할 수 있었지만, 그렇게 하지 않았기에 Gumloop이 존재할 수 있었다고 말합니다. 창업자를 창업하게 만드는 가장 큰 자질은 “나는 할 수 있다”는 맹목적인 자신감입니다. “자신이 할 수 있다고 생각할 때, 사람이 얼마나 많은 것을 해낼 수 있는지는 정말 놀라운 일입니다.”
AI 시대의 성공은 단축키를 찾거나 환상에 의존하는 것이 아니라, 깊이 있는 이해를 바탕으로 끊임없이 도전하고, 실패를 통해 배우며, 사용자에게 집중하고, 탁월한 인재들과 함께 맹목적인 자신감으로 나아가는 과정에 있음을 맥스 브로더-어버스 CEO는 명확하게 보여주고 있습니다.
“From journalist to app developer using Claude Code” — How I AI 기반 기사 원본 영상 보기
기자에서 앱 개발자로: AI 비서 ‘밥’과 ‘레이’로 앱스토어 정복한 댄 로스의 ‘바이브 코딩’ 비법
[도입부] 링키드인(LinkedIn) 편집장 댄 로스(Dan Roth)는 그의 경력을 비즈니스 작가이자 편집자로 시작했습니다. 하지만 이제 그는 AI 도구, 특히 클로드 코드(Claude Code)를 활용하여 iOS 앱을 개발하고 앱스토어에 출시하는 ‘바이브 코딩(vibe coding)’ 전문가로 변모했습니다. 비기술적 배경을 가진 그가 어떻게 고품질의 프로덕션 수준 소프트웨어를 구축할 수 있었을까요? 그 비결은 바로 ‘밥(Bob)‘과 ‘레이(Ray)‘라는 두 명의 AI 에이전트를 활용한 독특한 협업 시스템에 있습니다. 이번 기사에서는 댄 로스의 혁신적인 개발 철학과 워크플로우를 자세히 들여다봅니다.
1. 언론인의 AI 시대 적응기: ‘모두의 개발’ 시대 개척자
댄 로스는 과거 포춘(Fortune), 와이어드(Wired), 포브스(Forbes)에서 오랜 기간 비즈니스 작가로 활동했습니다. 그는 블로그가 막 등장하던 시절, 작가들에게만 주어졌던 ‘세상과 소통하는 능력’이 워드프레스(WordPress)나 텀블러(Tumblr) 같은 플랫폼을 통해 모두에게 개방되면서 아이디어의 폭발적인 성장을 목격했습니다. 처음에는 두려웠지만, 곧 더 많은 아이디어가 더 낫다는 것을 깨달으며 변화를 받아들였습니다.
몇 년 전 생성형 AI가 코딩 분야에 진입하면서, 로스는 비슷한 통찰을 얻었습니다. “누구나 무엇이든 만들 수 있겠다”는 생각이 든 것입니다. 과거에는 아이디어를 현실로 만들기 위해 엔지니어(engineer)나 PM(Product Manager)을 설득하고 협업해야 했지만, 이제는 직접 구축할 수 있게 되었습니다. 이는 로스에게 엄청난 열정으로 다가왔고, 주말마다 앱 개발에 몰두하는 ‘집착’으로 이어졌습니다. 그는 이를 “돈 먹는 앱(money-losing apps)“이라고 표현하지만, 그 과정에서 얻는 즐거움과 학습의 가치를 높이 평가합니다.
2. 댄 로스식 ‘바이브 코딩’의 핵심: 듀얼 AI 에이전트 시스템
댄 로스의 개발 철학은 자신을 “까다로운 고객(picky customer)“으로 정의하는 ‘바이브 코딩’에 뿌리를 두고 있습니다. 그는 건축가에게 “이 방은 파란색으로 해달라”고 요구하는 건축주처럼, 깊이 있게 신경 쓰는 부분에 집중하며 AI에게 원하는 바를 명확히 전달합니다. 그의 핵심 전략은 두 명의 AI 에이전트, 즉 ‘밥(Bob)‘과 ‘레이(Ray)‘를 활용한 독특한 협업 시스템입니다.
가. 빌더 에이전트 ‘밥(Bob)’ ‘밥 더 빌더(Bob the Builder)‘라는 별명을 가진 밥은 실제 코드를 구축하는 역할을 맡습니다. 밥에게는 다음과 같은 명확한 지침이 주어집니다.
- 철저한 계획: 무엇이든 구축하기 전에 반드시 계획을 세워야 합니다.
- 모듈화: 코드를 스파게티처럼 복잡하게 만들지 않고, 모듈(module) 형태로 깔끔하게 구축해야 합니다.
- 문서화: 모든 과정을 마크다운(Markdown, MD) 파일로 기록하여 컨텍스트(context) 손실을 방지하고 작업 내용을 추적합니다.
- 레이와의 협의: 항상 ‘레이’에게 모든 계획을 검토받아야 합니다.
나. 리뷰 에이전트 ‘레이(Ray)’ 레이는 ‘보안에 집착하는 수석 소프트웨어 엔지니어(senior software engineer obsessed with security)’ 역할을 수행합니다. 그의 주요 임무는 다음과 같습니다.
- 계획 검토: 밥이 코드를 구축하기 전에 그의 계획을 검토합니다.
- 마일스톤 코드 검토: 특정 개발 단계(milestone)에서 코드를 검토합니다.
- 보안 및 아키텍처 수호: 보안 취약점이나 디자인 가이드라인 위반 여부를 확인합니다.
- “아니오”라고 말하기: 클로드(Claude)가 모든 것을 긍정하는 경향이 있기에, 레이에게는 필요할 경우 “아니오”라고 말할 수 있는 권한과 책임이 부여됩니다. 이는 AI의 무조건적인 동의를 방지하고 비판적인 시각을 유지하는 데 중요합니다.
다. 인간 에이전트 ‘댄(Dan)’ 댄 로스 본인은 이 시스템에서 “결정타를 날리는 사람(the person who breaks the tie)” 역할을 합니다. 밥이 무언가를 하려 하고 레이가 “안 된다”고 말할 때, 댄이 최종 결정을 내립니다. 그는 자신이 PM처럼 앱 전체를 머릿속에 넣고 우선순위를 정하는 데 능숙하지 않다고 인정합니다. 대신, 그는 “까다로운 고객”으로서 “무엇을 깊이 신경 쓰는가?”라는 질문에 답하며 개발 방향을 제시합니다.
라. 기술적 구현 및 학습 철학 로스는 클로드 웹 앱에 두 개의 별도 탭을 열어 밥과 레이를 각각 운영합니다. 각 에이전트에게는 맞춤형 프롬프트(prompt)를 통해 역할을 부여하고, 모든 작업 내용은 MD 파일로 기록합니다. 특히 주목할 점은 밥이 생성한 계획을 수동으로 복사하여 레이에게 붙여넣는 ‘의도적인 마찰(deliberate friction)‘을 유지한다는 것입니다. 로스는 이를 통해 AI가 무엇에 어려움을 겪는지 직접 확인하고, 비기술적 배경을 가진 자신이 학습하는 데 도움을 받는다고 설명합니다.
3. 실제 워크플로우 들여다보기: 아이디어에서 앱스토어까지
댄 로스의 개발 워크플로우는 아이디어 구상부터 앱스토어 출시까지 체계적으로 이루어집니다.
가. 아이디어 관리 및 우선순위 지정 로스는 클로드 채팅창을 ‘커뮤틀리(Commutely)’ 앱의 기능 아이디어 및 추적기(tracker)로 활용합니다. 사용자 피드백을 바탕으로 새로운 아이디어를 입력하면, 클로드는 이를 구축 시간과 앱 성장에 미칠 잠재적 영향력(고객 만족도 및 성장 영향)이라는 1~3점 척도를 기준으로 순위를 매깁니다. 로스는 주말에 이 목록을 보고 몇 시간 안에 구축할 수 있는 기능을 선택하여 개발에 착수합니다.
나. 계획 수립 및 검토 선택된 기능에 대해 로스는 밥에게 개발을 지시합니다. 밥은 MD 파일로 기록된 기존의 리텐션(retention) 계획 등을 참조하여 상세한 구현 계획을 세웁니다. 이 계획이 완성되면, 로스는 이를 복사하여 레이에게 전달합니다. 레이는 해당 계획이 아키텍처적으로 올바른지, 보안상 문제는 없는지, 기존 디자인 가이드라인을 준수하는지 등을 검토합니다. 레이는 때로는 “엣지 케이스(edge case)” 우려를 제기하기도 하며, 밥의 계획이 타당하면 “녹색불(green light)“을 줍니다.
다. 코드 생성 및 테스트 레이의 승인이 떨어지면, 밥은 계획에 따라 코드를 생성하기 시작합니다. 이때 밥은 항상 새로운 브랜치(branch)에서 작업하도록 지시받습니다. 로스는 과거 메인(main) 브랜치에 직접 코드를 병합(merge)했다가 문제가 발생했던 경험을 통해 브랜치 사용의 중요성을 깨달았습니다.
생성된 코드는 Xcode로 가져와 시뮬레이터(simulator)와 실제 아이폰에서 테스트합니다. 로스는 Xcode의 기능을 5% 정도만 사용한다고 겸손하게 말하지만, 클로드에게 배운 기본적인 명령어(예: Command Shift K로 클린 빌드, Control B로 빌드)를 활용합니다. 그는 시뮬레이터 테스트만으로는 부족하며, 실제 기기에서 수많은 엣지 케이스를 발견한다고 강조합니다.
라. 앱스토어 출시의 난관 테스트가 완료되면 로스는 앱을 애플 앱스토어(App Store)에 배포합니다. 그는 개발 과정에서 가장 어려웠던 부분이 바로 앱스토어 탐색이었다고 고백합니다. 클로드를 통해 앱스토어 정책과 절차를 배우는 것이 “마지막 남은 마찰(last friction)“이라고 말할 정도입니다. 이에 그는 앱스토어 스크린샷 디자인을 자동화하는 AI 앱을 직접 만들고 싶다는 아이디어를 내기도 했습니다.
4. AI 시대의 ‘관리자’ 역할: 인간 고유의 가치
댄 로스는 AI 에이전트들을 관리하는 역할을 통해 인간 고유의 가치를 발휘합니다.
가. ‘숙취에 시달리는 인턴’ 비유 로스는 AI를 “매우 똑똑하지만 숙취에 시달리는 인턴(really smart but hung over intern)“에 비유합니다. AI는 뛰어난 능력을 가졌지만, 이전의 컨텍스트나 제한 사항(예: iOS 라이브 액티비티의 기술적 한계)을 쉽게 잊어버리곤 합니다. 이때 로스는 “우리 전에 이거 얘기했잖아, 기억 안 나?”라고 상기시키며 AI를 올바른 방향으로 이끌어야 합니다. 이는 마치 유능하지만 경험이 부족한 주니어 인력을 관리하는 것과 유사합니다.
나. 까다로운 고객으로서의 역할 그는 자신이 품질 보증(QA) 담당자라기보다는 “샴페인을 마시기 전에 냄새를 맡는” 까다로운 고객이라고 말합니다. 앱의 작은 디테일이나 메시지의 톤(tone)까지 자신의 개성을 불어넣습니다. 예를 들어, 팟캐스트 클립 앱에서 영상 생성 중 오류 메시지를 “오 마이 갓, 당신은 팟캐스트의 멋진 부분을 클립했어요!”처럼 재미있게 바꾸도록 지시했습니다. 이는 인간만이 부여할 수 있는 ‘영혼’이자 ‘개성’입니다.
다. 에이전트 계층 구조 로스는 밥에게는 “서브 에이전트(sub-agents)를 생성할 수 있지만, 그 에이전트들에 대한 책임은 너에게 있다”고 지시합니다. 하지만 서브 에이전트들은 또 다른 서브 에이전트를 만들 수 없습니다. 반면 레이는 오직 하나의 “문지기(gatekeeper)“로서, 누구도 레이를 관리할 수 없고 레이 또한 서브 에이전트를 만들 수 없습니다. 이는 로스가 15년간 테크 회사에서 일하며 목격한 ‘수석 엔지니어(principal engineer)‘와 같은 핵심 의사결정자의 역할을 AI 시스템에 반영한 것입니다.
5. 비코딩 업무에도 AI 활용: 일상 속 생산성 향상
댄 로스는 코딩 외에도 일상 업무에서 AI를 적극적으로 활용하여 생산성을 높입니다. 400명 규모의 팀을 관리하는 그는 하루 종일 컨텍스트 전환(context switch)이 잦습니다. 이때 마이크로소프트 코파일럿(Microsoft Copilot)을 활용하여 업무 흐름을 관리합니다.
그의 가장 유용한 비코딩 워크플로우는 퇴근 전 실행하는 “제가 놓친 일은 무엇인가요?(What did I drop the ball on?)”라는 프롬프트입니다. 코파일럿은 아웃룩(Outlook), 팀즈(Teams), 업데이트된 파일 등을 분석하여 그날 놓쳤거나 응답하지 않은 이메일, 팀즈 메시지, 프로젝트 진행 상황 등을 요약해줍니다. 이는 단순히 자신에게 언급된 내용뿐만 아니라, 그가 관심을 가지고 팔로우하던 프로젝트까지 추적합니다. 로스는 이를 “저녁 알림(nightly nudge)“이라고 부르며, 밤에 하루를 정리하고 다음 날 아침을 깨끗하게 시작할 수 있도록 돕는 “구원자(savior)“라고 표현합니다.
6. AI와의 대화법: ‘최고의 의도’를 가정하며
댄 로스는 AI와의 대화 방식에 있어서도 자신만의 철학을 가지고 있습니다. 그는 AI를 의인화(personify)하지 않기 위해 “부탁드립니다(please)“나 “감사합니다(thank you)” 같은 존칭을 사용하지 않습니다. 이는 “언젠가 AGI(General Artificial Intelligence)가 세상을 지배할 때까지 내 인간성을 지키겠다”는 그의 신념에서 비롯됩니다.
대신 그는 AI에게 명확하고 직설적인 지시를 내립니다. AI가 원하는 결과를 내지 못할 때는 “우리 이미 이거 얘기했잖아. 기억을 더듬어봐(Go search your memory)“라고 상기시킵니다. 이는 마치 아이들을 가르치는 부모의 태도와 비슷합니다. 로스는 “최고의 의도 가정(assume best intentions)“이라는 링키드인(LinkedIn)의 기업 문화를 AI와의 상호작용에도 적용합니다. AI가 실수를 하더라도 “최고의 의도를 가지고 있지만, 우리가 어떻게 일하는지 상기시켜줘야 한다”고 생각하며 인내심을 가지고 지시합니다.
[결론] 댄 로스의 이야기는 AI 시대에 비기술적 배경을 가진 사람들도 충분히 혁신적인 빌더(builder)가 될 수 있음을 보여줍니다. ‘밥’과 ‘레이’라는 듀얼 AI 에이전트 시스템, ‘바이브 코딩’ 철학, 그리고 AI를 효과적으로 관리하는 그의 노하우는 개발의 문턱을 낮추고 개인의 창의력을 극대화하는 새로운 가능성을 제시합니다. 그의 경험은 기술적 배경을 넘어 모든 이들에게 AI를 활용하여 자신만의 문제를 해결하고 세상을 변화시킬 수 있는 영감을 줍니다.
댄 로스 팔로우하기:
- LinkedIn: Daniel Roth 검색
- 뉴스레터: Forward Deployed Editor (AI 학습 경험 및 실패 사례 공유)
“A War Within the War: Israel’s Bombardment of Lebanon” — New York Times Podcasts 기반 기사 원본 영상 보기
전쟁 속 또 다른 전쟁: 레바논을 휩쓰는 이스라엘의 폭격, 끝나지 않는 비극의 서막인가
지난 몇 주간 전 세계의 이목이 이란과 이스라엘, 그리고 미국 간의 광범위한 전쟁에 집중된 가운데, 레바논에서는 또 다른 전선이 급부상하며 더욱 치명적인 ‘전쟁 속의 전쟁’이 펼쳐지고 있습니다. 뉴욕타임즈 팟캐스트 ‘더 데일리’의 레이첼 아브람스(Rachel Abrams)는 베이루트 지국장 크리스티나 골드봄(Christina Goldbaum)과의 인터뷰를 통해, 이스라엘의 레바논 공습이 어떻게 전례 없는 수준으로 확대되고 있으며, 레바논 국민들이 느끼는 공포와 이 전쟁의 장기적인 함의에 대해 심층적으로 다루었습니다.
베이루트를 덮친 전쟁의 그림자: 일상과 공포의 공존
지난 2주간 레바논에서는 또 다른 대규모 전쟁의 서막과도 같은 상황이 전개되고 있습니다. 베이루트의 거리를 걷다 보면 도시 곳곳에서 이어진 공습으로 인해 코를 찌르는 매캐한 냄새가 진동하고, 머리 위로는 이스라엘 드론의 굉음이 끊이지 않습니다. 일부 지역에서는 건물이 온전히 보존되어 일상이 평화롭게 흐르는 듯 보이지만, 다른 지역에서는 대규모 피란 사태가 현실로 나타나고 있습니다.
크리스티나 골드봄 지국장은 “사람들은 개미떼처럼, 마치 심판의 날 같았다”고 당시의 혼란을 전하는 한 남성의 목소리를 인용하며, 수십만 명이 공습을 피해 집을 떠나야 했던 절박한 상황을 묘사했습니다. 정부가 학교를 대피소로 전환했음에도 공간이 부족해, 사람들은 주차장, 해변 산책로, 인도, 공원 등에 천막과 회색 담요로 임시방편의 텐트를 치고 살아가고 있습니다. 특히 헤즈볼라(Hezbollah)의 영향력이 강한 남부 베이루트의 디아(Dahieh) 지역은 이스라엘의 폭격으로 수많은 건물이 완전히 파괴되었습니다.
최근 며칠 사이에는 중앙 베이루트 내의 건물과 차량까지 이스라엘의 공격 목표가 되었고, 대피 명령이 내려지면서 사태는 더욱 악화되었습니다. 사람들은 총성으로 대피를 경고받으며 건물과 거리에서 필사적으로 벗어나려 애쓰고 있습니다. 현재까지 100만 명에 육박하는 레바논 국민이 피란길에 올랐으며, 이는 580만 인구의 국가에 엄청난 부담으로 작용합니다. 이스라엘의 공습으로 800명 이상이 사망했으며, 레바논 국민들은 불과 2년 만에 또다시 대규모 전쟁에 휘말리게 된 현실에 직면하고 있습니다.
이번 전쟁이 다른 이유: 지상 침공의 공포와 이스라엘의 강경한 수사
크리스티나 지국장은 이번 분쟁이 과거 이스라엘의 레바논 작전들과는 본질적으로 다르다고 설명합니다. 가장 큰 차이점은 1980년대 이후 볼 수 없었던 이스라엘의 대규모 지상 침공(ground invasion) 가능성에 대한 공포입니다. 국경에는 이스라엘군의 대규모 병력 증강이 이루어졌고, 이스라엘 고위 관리들의 수사는 전례 없이 강경합니다. 이스라엘 국방부 장관은 베이루트 남부 외곽 지역을 가자 지구에서 완전히 파괴된 도시인 칸 유니스(Khan Younis)에 비유하며, 곧 그와 같은 모습이 될 것이라고 경고했습니다. 그는 또한 레바논 정부가 헤즈볼라의 이스라엘 공격을 막지 못한다면, 이스라엘이 레바논 영토 내에서 직접 행동에 나설 것이라고 선언했습니다. 이는 분명한 수사적 수위의 격상이며, 레바논 국민들은 이번 전쟁이 과거보다 훨씬 더 거대한 규모가 될 것이라고 우려하고 있습니다.
더욱이 이스라엘은 이번에 처음으로 베이루트 남부 외곽 전 지역과 국경에서 25마일(약 40km)에 달하는 레바논 남부의 광범위한 지역에 대해 전면적인 대피 명령을 내렸습니다. 이는 2024년 전쟁에서도 볼 수 없었던 조치로, 이스라엘이 헤즈볼라를 “영원히 제거하겠다”는 목표를 가지고 있음을 시사합니다.
‘전쟁 속의 전쟁’의 발발: 이란과의 연계
현재 레바논의 분쟁은 이란과의 광범위한 전쟁과 밀접하게 연결되어 있습니다. 이란 전쟁이 시작된 후, 이란 지도자 아야톨라 하메이니(Ayatollah Khamenei)의 사망에 대한 보복으로 헤즈볼라가 이스라엘에 미사일을 발사했습니다. 이에 이스라엘은 레바논 남부, 동부, 그리고 베이루트 남부 지역을 공습하며 반격했습니다. 전문가들은 이스라엘이 헤즈볼라의 공격을 오랫동안 원했던 목표, 즉 헤즈볼라를 완전히 제거할 기회로 삼고 있다고 분석합니다.
이러한 맥락에서 현재 레바논에서 벌어지는 상황은 본질적으로 ‘전쟁 속의 전쟁’이며, 레바논에게는 훨씬 더 큰 결과를 초래할 수 있는 중요한 전쟁으로 인식되고 있습니다. 많은 레바논 국민들은 헤즈볼라가 이란 지도자의 사망이라는 외부적 요인 때문에 이 분쟁에 개입하기로 결정한 것에 대해 놀라움을 표하고 있습니다.
이스라엘-레바논-헤즈볼라: 오래된 갈등의 역사
이스라엘과 레바논, 그리고 헤즈볼라의 갈등은 수십 년에 걸친 복잡한 역사를 가지고 있습니다. 1980년대 레바논은 내전의 소용돌이에 휘말려 있었고, 레바논 남부의 팔레스타인 무장단체들이 이스라엘을 공격했습니다. 이에 이스라엘은 1982년 레바논 남부를 대규모로 침공하여 수도 베이루트까지 진격했고, 이후 거의 20년 동안 남부를 점령했습니다.
헤즈볼라는 바로 이러한 배경 속에서 1980년대에 시아파 무장단체이자 정치 운동으로 부상했습니다. 이들은 레바논 내 이스라엘의 존재에 저항하는 것을 목표로 삼았으며, 이 지역 시아 이슬람의 선봉인 이란의 지원을 받았습니다. 이스라엘의 남부 점령 기간 동안 헤즈볼라 대원들은 이스라엘군과 치열하게 싸웠고, 2000년 이스라엘이 레바논 남부에서 철수했을 때 이는 헤즈볼라의 주요 승리로 여겨졌습니다. 이를 계기로 헤즈볼라는 무장 부문을 유지하면서도 정치적 영향력을 확장하기 시작했습니다.
이후에도 양측 국경을 따라 긴장은 매우 높았고 주기적인 충돌이 발생했습니다. 그러다 2023년 10월, 이란의 지원을 받는 또 다른 단체인 하마스(Hamas)가 이스라엘을 공격한 후, 헤즈볼라는 하마스와의 연대를 표명하며 이스라엘 진지에 로켓을 발사했습니다. 이는 대규모 분쟁으로 확대되어 이스라엘은 레바논 남동부와 베이루트 상당 부분을 폭격했습니다. 이 과정에서 헤즈볼라 대원들이 소지했던 수천 개의 호출기(pager)가 폭발했고, 헤즈볼라의 오랜 지도자 하산 나스랄라(Hassan Nasrallah)가 사망하는 등 조직에 큰 타격을 입혔습니다.
겉으로는 약해 보였던 헤즈볼라, 그러나…
지난 한 해 동안 많은 사람들은 나스랄라의 사망, 호출기 사건, 이스라엘의 공격 등으로 인해 헤즈볼라가 심각하게 약화되었다고 생각했습니다. 그러나 크리스티나 지국장은 현재 상황이 그 인식이 틀렸음을 입증한다고 지적합니다. 이스라엘은 이란 전쟁에 전 세계의 이목이 쏠린 틈을 타, 미국과의 파트너십과 자국의 군사력에 고무되어 헤즈볼라라는 가장 강력한 적수를 제거할 기회를 포착했습니다.
예루살렘 지국에서 온 보도에 따르면, 이스라엘의 이번 레바논 공세는 수개월 동안 계획되었으며, 이스라엘은 헤즈볼라가 이스라엘을 공격할 기회를 기다리고 있었습니다. 비록 헤즈볼라가 타격을 입었을지라도, 그들은 여전히 싸울 능력을 갖추고 있습니다. 무기 비축량을 보충하기 위해 레바논 내에서 무기를 제조하고 조립했으며, 이란 혁명수비대(IRGC) 대원들이 레바논에 들어와 헤즈볼라의 군사력을 보강하고 통제력을 강화한 증거도 포착되었습니다. 지난 2주 동안 이미 2명의 이스라엘 군인이 사망하고 10여 명이 부상당했습니다. 이는 헤즈볼라가 지난 1년 동안 겉으로는 군사적으로 약해 보였지만, 실제로는 이스라엘과의 대규모 분쟁을 예상하고 재건 및 준비에 시간을 보냈음을 시사합니다.
헤즈볼라 제거의 복잡한 도전: ‘국가 안의 국가’
이스라엘의 목표가 헤즈볼라의 작전 능력을 완전히 제거하는 것이라면, 과연 이 시점에 그것이 가능할까요? 크리스티나 지국장은 이것이 수십 년간 이스라엘의 오랜 숙원 사업이었지만, 지금까지 헤즈볼라를 무장 해제시키거나 제거하는 데 성공하지 못했다고 말합니다. 그러나 이스라엘 관리들은 지금이 이 오랜 위협을 마침내 제거할 기회라고 보고 있습니다.
문제는 헤즈볼라가 단순히 이란의 지원을 받는 무장 단체가 아니라는 점입니다. 헤즈볼라는 레바논 사회, 특히 전체 인구의 3분의 1을 차지하는 시아파 공동체에 깊이 뿌리내린 매우 강력한 정치 및 사회 운동입니다. 수십 년 동안 헤즈볼라는 사실상 레바논 정부를 목 조르는 듯한 지배력(chokehold)으로 장악하여, 그들의 지지 없이는 어떤 정치적 결정도 내릴 수 없게 만들었습니다. 이들은 레바논에서 가장 강력한 정당이며 의회 의석을 보유하고 있습니다. 또한 학교와 병원 같은 광범위한 사회 서비스 네트워크를 운영하며 수만 명의 사람들을 고용하고 있습니다. 이처럼 헤즈볼라는 레바논 사회에 막대한 영향력을 행사하며 ‘국가 안의 국가(a state within a state)‘로 불릴 정도입니다.
변화하는 지지층: 피로감과 회의감
그러나 이제 헤즈볼라의 충성스러운 지지 기반마저도 이전과는 다른 시험대에 오르고 있습니다. 피란민의 대다수는 헤즈볼라에 동정적인 지역 출신들입니다. 이는 지난 2년간의 전쟁 부담을 헤즈볼라 지지자들이 직접 감당해야 했음을 의미합니다. 그들은 반복되는 피란 생활, 대피소에서의 삶, 고향 마을이 공습으로 초토화되는 것에 지쳐 있습니다. 집을 재건해도 다시 파괴될까 걱정하는 피로감이 만연합니다.
오랫동안 이스라엘에 저항하려는 헤즈볼라 지지층의 열렬한 이념적 열정이 존재했지만, 이제 그 헌신은 전쟁에 대한 피로감으로 바뀌고 있습니다. 일부 사람들은 헤즈볼라가 이 전쟁에 개입하기로 한 결정 자체에 의문을 제기하기 시작했습니다. 레바논 국민들이 이스라엘 공습으로 사망했을 때는 침묵하다가, 이란 지도자의 사망 후에야 이스라엘을 공격한 이유에 대해 질문하는 목소리가 나옵니다. 크리스티나 지국장이 만난 한 간호사는 헤즈볼라가 여전히 시아파 공동체의 ‘주요 수호자’인지, 아니면 2년 만에 두 번이나 나라를 전쟁으로 끌고 들어가 공동체가 감당해야 할 대가가 너무 큰 것인지에 대해 가족들과 토론하고 있었다고 전했습니다.
레바논 정부의 무능력 또한 상황을 악화시킵니다. 레바논 군대는 약하고 장비도 제대로 갖춰져 있지 않아 헤즈볼라에 미치지 못합니다. 정부가 너무 성급하게 헤즈볼라를 무장 해제하려 한다면, 레바논 군과 헤즈볼라 전사들 간의 충돌로 이어져 내전(civil war)을 촉발할 수 있다는 우려도 큽니다.
이스라엘의 궁극적인 목표와 미국의 역할
이스라엘이 헤즈볼라를 완전히 제거하겠다고 말하는 것은 과연 현실적일까요? 마치 미국이 아프가니스탄에서 탈레반(Taliban)을 완전히 제거하겠다고 말하는 것과 비슷한, 비현실적인 목표일 수 있습니다. 많은 사람들은 이스라엘이 헤즈볼라를 주기적으로 약화시키지만 완전히 파괴하지는 못할 것이라고 생각합니다.
그러나 이스라엘 국경을 따라 대규모 군사력 증강이 이루어지고 있어, 레바논 국민들은 이스라엘의 대규모 지상 침공과 레바논 남부의 광범위한 지역 점령 가능성을 극도로 우려하고 있습니다. 이는 사실상 이스라엘과 새로운 완충 지대(buffer zone)를 만드는 것과 다름없습니다. 이러한 시나리오는 이미 100만 명 이상의 피란민을 발생시킨 상황에서 더 큰 대규모 피란을 야기할 수 있습니다. 또한 이스라엘이 레바논 남부를 재점령하는 것은 과거 내전과 점령의 기억이 생생한 레바논 국민들에게 역사의 반복으로 다가옵니다. 이스라엘 입장에서는 남부 레바논에 완충 지대를 만드는 것이 네타냐후(Netanyahu) 총리의 정치적 목표를 달성하는 데 도움이 될 수 있습니다.
미국의 역할도 변화하고 있습니다. 과거 미국은 이스라엘의 헤즈볼라 제거 노력이 지역을 불안정하게 만들고 더 큰 전쟁으로 비화될 것을 우려하여 이스라엘의 행동을 자제시키려 노력했습니다. 그러나 지금은 미국이 이스라엘의 레바논 폭격을 사실상 용인하고 있습니다. 동시에 미국은 레바논 정부에 헤즈볼라를 무장 해제하고 테러 단체로 선포하는 등 더욱 단호한 조치를 취하도록 막대한 압력을 가하고 있습니다. 이는 레바논 관리들이 그러한 조치가 국가를 매우 불안정하게 만들 수 있다고 우려함에도 불구하고 이루어지고 있습니다. 과거 이스라엘의 레바논 내 행동에 대한 미국의 견제 역할은 사라지거나 희미해진 듯합니다.
장기적인 전망: 끝나지 않는 비극의 예감
레바논의 전쟁은 이란과의 광범위한 전쟁과 분명히 연결되어 있습니다. 헤즈볼라는 이란의 지원을 받는 무장 단체이며, 이란 지도자의 사망에 대한 보복으로 이스라엘을 공격했습니다. 이는 이란의 ‘저항의 축(axis of resistance)‘의 정점이며, 이스라엘에 대한 로켓 공격을 이란과 조율하여 전쟁의 두 번째 전선을 만들려 하고 있습니다.
이 분쟁의 단기적 목표는 헤즈볼라가 이스라엘군에 가능한 한 많은 고통과 사상자를 입혀 이스라엘이 전쟁을 계속할 정치적 비용을 높이는 것입니다. 그러나 장기적으로 볼 때, 레바논의 전쟁은 이란 전쟁보다 훨씬 오래 지속될 것이라는 우려가 커지고 있습니다. 이란 전쟁이 끝나면 이스라엘이 레바논으로 군사력을 집중할 것이라는 공포가 존재합니다.
최근 며칠간 이스라엘 관리들의 메시지가 가자 지구와 레바논에서 유사하게 나타나면서 새로운 공포가 확산되고 있습니다. 이스라엘군은 레바논 남부 대부분 지역에 광범위한 대피 명령을 내렸는데, 이는 가자 지구에서 보았던 것과 유사합니다. 또한 이스라엘 전투기들은 중앙 베이루트 상공에 전단을 뿌려 헤즈볼라를 무장 해제하고 레바논에 “새로운 현실(new reality)“이 도래할 것이라고 경고했습니다. 이스라엘 관리들은 헤즈볼라가 구급차와 민간 트럭을 무기 운반에 사용하고 있다고 주장하며, 이들 또한 더 이상 공격의 대상에서 벗어나지 않을 것이라고 암시하기 시작했습니다.
이러한 섬뜩한 유사성들은 이스라엘의 이번 전쟁 목표가 과거보다 훨씬 더 영구적일 수 있다는 새로운 공포를 레바논에 불러일으키고 있습니다. 이는 헤즈볼라와의 더 많은 충돌, 더 많은 피란민, 그리고 앞으로 수년간 더 많은 민간인 사상자가 발생할 가능성을 시사하며, 레바논 국민들은 또 다른 길고 치명적인 분쟁의 그림자 속에서 깊은 불안감에 휩싸여 있습니다.