YouTube Digest

March 13, 2026

Listen to this issue

English

Based on โ€œThis Founder is Making 1B+ Excel Workers 20x Faster | Meridian, John Lingโ€ from EO Watch the original video

Beyond the Grid: How John Ling is Turning One Billion Excel Users into AI Powerhouses

In the glass towers of Manhattan, the lights rarely go out. Inside, thousands of the worldโ€™s brightest young minds are hunched over glowing monitors, performing a ritual that has remained largely unchanged for thirty years: manual data entry in Microsoft Excel. They are building Leveraged Buyout (LBO) models, three-statement financials, and complex valuations, one cell at a time.

To John Ling, the co-founder and CEO of Meridian, this isnโ€™t just a grueling rite of passage for junior bankersโ€”itโ€™s a massive, untapped opportunity for a technological revolution.

โ€œI donโ€™t believe any person on the planet has spent a thousand hours trying to build financial models with AI,โ€ Ling says. โ€œIf you think about the bankers, theyโ€™re just like, โ€˜Weโ€™re just going to do it by hand.โ€™ And if you donโ€™t know how to do it, you probably just donโ€™t know how to do it.โ€

Ling is on a mission to change that. Backed by over $15 million in funding led by Andreessen Horowitz, his startup, Meridian, is building the โ€œAI for spreadsheets.โ€ His goal is bold: to take the one billion people who live in Excel and make them 20 times faster.

The First Principles of a High Performer

Lingโ€™s path to Meridian wasnโ€™t accidental. It was forged at Scale AI, the data-labeling juggernaut where he earned a reputation as a โ€œtop 1% performer.โ€ At Scale, Ling was known for a relentless, first-principles approach to business. He didnโ€™t just stay in his lane; he treated every problem in the company as his own, viewing โ€œmore workโ€ as simply โ€œmore opportunities to learn.โ€

โ€œI think I just really enjoyed learning new things,โ€ Ling explains. โ€œThere are 50 problems, and with each problem, you learn a little bit more about something completely different.โ€

This intellectual curiosity allowed him to witness the development of Large Language Models (LLMs) from the front row. He spent his days obsessing over data quality, benchmarks, and internal efficiencies. It was here that he realized the trajectory of the world was about to changeโ€”and that most people were still looking in the wrong direction.

The โ€œVibe Codingโ€ Revelation

The spark for Meridian came from a personal โ€œahaโ€ moment. Ling began using Cursor, an AI-powered code editor, and watched his productivity explode. Tasks that previously took two weeks were suddenly being completed in thirty minutes. He calls this โ€œvibe codingโ€โ€”a state where the barrier between thought and execution vanishes, mediated by a โ€œsuper-powerful calculator.โ€

โ€œI had a moment where I was just like, โ€˜Wow, this thing is magical,โ€™โ€ Ling says. โ€œI wanted everyone to use it. I felt like if you donโ€™t know how to vibe code, youโ€™re going to be left behind.โ€

However, when he traveled from the tech-heavy atmosphere of San Francisco to New York, he found a jarring disconnect. While SF was buzzing with โ€œcrazy architecturesโ€ and AI unlocks, the finance world in New York was still operating on manual labor.

The reason, Ling realized, was a lack of decomposition. In coding, developers build the tools they use; they know exactly where a model fails and why. In finance, the workflows are so specialized and opaque that most people assume AI simply โ€œcanโ€™t do it.โ€ Ling disagreed. He saw that if you break a complex financial workflow into its constituent parts, AI can handle many of them exceptionally wellโ€”if someone is willing to do the investigation.

Excel: The Worldโ€™s Most Distributed Programming Language

At the heart of Meridianโ€™s vision is a fundamental reclassification of what a spreadsheet actually is. Ling doesnโ€™t see Excel as just a grid for numbers; he sees it as the most widely used programming language on Earth.

โ€œOur goal is to say, โ€˜Hey, how can we help all the people that spend a lot of time in spreadsheet software today move 20 times faster?โ€™โ€

Meridian isnโ€™t just about automating a few formulas. Itโ€™s about โ€œinfusing the work done in spreadsheets with meaningful intelligence.โ€ By augmenting the knowledge worker in the same way AI has augmented the software developer, Meridian aims to transform the very nature of white-collar work.

The market is staggering. Knowledge work is one of the largest sectors of the global economy, and spreadsheets are its primary tool. If Meridian can successfully apply the โ€œCursor modelโ€ to the financial world, they arenโ€™t just building a toolโ€”they are redefining a profession.

The Philosophy of โ€œThe Askโ€ and the Power of Failure

Lingโ€™s leadership style is as unconventional as his product. He promotes an environment where experimentation is mandatory and failure is a recognized part of the process.

โ€œYou learn by trying things youโ€™ve never tried before. And if you [mess] up, you [mess] up. Itโ€™s okay,โ€ he says. At Meridian, if an employee fails at a task, the response isnโ€™t a reprimand; itโ€™s a surge of support from the rest of the team to find a new way forward.

This โ€œsuspension of disbeliefโ€ is what Ling believes separates entrepreneurs from everyone else. Itโ€™s the same mindset that leads him to encourage others to reach out to anyoneโ€”even industry titans.

โ€œDonโ€™t think that Satya Nadella will never respond to your email,โ€ Ling advises. โ€œIf you think that way, he obviously never will. But if you reach out, you might be surprised.โ€

The Future: 10,000 Hours of AI Finance

As Meridian grows, Ling remains obsessed with the idea of โ€œintuition.โ€ He believes that by spending thousands of hours playing with the technology, one develops a โ€œsixth senseโ€ for what AI will be able to do three, six, or twelve months down the line.

He also offers a surprising insight into the value of prompting. While many see prompting as a chore, Ling sees it as a tool for self-clarity. Much like the Y Combinator application process forces founders to articulate their business, explaining a task to an LLM forces a human to actually understand what they want to achieve.

โ€œLearning how to be specific about your ask gives yourself a lot of clarity,โ€ Ling says.

For the one billion people currently โ€œdoing it by hand,โ€ that clarityโ€”combined with Meridianโ€™s technologyโ€”might just be the key to the next great leap in human productivity. John Ling isnโ€™t just building a company; heโ€™s trying to build a masterpiece that transforms how the world thinks, one cell at a time.


Based on โ€œFrom Coder to Manager: Navigating the Shift to Agentic Engineering with Notion Co-Founder Simon Lastโ€ from No Priors: AI, Machine Learning, Tech, & Startups Watch the original video

The Agentic Office: How Notion is Reimagining Work for a World of Autonomous Agents

In late 2022, during a company offsite in Mexico, Simon Last and his Notion co-founder Ivan Zhao sat down with an early version of GPT-4. It wasnโ€™t just a better chatbot; it was a revelation. For the founders of a company built on the philosophy of โ€œtools for thought,โ€ the realization was instantaneous: the nature of โ€œthoughtโ€ in the digital workspace was about to change forever.

โ€œIt was immediately clear that the time was now,โ€ Last recalls. โ€œIt was smart enough to follow complicated instructions, and the scope of its knowledge was incredibly deep. We knew it was only going to get better.โ€

That moment sparked a fundamental pivot for Notion. What began as a versatile workspace for humans to organize their notes and databases is rapidly evolving into a collaborative platform where humans and AI agents work side-by-side. In a wide-ranging conversation on the No Priors podcast, Last outlined how this โ€œagenticโ€ shift is rewriting the rules of engineering, product design, and personal productivity.

From Writing Assistant to General Assistant

Notionโ€™s AI journey began with what Last calls the โ€œshort-term visionโ€: the AI Writer. Launched in early 2023, it served as a sophisticated editing partnerโ€”rewriting paragraphs, changing tones, and summarizing meeting notes. It was a โ€œsingle-stepโ€ task that required no external memory.

But the โ€œlong-term visionโ€ was far more ambitious. Notion wanted to build a general assistant capable of using the platform exactly like a human does: querying databases, creating documents, and weaving together disparate pieces of information to complete complex, multi-day tasks.

The bridge between these two visions was โ€œQ&A,โ€ a feature that performs a semantic index of a userโ€™s entire workspace. Unlike traditional keyword searches, which often fail in the messy reality of corporate wikis, semantic indexing allows the AI to understand the meaning of a query.

โ€œThe AI doesnโ€™t really care what the tree structure of your folders is,โ€ Last explains. โ€œAll it cares about is finding the snippet of text that has the context you need. We tell people now: donโ€™t worry as much about organization. Just get the data in there.โ€

The โ€œSwitzerlandโ€ of Models

One of Notionโ€™s core strategies in the AI arms race is neutrality. Rather than tethering itself to a single provider, Notion views itself as the โ€œSwitzerland for models.โ€

Whether itโ€™s OpenAIโ€™s latest frontier model, Anthropicโ€™s Claude, or high-performing open-source models from China, Notionโ€™s infrastructure is designed to be model-agnostic. This allows users to swap the โ€œbrainโ€ of their assistant as technology evolves, ensuring they arenโ€™t locked into a provider that might be leapfrogged next month.

โ€œOur customers donโ€™t want to be locked in,โ€ says Last. โ€œWe want to be the place where you can get access to all the best models at any time, in a collaborative workspace that is designed for humans and agents to coordinate.โ€

Designing for a New Kind of User: The Agent

Perhaps the most surprising technical shift at Notion is the realization that their software now has a new type of customer: the AI itself.

Historically, Notionโ€™s internal data structures (APIs) were designed for humans writing code. They were verbose, complex, and โ€œhorrible for an agent,โ€ according to Last. To solve this, the engineering team had to rethink their primitives. They developed a specific โ€œMarkdown dialectโ€ that translates Notionโ€™s complex block-based structure into a format LLMs can read and write naturally. They even integrated SQLite interfaces to help agents interact with databases more efficiently.

โ€œWe essentially have a new customer, which is the agent,โ€ Last says. โ€œWe had to take on the engineering challenge of making our APIs convenient for them. Now, the agents are naturally good at it.โ€

The Death of Manual Coding

The shift to โ€œagentic engineeringโ€ isnโ€™t just something Notion sells; itโ€™s how they work. Last, a co-founder of a multi-billion dollar tech company, admits he has undergone a radical personal transformation.

โ€œI havenโ€™t written code since last summer,โ€ Last reveals. โ€œI donโ€™t type code anymore. Iโ€™ve gone from being a coder to an agent manager.โ€

In the modern Notion engineering workflow, humans no longer spend their days typing out syntax. Instead, they design end-to-end tasks, set up verification loops, and act as the โ€œouter verifierโ€ to ensure the agentโ€™s output is correct. This hasnโ€™t necessarily reduced team sizes, but it has drastically widened the gap between the median engineer and the โ€œ100xโ€ engineer who can effectively harness these tools.

Last shares a personal record: he once had a coding agent running for 13 days straight, autonomously working through a backlog of tasks while he monitored its progress. โ€œMy goal these days is to have as many agents running as possible,โ€ he says. โ€œEvery night before I go to bed, I make sure Iโ€™ve given them enough work so that theyโ€™re still spinning when I wake up.โ€

The Future: Custom Agents for Every Task

The latest milestone in this evolution is the launch of Custom Agents. Unlike a general personal assistant, a Custom Agent can be assigned a specific โ€œjobโ€โ€”like a digital employee.

Last uses several himself. One is an email triage agent that has access to his work and personal accounts. Through a process of โ€œinterviewsโ€ and feedback, the agent learned his preferences and now automatically archives 95% of his inbox, leaving only the essential messages. Another agent lives in a Slack channel, listening for bug reports and product feedback, then autonomously routing them to the correct engineering teamโ€™s database.

โ€œThe concept is very intuitive once you get past the technical barrier,โ€ Last notes. โ€œItโ€™s a very human-like interface.โ€

A New Philosophy of Work

Before the AI revolution, Notionโ€™s mission was to be the best tool for humans to perform their work. Today, that mission has been updated: Notion aims to be the best tool for humans to manage agents who do the work for them.

While the mission has shifted, the core building blocks of Notionโ€”the pages, the databases, the Kanban boardsโ€”remain as relevant as ever. In a world of autonomous digital workers, humans still need a โ€œcoordination structureโ€ to see what is happening.

โ€œIf youโ€™re working with a swarm of 100 background coding agents, you donโ€™t want 100 chat threads,โ€ Last concludes. โ€œYou want a Kanban board. Itโ€™s the same coordination structure as before, just with a much faster set of workers.โ€


Based on โ€œPalantir CEO on Iran, AI Weapons and Americaโ€™s Advantage | a16z American Dynamism Summitโ€ from a16z Watch the original video

The Architect of Deterrence: Alex Karp on AI, the โ€œFreak Show,โ€ and the Moral Necessity of American Might

For two decades, Palantir was the โ€œfreak showโ€ of Silicon Valley. While other startups were busy optimizing ad clicks or disrupting laundry services, Alex Karp and his team were embedded in the dark, unfashionable corners of the world: intelligence agencies, war zones, and the logistical labyrinths of the Department of Defense.

At the fourth annual a16z American Dynamism Summit, Karp, the CEO of Palantir and the โ€œOGโ€ of the movement to relink technology with national interest, didnโ€™t look like a man seeking vindication. He looked like a man who had already found it. In a wide-ranging conversation, Karp laid out a stark vision for the future: one where American military superiority is the only thing standing between Western liberalism and a new era of authoritarian dominance.

The Return of Deterrence

The conversation began against a backdrop of escalating global tension. Addressing the volatile state of the Middle East and the necessity of Western intervention, Karp was unapologetic. For years, he argued, American deterrence had been โ€œeviscerated.โ€ Now, through a combination of raw courage and cutting-edge technology, that deterrence is being rebuilt in real-time.

โ€œWe are the power that actually has the decisive vote,โ€ Karp stated, โ€œand that is with military superiority.โ€

He pointed to recent operations as evidence of a new technological reality. In Karpโ€™s view, the dominance displayed by the U.S. and its allies isnโ€™t just about having more boots on the ground; itโ€™s about a โ€œtrinityโ€ of software, hardware, and AI. This isnโ€™t the โ€œparasiticโ€ software of the pastโ€”the kind Karp mocks as โ€œsupplying a steak dinnerโ€ to bureaucratsโ€”but a specialized orchestration of data that allows the West to see, decide, and act faster than any adversary.

โ€œI literally believe weโ€™re doing the work of a higher purpose,โ€ Karp said. โ€œItโ€™s us, or China, or Russia. I donโ€™t know how you feel about those decisions, but I believe we are making sure we have the decisive vote.โ€

The Silicon Valley Blind Spot

Karpโ€™s sharpest barbs were reserved for his neighbors in Palo Alto. He identified a dangerous disconnect between the tech elite and the rest of the country. Silicon Valley, he argued, is obsessed with โ€œpositive-sumโ€ gamesโ€”the idea that everyone can win through innovation. But in the realm of global power, Karp insists the game is โ€œzero-sum.โ€

He issued a dire warning to the creators of Large Language Models (LLMs) and white-collar automation: if the tech industry is perceived as a force that destroys domestic jobs while refusing to support the American warfighter, the backlash will be existential.

โ€œIf Silicon Valley believes we are going to take away everyoneโ€™s white-collar jobโ€ฆ and youโ€™re going to screw the military, if you donโ€™t think thatโ€™s going to lead to the nationalization of our technology, youโ€™re [expletive],โ€ Karp warned.

He described a โ€œhorseshoe effectโ€ where the far left and the far right eventually agree on only one thing: that the tech industry is not โ€œpaying the billsโ€ for society and should be seized by the state. To avoid this, Karp argues that the tech elite must stop being โ€œeffing spoiledโ€ and start empathizing with the โ€œsoldier from Iowaโ€ who risks everything to protect the system that allowed them to get rich.

The Meritocracy of the Battlefield

One of the more surprising moments of the talk was Karpโ€™s defense of the Department of Defense as a social institution. He described the military as the most meritocratic environment in Americaโ€”a place that integrated long before the rest of society and remains the only institution revered across every demographic.

โ€œIf you were a Black American, you got your break in America by going to the military,โ€ Karp noted. โ€œItโ€™s the only institution revered by the American peopleโ€ฆ precisely because itโ€™s been meritocratic.โ€

For Karp, the moral mission of Palantir is simple: ensure that the American warfighter is the most likely person to come home safely, and that those trying to harm them know they wonโ€™t. This isnโ€™t just about lethality; itโ€™s about the preservation of a culture that allows for dissent, individuality, and freedom.

The Neurodivergent Advantage

Karp, who is famously open about his own dyslexia and โ€œintrovertedโ€ nature, linked Americaโ€™s technological edge directly to its tolerance for โ€œfreaks.โ€ He views the United States as a sanctuary for the โ€œneurologically divergentโ€โ€”those whose minds donโ€™t fit the standard mold but who possess the โ€œoutlier IQโ€ necessary to build world-changing technology.

โ€œYou could define our wonderful country as a place where everyone who was divergent in ideology, thought, religion, or neurologically came to have a better place where they could express their freedom,โ€ he said.

This philosophy extends to the culture at Palantir. Karp describes his role as an โ€œartist,โ€ managing a collection of brilliant, often difficult individuals who wouldnโ€™t thrive in a traditional corporate environment. He pushes back against โ€œwokeโ€ corporate culture, which he views as a performance of being different while everyone actually thinks the same.

โ€œI tend to gravitate towards people who are unique,โ€ Karp explained. โ€œI donโ€™t really care about their politics. I care about their ability to think and do.โ€

The โ€œMarkโ€ in the Room

As the summit concluded, Karp offered a piece of advice for the new generation of โ€œAmerican Dynamismโ€ foundersโ€”those building the next generation of defense tech. His warning was against the arrogance of intelligence.

โ€œThe single biggest mistake people make in this area is because theyโ€™re intelligent in one area, they assume theyโ€™re intelligent in all areas,โ€ Karp said. โ€œIf you donโ€™t know who the mark is, youโ€™re the mark.โ€

For Karp, success in the high-stakes world of national defense requires more than just a high IQ; it requires the humility to talk to a general or a soldier and realize that their expertise is just as vital as a line of code.

Palantir may have started as a โ€œfreak show,โ€ but in Alex Karpโ€™s eyes, it has become the blueprint for how America wins the 21st century: by embracing its outliers, arming its soldiers with the best software on earth, and never apologizing for holding the decisive vote.


ํ•œ๊ตญ์–ด

โ€œThis Founder is Making 1B+ Excel Workers 20x Faster | Meridian, John Lingโ€ โ€” EO ๊ธฐ๋ฐ˜ ๊ธฐ์‚ฌ ์›๋ณธ ์˜์ƒ ๋ณด๊ธฐ

โ€์—‘์…€ ๋…ธ๋™์ž 10์–ต ๋ช…์„ ์œ„ํ•œ ํ•ด๋ฐฉ๊ตฐโ€ โ€“ ๋ฉ”๋ฆฌ๋””์•ˆ ์กด ๋ง์ด ๊ทธ๋ฆฌ๋Š” AI ์Šคํ”„๋ ˆ๋“œ์‹œํŠธ์˜ ๋ฏธ๋ž˜

์ „ ์„ธ๊ณ„์—์„œ ๊ฐ€์žฅ ๋„๋ฆฌ ์“ฐ์ด๋Š” ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์–ธ์–ด๋Š” ๋ฌด์—‡์ผ๊นŒ? ํŒŒ์ด์ฌ(Python)๋„, ์ž๋ฐ”(Java)๋„ ์•„๋‹ˆ๋‹ค. ๋ฐ”๋กœ ๋งˆ์ดํฌ๋กœ์†Œํ”„ํŠธ์˜ **์—‘์…€(Excel)**์ด๋‹ค. ์ „ ์„ธ๊ณ„ 10์–ต ๋ช… ์ด์ƒ์˜ ์‚ฌ์šฉ์ž๊ฐ€ ๋งค์ผ ์Šคํ”„๋ ˆ๋“œ์‹œํŠธ ์œ„์—์„œ ์ˆซ์ž๋ฅผ ๋‹ค๋ฃจ๊ณ  ๋น„์ฆˆ๋‹ˆ์Šค ์˜์‚ฌ๊ฒฐ์ •์„ ๋‚ด๋ฆฐ๋‹ค. ํ•˜์ง€๋งŒ ๊ธฐ์ˆ ์˜ ์ตœ์ „์„ ์ธ ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ์—์„œ ์ฝ”๋”ฉ ๋ฐฉ์‹์ด ํ˜๋ช…์ ์œผ๋กœ ๋ณ€ํ•˜๋Š” ๋™์•ˆ, ๊ธˆ์œต๊ณผ ์ปจ์„คํŒ…์˜ ์ค‘์‹ฌ์ง€์ธ ์›”์ŠคํŠธ๋ฆฌํŠธ์˜ ํ’๊ฒฝ์€ ์ˆ˜์‹ญ ๋…„ ์ „๊ณผ ํฌ๊ฒŒ ๋‹ค๋ฅด์ง€ ์•Š๋‹ค.

์Šค์ผ€์ผ AI(Scale AI) ์ถœ์‹ ์˜ ์—ฐ์‡„ ์ฐฝ์—…๊ฐ€์ด์ž **๋ฉ”๋ฆฌ๋””์•ˆ(Meridian)**์˜ ๊ณต๋™ ์ฐฝ์—…์ž์ธ ์กด ๋ง(John Ling) CEO๋Š” ์ด ์ง€์ ์— ์ฃผ๋ชฉํ–ˆ๋‹ค. ๊ทธ๋Š” ์™œ ๋ชจ๋‘๊ฐ€ ์ˆ˜์ž‘์—…์œผ๋กœ ๊ธˆ์œต ๋ชจ๋ธ์„ ๋งŒ๋“œ๋Š” ๋ฐ ์ˆ˜๋ฐฑ ์‹œ๊ฐ„์„ ํ—ˆ๋น„ํ•˜๊ณ  ์žˆ๋Š”์ง€ ์˜๋ฌธ์„ ๋˜์ง„๋‹ค. ์•ˆ๋“œ๋ ˆ์„ผ ํ˜ธ๋กœ์œ„์ธ (a16z)๋กœ๋ถ€ํ„ฐ 1,500๋งŒ ๋‹ฌ๋Ÿฌ(์•ฝ 200์–ต ์›) ์ด์ƒ์˜ ํˆฌ์ž๋ฅผ ์œ ์น˜ํ•˜๋ฉฐ ์ฃผ๋ชฉ๋ฐ›๊ณ  ์žˆ๋Š” ๊ทธ๋ฅผ ํ†ตํ•ด, AI๊ฐ€ ๋ฐ”๊ฟ€ ์ง€์‹ ๋…ธ๋™์˜ ๋ฏธ๋ž˜๋ฅผ ๋“ค์—ฌ๋‹ค๋ณด์•˜๋‹ค.


1. ์Šค์ผ€์ผ AI์—์„œ์˜ ๊นจ๋‹ฌ์Œ: โ€œ๋ฐฐ์›€์€ ์—…๋ฌด์˜ ์—ฐ์žฅ์„ ์ด๋‹คโ€

์กด ๋ง์€ ๋ฉ”๋ฆฌ๋””์•ˆ์„ ์„ธ์šฐ๊ธฐ ์ „, ๋ฐ์ดํ„ฐ ๋ผ๋ฒจ๋ง ๋ฐ AI ์†”๋ฃจ์…˜ ๋ถ„์•ผ์˜ ์œ ๋‹ˆ์ฝ˜ ๊ธฐ์—…์ธ **์Šค์ผ€์ผ AI(Scale AI)**์—์„œ ํ•ต์‹ฌ ์ธ์žฌ๋กœ ํ™œ์•ฝํ–ˆ๋‹ค. ๊ทธ๊ณณ์—์„œ ๊ทธ๋Š” AI ๊ธฐ์ˆ ์ด ์„ธ์ƒ์„ ์–ด๋–ป๊ฒŒ ๋ฐ”๊ฟ€์ง€ ๊ฐ€์žฅ ๊ฐ€๊นŒ์ด์—์„œ ์ง€์ผœ๋ณด๋ฉฐ โ€˜์ œ1์›์น™(First Principles)โ€˜์— ๊ธฐ๋ฐ˜ํ•œ ์‚ฌ๊ณ ๋ฐฉ์‹์„ ์ตํ˜”๋‹ค.

โ€œ์ €๋Š” ์ƒˆ๋กœ์šด ๊ฒƒ์„ ๋ฐฐ์šฐ๋Š” ๊ณผ์ •์„ ์ •๋ง ์ฆ๊น๋‹ˆ๋‹ค. ์—…๋ฌด๊ฐ€ ๋งŽ์•„์ง„๋‹ค๋Š” ๊ฑด ๊ทธ๋งŒํผ ๋” ๋งŽ์ด ๋ฐฐ์šธ ๊ธฐํšŒ๊ฐ€ ์ƒ๊ธด๋‹ค๋Š” ๋œป์ด์—ˆ์ฃ .โ€

๊ทธ๋Š” ์Šค์ผ€์ผ AI์—์„œ ์ง๋ฌด์˜ ๊ฒฝ๊ณ„์— ๊ฐ‡ํžˆ์ง€ ์•Š์•˜๋‹ค. ๋ฐ์ดํ„ฐ์˜ ํ’ˆ์งˆ์„ ์ •์˜ํ•˜๊ณ , ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM)์˜ ๋ฒค์น˜๋งˆํฌ(์„ฑ๋Šฅ ์ธก์ • ๊ธฐ์ค€)๋ฅผ ์„ค์ •ํ•˜๋ฉฐ, ๋‚ด๋ถ€ ํ”„๋กœ์„ธ์Šค๋ฅผ AI๋กœ ํšจ์œจํ™”ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋Š์ž„์—†์ด ์—ฐ๊ตฌํ–ˆ๋‹ค. ๋™๋ฃŒ๋“ค์€ ๊ทธ๋ฅผ โ€˜์ƒ์œ„ 1%์˜ ์„ฑ๊ณผ์žโ€™๋กœ ๊ธฐ์–ตํ•œ๋‹ค. ๊ทธ๋Š” ๋‹จ์ˆœํžˆ ์ฃผ์–ด์ง„ ์ผ์„ ํ•˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ, โ€œ์ด ์ผ์„ ์™œ ํ•ด์•ผ ํ•˜๋Š”๊ฐ€?โ€๋ฅผ ๋Š์ž„์—†์ด ์งˆ๋ฌธํ•˜๋ฉฐ ๊ธฐ์ˆ ์˜ ๊ถค์ ์„ ์ถ”์ ํ–ˆ๋‹ค.


2. โ€˜๋ฐ”์ด๋ธŒ ์ฝ”๋”ฉ(Vibe Coding)โ€˜์˜ ์ถฉ๊ฒฉ: 2์ฃผ์˜ ์ž‘์—…์„ 30๋ถ„์œผ๋กœ

์กด ๋ง์ด AI์˜ ๋งˆ๋ฒ•์„ ์‹ค๊ฐํ•œ ์ˆœ๊ฐ„์€ ์ฝ”๋”ฉ ๋„๊ตฌ์ธ โ€˜์ปค์„œ(Cursor)โ€˜๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด์„œ๋ถ€ํ„ฐ์˜€๋‹ค. ์†Œ์œ„ **โ€˜๋ฐ”์ด๋ธŒ ์ฝ”๋”ฉ(Vibe Coding)โ€˜**์ด๋ผ ๋ถˆ๋ฆฌ๋Š” ์ด ๋ฐฉ์‹์€ ๊ฐœ๋ฐœ์ž๊ฐ€ ๋ณต์žกํ•œ ๋ฌธ๋ฒ•์— ์–ฝ๋งค์ด์ง€ ์•Š๊ณ , AI์™€ ๋Œ€ํ™”ํ•˜๋ฉฐ ์ง๊ด€์ ์œผ๋กœ ์ฝ”๋”ฉํ•˜๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•œ๋‹ค.

โ€œ๊ณผ๊ฑฐ์— 2์ฃผ์ผ์ด ๊ฑธ๋ฆฌ๋˜ ์ž‘์—…์ด ๋‹จ 30๋ถ„, ๊ธธ์–ด์•ผ ๋ฐ˜๋‚˜์ ˆ ๋งŒ์— ๋๋‚˜๋Š” ๊ฒƒ์„ ๋ณด์•˜์Šต๋‹ˆ๋‹ค. ์ด๊ฑด ๋‹จ์ˆœํ•œ ๋„๊ตฌ๊ฐ€ ์•„๋‹ˆ๋ผ โ€˜์Šˆํผ ๊ณ„์‚ฐ๊ธฐโ€™๊ฐ€ ๋“ฑ์žฅํ•œ ๊ฒƒ๊ณผ ๊ฐ™์•˜์ฃ .โ€

ํ•˜์ง€๋งŒ ๊ทธ๋Š” ๋‰ด์š•์˜ ๊ธˆ์œต๊ฐ€ ์นœ๊ตฌ๋“ค์„ ๋งŒ๋‚  ๋•Œ๋งˆ๋‹ค ๊ธฐ๋ฌ˜ํ•œ ๊ดด๋ฆฌ๊ฐ์„ ๋А๊ผˆ๋‹ค. ์ƒŒํ”„๋ž€์‹œ์Šค์ฝ”์˜ ๊ฐœ๋ฐœ์ž๋“ค์€ AI๋กœ ์ƒ์‚ฐ์„ฑ์„ ์ˆ˜์‹ญ ๋ฐฐ์”ฉ ๋†’์ด๊ณ  ์žˆ๋Š”๋ฐ, ๋‰ด์š•์˜ ์€ํ–‰๊ฐ€๋“ค์€ ์—ฌ์ „ํžˆ ์ˆ˜์ž‘์—…์œผ๋กœ LBO(์ฐจ์ž…๋งค์ˆ˜, Leveraged Buyout) ๋ชจ๋ธ์„ ๋งŒ๋“ค๋ฉฐ ๋ฐค์„ ์ง€์ƒˆ์šฐ๊ณ  ์žˆ์—ˆ๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค.


3. โ€œ์•„๋ฌด๋„ AI๋กœ ๊ธˆ์œต ๋ชจ๋ธ์„ 1,000์‹œ๊ฐ„ ๋™์•ˆ ๋งŒ๋“ค์–ด๋ณด์ง€ ์•Š์•˜๋‹คโ€

์กด ๋ง์€ ํ•ต์‹ฌ์ ์ธ ํ†ต์ฐฐ์„ ์ œ์‹œํ•œ๋‹ค. ์ฝ”๋”ฉ ๋ถ„์•ผ์—์„œ๋Š” ๊ฐœ๋ฐœ์ž๋“ค์ด ์Šค์Šค๋กœ๋ฅผ ์œ„ํ•œ AI ๋„๊ตฌ๋ฅผ ์ง์ ‘ ๋งŒ๋“ค๊ธฐ ๋•Œ๋ฌธ์— ๋ฐœ์ „์ด ๋น ๋ฅด์ง€๋งŒ, ๊ธˆ์œต์ด๋‚˜ ์ปจ์„คํŒ… ๊ฐ™์€ ์ „๋ฌธ ์˜์—ญ์—์„œ๋Š” ์ „๋ฌธ๊ฐ€๋“ค์ด AI๋ฅผ ๊นŠ์ด ์žˆ๊ฒŒ ํŒŒ๊ณ ๋“ค ์‹œ๊ฐ„์ด ์—†์—ˆ๋‹ค๋Š” ๊ฒƒ์ด๋‹ค.

โ€œ๋‹จ์–ธ์ปจ๋Œ€, ์ง€๊ตฌ์ƒ์—์„œ AI๋งŒ์„ ์ด์šฉํ•ด ๊ธˆ์œต ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ•˜๋Š” ๋ฐ 1,000์‹œ๊ฐ„์„ ์Ÿ์•„๋ถ€์€ ์‚ฌ๋žŒ์€ ๋‹จ ํ•œ ๋ช…๋„ ์—†์„ ๊ฒ๋‹ˆ๋‹ค. ์€ํ–‰๊ฐ€๋“ค์€ ๊ทธ์ € โ€˜์ง์ ‘ ์†์œผ๋กœ ํ•˜๋Š” ๊ฒŒ ๋น ๋ฅด๋‹คโ€™๊ณ  ์ƒ๊ฐํ•˜์ฃ . ํ•˜์ง€๋งŒ ์—…๋ฌด ํ๋ฆ„์„ ์„ธ๋ถ„ํ™”(Decomposition)ํ•ด๋ณด๋ฉด, AI๊ฐ€ ๋†€๋ผ์šธ ์ •๋„๋กœ ์ž˜ ํ•ด๋‚ผ ์ˆ˜ ์žˆ๋Š” ์˜์—ญ์ด ๋ถ„๋ช…ํžˆ ์กด์žฌํ•ฉ๋‹ˆ๋‹ค.โ€

๊ทธ๋Š” ์‚ฌ๋žŒ๋“ค์ด AI๊ฐ€ ๋‚ด๋†“์€ ๋‹ต์ด ํ‹€๋ ธ์„ ๋•Œ โ€œ์ด ๋ชจ๋ธ์€ ๋ชป ์“ฐ๊ฒ ๋„คโ€๋ผ๊ณ  ์‰ฝ๊ฒŒ ํฌ๊ธฐํ•˜๋Š” ๊ฒฝํ–ฅ์„ ์ง€์ ํ•œ๋‹ค. ์™œ ์ˆซ์ž๊ฐ€ ํ‹€๋ ธ๋Š”์ง€, ์–ด๋–ค ๋…ผ๋ฆฌ์  ๋‹จ๊ณ„์—์„œ ์˜ค๋ฅ˜๊ฐ€ ๋‚ฌ๋Š”์ง€ ์ง‘์š”ํ•˜๊ฒŒ ํŒŒ๊ณ ๋“œ๋Š” โ€˜์กฐ์‚ฌ(Investigation)โ€˜์˜ ๊ณผ์ •์ด ์—†์—ˆ๋‹ค๋Š” ๊ฒƒ์ด๋‹ค. ๋ฉ”๋ฆฌ๋””์•ˆ์€ ๋ฐ”๋กœ ์ด ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด ํƒ„์ƒํ–ˆ๋‹ค.


4. ๋ฉ”๋ฆฌ๋””์•ˆ์˜ ๋น„์ „: ์Šคํ”„๋ ˆ๋“œ์‹œํŠธ์˜ 20๋ฐฐ ๊ฐ€์†ํ™”

๋ฉ”๋ฆฌ๋””์•ˆ์˜ ๋ชฉํ‘œ๋Š” ๋ช…ํ™•ํ•˜๋‹ค. ์Šคํ”„๋ ˆ๋“œ์‹œํŠธ ์‚ฌ์šฉ์ž๋ฅผ ์œ„ํ•œ AI๋ฅผ ๊ตฌ์ถ•ํ•˜์—ฌ ๊ทธ๋“ค์˜ ์—…๋ฌด ์†๋„๋ฅผ 20๋ฐฐ ์ด์ƒ ๋†’์ด๋Š” ๊ฒƒ์ด๋‹ค.

  • ์—…๋ฌด์˜ ๋ถ„ํ•ด์™€ ์žฌ๊ตฌ์„ฑ: ๋ณต์žกํ•œ ๊ธˆ์œต ๋ชจ๋ธ๋ง ๊ณผ์ •์„ AI๊ฐ€ ์ฒ˜๋ฆฌ ๊ฐ€๋Šฅํ•œ ๋‹จ์œ„๋กœ ๋‚˜๋ˆ„์–ด ์ž๋™ํ™”ํ•œ๋‹ค.
  • ์ „๋ฌธ ์ง€์‹์˜ ์ด์‹: ์ฝ”๋”ฉ ๋„๊ตฌ๊ฐ€ ๊ฐœ๋ฐœ์ž์˜ ์˜๋„๋ฅผ ์ฝ๋“ฏ, ๋ฉ”๋ฆฌ๋””์•ˆ์€ ์žฌ๋ฌด ๋ถ„์„๊ฐ€๋‚˜ ์ปจ์„คํ„ดํŠธ์˜ ๋…ผ๋ฆฌ๋ฅผ ์ดํ•ดํ•˜๋Š” AI๋ฅผ ์ง€ํ–ฅํ•œ๋‹ค.
  • ์ง€์‹ ๋…ธ๋™์˜ ๋ณ€ํ˜: ์—‘์…€์„ ์‚ฌ์šฉํ•˜๋Š” ์ „ ์„ธ๊ณ„ ์ˆ˜์–ต ๋ช…์˜ ์ง€์‹ ๋…ธ๋™์ž๋“ค์ด ๋‹จ์ˆœ ๋ฐ˜๋ณต ์ž‘์—…์—์„œ ๋ฒ—์–ด๋‚˜ ๋” ๊ฐ€์น˜ ์žˆ๋Š” ์˜์‚ฌ๊ฒฐ์ •์— ์ง‘์ค‘ํ•˜๊ฒŒ ๋งŒ๋“ ๋‹ค.

5. ์ฐฝ์—…๊ฐ€ ์ •์‹ : โ€œ์•ˆ ๋  ๊ฑฐ๋ผ๋Š” ๋ฏฟ์Œ์„ ๊ฑฐ๋ถ€ํ•˜๋ผโ€

์กด ๋ง์€ ์˜ˆ๋น„ ์ฐฝ์—…๊ฐ€๋“ค์—๊ฒŒ **โ€˜๋ถˆ์‹ ์„ ์œ ์˜ˆํ•˜๋Š” ๋Šฅ๋ ฅ(Suspension of Disbelief)โ€˜**์ด ํ•„์š”ํ•˜๋‹ค๊ณ  ๊ฐ•์กฐํ•œ๋‹ค. ๋‚จ๋“ค์ด โ€œ๊ทธ๊ฑด ๋ถˆ๊ฐ€๋Šฅํ•ดโ€, โ€œ๋ฏธ์นœ ์ง“์ด์•ผโ€๋ผ๊ณ  ๋งํ•  ๋•Œ, โ€œ์™œ ์•ˆ ๋ผ? ํ•œ๋ฒˆ ํ•ด๋ณด์žโ€๋ผ๊ณ  ๋งํ•  ์ˆ˜ ์žˆ๋Š” ์šฉ๊ธฐ๋‹ค.

๊ทธ๋Š” ์‚ฌํ‹ฐ์•„ ๋‚˜๋ธ๋ผ(๋งˆ์ดํฌ๋กœ์†Œํ”„ํŠธ CEO) ๊ฐ™์€ ๊ฑฐ๋ฌผ์—๊ฒŒ๋„ ์ด๋ฉ”์ผ์„ ๋ณด๋‚ด๋Š” ๊ฒƒ์„ ๋‘๋ ค์›Œํ•˜์ง€ ๋ง๋ผ๊ณ  ์กฐ์–ธํ•œ๋‹ค. โ€œ๋‹ต์žฅ์ด ์•ˆ ์˜ฌ ๊ฑฐ๋ผ๊ณ  ๋ฏฟ์œผ๋ฉด ๋‹น์—ฐํžˆ ์•ˆ ์˜ค๊ฒ ์ง€๋งŒ, ๋ณด๋‚ด๋ณด๋ฉด ์˜์™ธ์˜ ๊ฒฐ๊ณผ๊ฐ€ ์ƒ๊ธธ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹œ๋„ํ•˜์ง€ ์•Š์œผ๋ฉด ์•„๋ฌด ์ผ๋„ ์ผ์–ด๋‚˜์ง€ ์•Š์ฃ .โ€

๋˜ํ•œ, ์‹คํŒจ๋ฅผ ์šฉ์ธํ•˜๋Š” ๋ฌธํ™”๋ฅผ ๊ฐ•์กฐํ•œ๋‹ค. โ€œ์šฐ๋ฆฌ ํšŒ์‚ฌ๋Š” ์ง์›์ด ์Šค์Šค๋กœ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋ ค๋‹ค ์‹คํŒจํ•˜๋Š” ๊ฒƒ์„ ์žฅ๋ คํ•ฉ๋‹ˆ๋‹ค. ์‹คํŒจํ•˜๋ฉด ๋™๋ฃŒ๋“ค์ด ๋‹ฌ๋ ค๊ฐ€ ๋„์™€์ฃผ๋ฉด ๋ฉ๋‹ˆ๋‹ค. ์‹คํ—˜ํ•˜๊ณ  ์‹คํŒจํ•ด๋„ ๊ดœ์ฐฎ์€ ํ™˜๊ฒฝ์„ ๋งŒ๋“ค์–ด์•ผ ํ˜์‹ ์ด ์ผ์–ด๋‚ฉ๋‹ˆ๋‹ค.โ€


๊ฒฐ๋ก : AI ์‹œ๋Œ€, ์ธ๊ฐ„์—๊ฒŒ ํ•„์š”ํ•œ ๊ฒƒ์€ โ€˜๋ช…ํ™•ํ•œ ์งˆ๋ฌธโ€™

์กด ๋ง์€ AI์™€ ๋” ๋งŽ์€ ์‹œ๊ฐ„์„ ๋ณด๋‚ผ์ˆ˜๋ก ๊ธฐ์ˆ ์— ๋Œ€ํ•œ ์ง๊ด€์ด ์ƒ๊ธฐ๋ฉฐ, ์ด๋Š” 3๊ฐœ์›” ํ›„, 6๊ฐœ์›” ํ›„์˜ ๋ฏธ๋ž˜๋ฅผ ์˜ˆ์ธกํ•˜๋Š” ํž˜์ด ๋œ๋‹ค๊ณ  ๋งํ•œ๋‹ค. ํŠนํžˆ ๊ทธ๋Š” AI์—๊ฒŒ ์ž‘์—…์„ ์„ค๋ช…ํ•˜๋Š” ๊ณผ์ • ์ž์ฒด๊ฐ€ ์ธ๊ฐ„์—๊ฒŒ ํฐ ์œ ์ต์„ ์ค€๋‹ค๊ณ  ๊ฐ•์กฐํ•œ๋‹ค.

โ€œAI์—๊ฒŒ ๋ฌด์—‡์„ ์›ํ•˜๋Š”์ง€ ๊ตฌ์ฒด์ ์œผ๋กœ ์„ค๋ช…ํ•˜๋ ค๊ณ  ๋…ธ๋ ฅํ•˜๋‹ค ๋ณด๋ฉด, ์˜คํžˆ๋ ค ์ž๊ธฐ ๋จธ๋ฆฟ์†์ด ๋ช…ํ™•ํ•ด์ง‘๋‹ˆ๋‹ค. ๋‚ด๊ฐ€ ์ •๋ง๋กœ ๋ฌด์—‡์„ ํ•˜๋ ค๋Š”์ง€ ์Šค์Šค๋กœ ๊นจ๋‹ซ๊ฒŒ ๋˜๋Š” ๊ฒƒ์ด์ฃ .โ€

๋ฉ”๋ฆฌ๋””์•ˆ์ด ๊ทธ๋ฆฌ๋Š” ๋ฏธ๋ž˜๋Š” ๋‹จ์ˆœํžˆ ์—‘์…€ ์นธ์„ ์ž๋™์œผ๋กœ ์ฑ„์šฐ๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋‹ค. ๊ทธ๊ฒƒ์€ ์ธ๊ฐ„์˜ ๋ณต์žกํ•œ ๋…ผ๋ฆฌ๋ฅผ AI์™€ ํ˜‘์—…ํ•˜์—ฌ ๋” ๋น ๋ฅด๊ณ  ์ •ํ™•ํ•˜๊ฒŒ ํ˜„์‹คํ™”ํ•˜๋Š”, ์ƒˆ๋กœ์šด ์ฐจ์›์˜ ์ง€์‹ ๋…ธ๋™ ์‹œ๋Œ€๋ฅผ ์—ฌ๋Š” ๊ฒƒ์ด๋‹ค. 10์–ต ๋ช…์˜ ์—‘์…€ ์‚ฌ์šฉ์ž๋“ค์ด โ€˜20๋ฐฐ ๋น ๋ฅธ ์†๋„โ€™๋ฅผ ๊ฐ–๊ฒŒ ๋  ๋•Œ, ์„ธ์ƒ์€ ๊ณผ์—ฐ ์–ด๋–ค ๋ชจ์Šต์œผ๋กœ ๋ณ€ํ• ๊นŒ? ์กด ๋ง๊ณผ ๋ฉ”๋ฆฌ๋””์•ˆ์˜ ํ–‰๋ณด๊ฐ€ ๊ธฐ๋Œ€๋˜๋Š” ์ด์œ ๋‹ค.


โ€œFrom Coder to Manager: Navigating the Shift to Agentic Engineering with Notion Co-Founder Simon Lastโ€ โ€” No Priors: AI, Machine Learning, Tech, & Startups ๊ธฐ๋ฐ˜ ๊ธฐ์‚ฌ ์›๋ณธ ์˜์ƒ ๋ณด๊ธฐ

[๋งค๊ฑฐ์ง„ ํ…Œํฌ] ์ฝ”๋”์—์„œ ๊ด€๋ฆฌ์ž๋กœ: ๋…ธ์…˜ ๊ณต๋™ ์ฐฝ์—…์ž ์‚ฌ์ด๋จผ ๋ผ์ŠคํŠธ๊ฐ€ ๊ทธ๋ฆฌ๋Š” โ€˜์—์ด์ „ํŠธ ์—”์ง€๋‹ˆ์–ด๋งโ€™์˜ ๋ฏธ๋ž˜

โ€œ์ด์ œ ์ €๋Š” ์ฝ”๋“œ๋ฅผ ์ง์ ‘ ํƒ€์ดํ•‘ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹  ์—์ด์ „ํŠธ ๊ตฐ๋‹จ์„ ๊ด€๋ฆฌํ•˜์ฃ .โ€

์ „ ์„ธ๊ณ„ ์ˆ˜์ฒœ๋งŒ ๋ช…์˜ ์ƒ์‚ฐ์„ฑ์„ ์ฑ…์ž„์ง€๋Š” ๋„๊ตฌ, ๋…ธ์…˜(Notion)์˜ ๊ณต๋™ ์ฐฝ์—…์ž ์‚ฌ์ด๋จผ ๋ผ์ŠคํŠธ(Simon Last)๊ฐ€ ๋˜์ง„ ์ด ํ•œ๋งˆ๋””๋Š” ์˜ค๋Š˜๋‚  ์†Œํ”„ํŠธ์›จ์–ด ์—”์ง€๋‹ˆ์–ด๋ง๊ณผ ์ƒ์‚ฐ์„ฑ ๋„๊ตฌ๊ฐ€ ์ง๋ฉดํ•œ ๊ฑฐ๋Œ€ํ•œ ํŒจ๋Ÿฌ๋‹ค์ž„ ๋ณ€ํ™”๋ฅผ ์ƒ์ง•ํ•ฉ๋‹ˆ๋‹ค. ํŒŸ์บ์ŠคํŠธ โ€˜No Priorsโ€™์— ์ถœ์—ฐํ•œ ๊ทธ๋Š” ๋…ธ์…˜์ด ๋‹จ์ˆœํ•œ ๋ฉ”๋ชจ ์•ฑ์„ ๋„˜์–ด ์ธ๊ฐ„๊ณผ AI ์—์ด์ „ํŠธ๊ฐ€ ํ˜‘์—…ํ•˜๋Š” ํ”Œ๋žซํผ์œผ๋กœ ์ง„ํ™”ํ•˜๋Š” ๊ณผ์ •๊ณผ, ๊ทธ ๊ณผ์ •์—์„œ ๋ชฉ๊ฒฉํ•œ ๊ธฐ์ˆ ์ ยท์กฐ์ง์  ๋ณ€ํ™”๋ฅผ ์‹ฌ๋„ ์žˆ๊ฒŒ ๊ณต์œ ํ–ˆ์Šต๋‹ˆ๋‹ค.


1. ๋ฉ•์‹œ์ฝ”์—์„œ์˜ ๊นจ๋‹ฌ์Œ: โ€œ์‹œ๊ฐ„์ด ์™”๋‹คโ€

๋…ธ์…˜์ด AI๋ฅผ ๋„์ž…ํ•˜๊ฒŒ ๋œ ๊ฒฐ์ •์ ์ธ ๊ณ„๊ธฐ๋Š” 2022๋…„ ๋ง, ๋ฉ•์‹œ์ฝ”์—์„œ ์—ด๋ฆฐ ํšŒ์‚ฌ ์›Œํฌ์ˆ์œผ๋กœ ๊ฑฐ์Šฌ๋Ÿฌ ์˜ฌ๋ผ๊ฐ‘๋‹ˆ๋‹ค. ๋‹น์‹œ ์‚ฌ์ด๋จผ๊ณผ ๊ณต๋™ ์ฐฝ์—…์ž ์ด๋ฐ˜ ์Šˆ(Ivan Zhao)๋Š” ์ถœ์‹œ ์ „์ด์—ˆ๋˜ GPT-4๋ฅผ ์ฒ˜์Œ ์ ‘ํ–ˆ์Šต๋‹ˆ๋‹ค.

์‚ฌ์ด๋จผ์€ ๋‹น์‹œ๋ฅผ ์ด๋ ‡๊ฒŒ ํšŒ์ƒํ•ฉ๋‹ˆ๋‹ค. โ€œ๋ณต์žกํ•œ ์ง€์‹œ๋ฅผ ์™„๋ฒฝํ•˜๊ฒŒ ๋”ฐ๋ฅด๋Š” ์ง€๋Šฅ, ๊ทธ๋ฆฌ๊ณ  ๋ฐฉ๋Œ€ํ•˜๊ณ  ๊นŠ์€ ์„ธ๊ณ„ ์ง€์‹์— ์ „์œจ์„ ๋А๊ผˆ์Šต๋‹ˆ๋‹ค. ๊ทธ ์ˆœ๊ฐ„ ์ด๋ฐ˜๊ณผ ์ €๋Š” ํ™•์‹ ํ–ˆ์ฃ . โ€˜์ง€๊ธˆ์ด ๋ฐ”๋กœ AI๋ฅผ ์ ์šฉํ•  ๋•Œโ€™๋ผ๋Š” ๊ฒƒ์„์š”.โ€

๋…ธ์…˜์€ ์ฆ‰์‹œ โ€˜ํƒ€์ด๊ฑฐ ํŒ€(Tiger Team, ํŠน์ • ๋ชฉํ‘œ๋ฅผ ์œ„ํ•ด ๊ตฌ์„ฑ๋œ ์†Œ์ˆ˜ ์ •์˜ˆ ํŒ€)โ€˜์„ ๊พธ๋ ค ๋‘ ๊ฐ€์ง€ ํŠธ๋ž™์˜ ๋น„์ „์„ ์„ค์ •ํ–ˆ์Šต๋‹ˆ๋‹ค.

  • ๋‹จ๊ธฐ ๋น„์ „: ๋ฌธ์„œ ์ž‘์„ฑ์„ ๋•๋Š” โ€˜AI ๋ผ์ดํ„ฐ(AI Writer)โ€™. (2023๋…„ 2์›” ์ถœ์‹œ)
  • ์žฅ๊ธฐ ๋น„์ „: ์ธ๊ฐ„์ฒ˜๋Ÿผ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ์ฟผ๋ฆฌํ•˜๊ณ , ๋ฌธ์„œ๋ฅผ ์ˆ˜์ •ํ•˜๋ฉฐ, ๋ณต์žกํ•œ ์—…๋ฌด๋ฅผ ์ž์œจ์ ์œผ๋กœ ์ˆ˜ํ–‰ํ•˜๋Š” โ€˜๋ฒ”์šฉ ์—์ด์ „ํŠธ(General Assistant)โ€˜.

2. ๊ฒ€์ƒ‰์˜ ๋ฏธํ•™: ์™œ ๋…ธ์…˜์€ ๋‹ค๋ฅธ ์„œ๋น„์Šค๋ณด๋‹ค ๊ฒ€์ƒ‰์„ ์ž˜ํ• ๊นŒ?

๋…ธ์…˜ AI์˜ ํ•ต์‹ฌ ๊ธฐ๋Šฅ ์ค‘ ํ•˜๋‚˜๋Š” โ€˜Q&Aโ€™์ž…๋‹ˆ๋‹ค. ์‚ฌ์šฉ์ž๊ฐ€ ์งˆ๋ฌธํ•˜๋ฉด ์›Œํฌ์ŠคํŽ˜์ด์Šค ๋‚ด์˜ ๋ฐฉ๋Œ€ํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ๊ฒ€์ƒ‰ํ•ด ๋‹ต์„ ๋‚ด๋†“๋Š” ๋ฐฉ์‹์ด์ฃ . ํฅ๋ฏธ๋กœ์šด ์ ์€ ๋…ธ์…˜์ด ์Šฌ๋ž™(Slack)์ด๋‚˜ ๊ตฌ๊ธ€ ๋“œ๋ผ์ด๋ธŒ(Google Drive) ๊ฐ™์€ ์™ธ๋ถ€ ํˆด๊นŒ์ง€ ์ธ๋ฑ์‹ฑ(Indexing)ํ•˜์—ฌ ํ†ตํ•ฉ ๊ฒ€์ƒ‰ ์„œ๋น„์Šค๋ฅผ ์ œ๊ณตํ•œ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

โ€œ์›๋ž˜ ๊ทธ ์„œ๋น„์Šค๋ฅผ ๋งŒ๋“  ํšŒ์‚ฌ๋“ค๋ณด๋‹ค ๋…ธ์…˜์ด ๊ฒ€์ƒ‰์„ ๋” ์ž˜ํ•˜๋Š” ๋น„๊ฒฐ์ด ๋ญ๋ƒโ€๋Š” ์งˆ๋ฌธ์— ์‚ฌ์ด๋จผ์€ **โ€˜์žฅ์ธ์ •์‹ (Craft)โ€˜**๊ณผ **โ€˜๊ฒฝํ—˜์  ๋ฐ˜๋ณต(Empirical Iteration)โ€˜**์„ ๊ผฝ์•˜์Šต๋‹ˆ๋‹ค.

โ€œ๋ฐ์ดํ„ฐ ์†Œ์Šค๋งˆ๋‹ค ํŠน์„ฑ์ด ๋‹ค๋ฆ…๋‹ˆ๋‹ค. ์Šฌ๋ž™์˜ ๋ฉ”์‹œ์ง€์™€ ๊ตฌ๊ธ€ ๋“œ๋ผ์ด๋ธŒ์˜ ๋ฌธ์„œ๋Š” ์™„์ „ํžˆ ๋‹ค๋ฅธ ํ˜•ํƒœ์ฃ . ์šฐ๋ฆฌ๋Š” ๋‹จ์ˆœํžˆ ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM)์„ ์—ฐ๊ฒฐํ•˜๋Š” ๋ฐ ๊ทธ์น˜์ง€ ์•Š๊ณ , ๋งค์ผ ์ง์ ‘ ์ฟผ๋ฆฌ๋ฅผ ๋˜์ ธ๋ณด๋ฉฐ ๊ฒ€์ƒ‰ ํ’ˆ์งˆ์„ ํŠœ๋‹ํ•ฉ๋‹ˆ๋‹ค. โ€˜AI์— ์ค‘๋…๋œ(AI-pilled)โ€™ ๊ฐ๊ฐ๊ณผ ๋””ํ…Œ์ผ์— ๋Œ€ํ•œ ์ง‘์ฐฉ์ด ์ฐจ์ด๋ฅผ ๋งŒ๋“ญ๋‹ˆ๋‹ค.โ€

์ด์ œ ๋…ธ์…˜์€ ์‚ฌ์šฉ์ž๋“ค์—๊ฒŒ โ€œ์ •๋ฆฌ์— ๋„ˆ๋ฌด ๊ณต๋“ค์ด์ง€ ๋งˆ๋ผโ€๊ณ  ์กฐ์–ธํ•ฉ๋‹ˆ๋‹ค. AI๊ฐ€ ์ž„๋ฒ ๋”ฉ(Embedding, ๋ฐ์ดํ„ฐ๋ฅผ ๋ฒกํ„ฐ ํ˜•ํƒœ๋กœ ๋ณ€ํ™˜ํ•˜๋Š” ๊ธฐ์ˆ )์„ ํ†ตํ•ด ๊ตฌ์กฐ์™€ ์ƒ๊ด€์—†์ด ํ•„์š”ํ•œ ๋งฅ๋ฝ์„ ์ฐพ์•„๋‚ผ ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.

3. ์—”์ง€๋‹ˆ์–ด๋ง์˜ ์ข…๋ง? ์•„๋‹ˆ, โ€˜1,000๋ฐฐ ์—”์ง€๋‹ˆ์–ดโ€™์˜ ํƒ„์ƒ

์‚ฌ์ด๋จผ ๋ผ์ŠคํŠธ๋Š” ๋…ธ์…˜ ๋‚ด๋ถ€์˜ ๊ฐœ๋ฐœ ๋ฌธํ™”๊ฐ€ ์ง€๋‚œ 2~3๋…„ ์‚ฌ์ด ์™„์ „ํžˆ ๋ฐ”๋€Œ์—ˆ๋‹ค๊ณ  ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค. ๊ฐ€์žฅ ๋†€๋ผ์šด ๋ณ€ํ™”๋Š” โ€˜AI ํ•˜๋‹ˆ์Šค(AI Harness, AI ๋ชจ๋ธ์„ ์ œํ’ˆ์— ์—ฐ๊ฒฐํ•˜๋Š” ์‹œ์Šคํ…œ ์ธํ”„๋ผ)โ€˜๋ฅผ 6๊ฐœ์›”๋งˆ๋‹ค ํ†ต์งธ๋กœ ์ƒˆ๋กœ ์“ด๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค.

โ€œ๊ธฐ์ˆ  ๋ฐœ์ „ ์†๋„๊ฐ€ ๋„ˆ๋ฌด ๋นจ๋ผ์„œ ์˜ˆ์ „ ๋ฐฉ์‹์„ ๊ณ ์ˆ˜ํ•˜๋ฉด ๋’ค์ฒ˜์ง‘๋‹ˆ๋‹ค. ๋งค๋ฒˆ ์ƒˆ๋กœ ์‹œ์ž‘ํ•˜๋Š” ๊ฒŒ ์˜คํžˆ๋ ค ๋” ํšจ์œจ์ ์ด์ฃ .โ€

์ด๋Ÿฌํ•œ ๊ณผ๊ฐํ•œ ์‹œ๋„๊ฐ€ ๊ฐ€๋Šฅํ•œ ์ด์œ ๋Š” ๋…ธ์…˜ ์—”์ง€๋‹ˆ์–ด๋“ค์ด โ€˜์ฝ”๋”ฉ ์—์ด์ „ํŠธโ€™๋ฅผ ์ ๊ทน์ ์œผ๋กœ ํ™œ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์‚ฌ์ด๋จผ ๋ณธ์ธ๋„ ์ž‘๋…„ ์—ฌ๋ฆ„ ์ดํ›„๋กœ ์ง์ ‘ ์ฝ”๋“œ๋ฅผ ํƒ€์ดํ•‘ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹  ์—์ด์ „ํŠธ์—๊ฒŒ ์ „์ฒด์ ์ธ ์„ค๊ณ„๋ฅผ ๋งก๊ธฐ๊ณ , ๊ฒฐ๊ณผ๋ฌผ์„ ๊ฒ€์ฆ(Verify)ํ•˜๊ณ  ๋ชจ๋‹ˆํ„ฐ๋งํ•˜๋Š” โ€˜๊ด€๋ฆฌ์žโ€™ ์—ญํ• ์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋Š” ์ด ์ง€์ ์—์„œ **โ€˜1,000๋ฐฐ ์—”์ง€๋‹ˆ์–ดโ€™**์˜ ๊ฐœ๋…์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.

  • ์ค‘๊ฐ„ ์ˆ˜์ค€์˜ ์—”์ง€๋‹ˆ์–ด: ๋„๊ตฌ ํ™œ์šฉ ์—ญ๋Ÿ‰์— ๋”ฐ๋ผ ์„ฑ๊ณผ๊ฐ€ ๊ฒฐ์ •๋จ.
  • ์ƒ์œ„ ์—”์ง€๋‹ˆ์–ด: ์—์ด์ „ํŠธ๋ฅผ ์กฐ์œจํ•˜์—ฌ ํ˜ผ์ž์„œ ์ˆ˜์‹ญ ๋ช…๋ถ„์˜ ๊ฒฐ๊ณผ๋ฌผ์„ ๋งŒ๋“ค์–ด๋ƒ„.
  • ๊ฒฐ๊ณผ: ํ•˜ํ•œ์„ (Minimum bar)์€ ๊ทธ๋Œ€๋กœ์ง€๋งŒ, ์ƒํ•œ์„ (Maximum bar)์ด ๋ฌดํ•œ๋Œ€๋กœ ํ™•์žฅ๋จ.

4. ์ž์œจ ์ฃผํ–‰ํ•˜๋Š” ์›Œํฌ์ŠคํŽ˜์ด์Šค: โ€˜์ปค์Šคํ…€ ์—์ด์ „ํŠธโ€™์˜ ๋“ฑ์žฅ

์ตœ๊ทผ ๋…ธ์…˜์€ โ€˜์ปค์Šคํ…€ ์—์ด์ „ํŠธ(Custom Agents)โ€™ ๊ธฐ๋Šฅ์„ ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์‚ฌ์šฉ์ž๊ฐ€ ํŠน์ • ๋ชฉ์ ์„ ์œ„ํ•ด ๋งŒ๋“  AI ๋น„์„œ๋กœ, ๋ฐฐ๊ฒฝ์—์„œ ์ž์œจ์ ์œผ๋กœ ์ž‘๋™ํ•ฉ๋‹ˆ๋‹ค. ์‚ฌ์ด๋จผ์€ ์ž์‹ ์˜ ์‹ค์ œ ํ™œ์šฉ ์‚ฌ๋ก€๋ฅผ ์†Œ๊ฐœํ–ˆ์Šต๋‹ˆ๋‹ค.

  1. ์ด๋ฉ”์ผ ์ •๋ฆฌ ์—์ด์ „ํŠธ: ๋ชจ๋“  ์—…๋ฌด ๋ฐ ๊ฐœ์ธ ์ด๋ฉ”์ผ์„ ์ฝ๊ณ , ์ค‘์š”ํ•˜์ง€ ์•Š์€ ๋ฉ”์ผ์€ ์ž๋™์œผ๋กœ ์•„์นด์ด๋ธŒํ•ฉ๋‹ˆ๋‹ค. ์ดˆ๊ธฐ์—๋Š” ์‚ฌ์ด๋จผ์˜ ํ”ผ๋“œ๋ฐฑ์„ ๋ฐ›์œผ๋ฉฐ ํ•™์Šต(Memory)์„ ์Œ“์•˜๊ณ , ์ด์ œ๋Š” ์™„๋ฒฝํ•˜๊ฒŒ ๊ทธ์˜ ์ทจํ–ฅ์„ ํŒŒ์•…ํ•ด ์ด๋ฉ”์ผ ์ŠคํŠธ๋ ˆ์Šค๋ฅผ ์—†์• ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค.
  2. ๋ฒ„๊ทธ ๋ผ์šฐํŒ… ์—์ด์ „ํŠธ: ์Šฌ๋ž™ ์ฑ„๋„์— ์˜ฌ๋ผ์˜ค๋Š” ์ˆ˜๋งŽ์€ ๊ณ ๊ฐ ํ”ผ๋“œ๋ฐฑ๊ณผ ๋ฒ„๊ทธ ์‹ ๊ณ ๋ฅผ ๋ถ„์„ํ•ด, ๋‹ด๋‹น ํŒ€์˜ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์— ์ž๋™์œผ๋กœ ํ• ๋‹นํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋Š” โ€œ์—์ด์ „ํŠธ๊ฐ€ ์Šค์Šค๋กœ์˜ ๋Šฅ๋ ฅ์„ ํ™•์žฅํ•˜๋Š” โ€˜๋ถ€ํŠธ์ŠคํŠธ๋ž˜ํ•‘(Bootstrapping)โ€˜์ด ๋ฏธ๋ž˜์˜ ํ•ต์‹ฌโ€์ด๋ผ๊ณ  ๋งํ•ฉ๋‹ˆ๋‹ค. ์—์ด์ „ํŠธ๊ฐ€ ํ•„์š”ํ•œ ๋„๊ตฌ๊ฐ€ ์—†๋‹ค๋ฉด ์Šค์Šค๋กœ ์ฝ”๋“œ๋ฅผ ์งœ์„œ ํ†ตํ•ฉ ๊ธฐ๋Šฅ์„ ๋งŒ๋“ค๊ณ  ์‹คํ–‰ํ•˜๋Š” ๋‹จ๊ณ„๊นŒ์ง€ ๋‚˜์•„๊ฐ€๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

5. ๊ฒฐ๋ก : ๋„๊ตฌ์—์„œ ํŒŒํŠธ๋„ˆ๋กœ, ๋…ธ์…˜์˜ ์ƒˆ๋กœ์šด ์ฒ ํ•™

๋…ธ์…˜์˜ ๊ทผ๊ฐ„์€ ์ด๋ฅธ๋ฐ” โ€˜์ƒ๊ฐ์„ ์œ„ํ•œ ๋„๊ตฌ(Tools for Thought)โ€™ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ์žˆ์Šต๋‹ˆ๋‹ค. AI ์‹œ๋Œ€ ์ด์ „์˜ ๋…ธ์…˜์ด โ€˜์ธ๊ฐ„์ด ์ง์ ‘ ์ผ์„ ํ•˜๊ธฐ ์œ„ํ•œ ์ตœ๊ณ ์˜ ๋„๊ตฌโ€™์˜€๋‹ค๋ฉด, ์ด์ œ ๋…ธ์…˜์˜ ๋ชฉํ‘œ๋Š” **โ€˜์ธ๊ฐ„์ด ์—์ด์ „ํŠธ๋ฅผ ๊ด€๋ฆฌํ•˜๋ฉฐ ํ•จ๊ป˜ ์ผํ•˜๋Š” ์ตœ๊ณ ์˜ ํ”Œ๋žซํผโ€™**์ด ๋˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์‚ฌ์ด๋จผ ๋ผ์ŠคํŠธ๋Š” ๋งํ•ฉ๋‹ˆ๋‹ค. โ€œ์šฐ๋ฆฌ๋Š” ์—ฌ์ „ํžˆ ๋ฌธ์„œ(Document)์™€ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค(Database)๋ผ๋Š” ๊ธฐ๋ณธ ๋‹จ์œ„๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ๋‹ค๋งŒ ์ด์ œ ๊ทธ๊ฒƒ๋“ค์€ ์ธ๊ฐ„๋ฟ๋งŒ ์•„๋‹ˆ๋ผ AI ์—์ด์ „ํŠธ๊ฐ€ ์ฝ๊ณ  ์“ฐ๊ธฐ์—๋„ ๊ฐ€์žฅ ํŽธํ•œ ํ˜•ํƒœ์—ฌ์•ผ ํ•˜์ฃ .โ€

๋…ธ์…˜์€ ํ˜„์žฌ ์—์ด์ „ํŠธ๊ฐ€ ๋” ์ž˜ ์ดํ•ดํ•  ์ˆ˜ ์žˆ๋„๋ก โ€˜๋งˆํฌ๋‹ค์šด(Markdown) ๋ณ€ํ˜• ์–ธ์–ดโ€™๋ฅผ ์„ค๊ณ„ํ•˜๊ณ , ์—์ด์ „ํŠธ ์ „์šฉ API๋ฅผ ์ตœ์ ํ™”ํ•˜๋Š” ๋ฐ ์ง‘์ค‘ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ธ๊ฐ„๊ณผ AI๊ฐ€ ํ•œ ๊ณต๊ฐ„์—์„œ ์บ”๋ฒ„์Šค(Canvas)๋ฅผ ๊ณต์œ ํ•˜๋ฉฐ ํ˜‘์—…ํ•˜๋Š” ์‹œ๋Œ€, ๋…ธ์…˜์€ ๊ทธ ๊ฑฐ๋Œ€ํ•œ ๋ณ€ํ™”์˜ ์ค‘์‹ฌ์—์„œ โ€˜์—์ด์ „ํŠธ ๋งค๋‹ˆ์ €โ€™๊ฐ€ ๋œ ์šฐ๋ฆฌ๋ฅผ ๊ธฐ๋‹ค๋ฆฌ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.


์ด ๊ธฐ์‚ฌ๋Š” ํŒŸ์บ์ŠคํŠธ โ€˜No Priorsโ€™์˜ ์‚ฌ์ด๋จผ ๋ผ์ŠคํŠธ ์ธํ„ฐ๋ทฐ ๋‚ด์šฉ์„ ๋ฐ”ํƒ•์œผ๋กœ ์ž‘์„ฑ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.


โ€œPalantir CEO on Iran, AI Weapons and Americaโ€™s Advantage | a16z American Dynamism Summitโ€ โ€” a16z ๊ธฐ๋ฐ˜ ๊ธฐ์‚ฌ ์›๋ณธ ์˜์ƒ ๋ณด๊ธฐ

โ€์ „์Ÿ์˜ ์ŠนํŒจ๋Š” AI๊ฐ€ ๊ฒฐ์ •ํ•œ๋‹คโ€ ํŒ”๋ž€ํ‹ฐ์–ด CEO ์•Œ๋ ‰์Šค ์นดํ”„๊ฐ€ ๋˜์ง€๋Š” ์„œ๊ตฌ ๋ฌธ๋ช…์˜ ๊ฒฝ๊ณ ์™€ ํฌ๋ง

์ตœ๊ทผ ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ์—์„œ ๊ฐ€์žฅ ๋œจ๊ฑฐ์šด ํ™”๋‘๋Š” ๋‹จ์—ฐ โ€˜์•„๋ฉ”๋ฆฌ์นธ ๋‹ค์ด๋‚ด๋ฏธ์ฆ˜(American Dynamism)โ€˜์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๊ธฐ์ˆ  ํ˜์‹ ์„ ํ†ตํ•ด ๋ฏธ๊ตญ์˜ ๊ตญ๊ฐ€ ๊ฒฝ์Ÿ๋ ฅ๊ณผ ์•ˆ๋ณด๋ฅผ ๊ฐ•ํ™”ํ•˜๋ ค๋Š” ์›€์ง์ž„์„ ๋œปํ•ฉ๋‹ˆ๋‹ค. ์ง€๋‚œ 20๋…„๊ฐ„ ์ด ๋ถ„์•ผ์˜ โ€˜์ด๋‹จ์•„โ€™๋กœ ๋ถˆ๋ฆฌ๋ฉฐ ๋ฌต๋ฌตํžˆ ๊ธธ์„ ๊ฑธ์–ด์˜จ ๊ธฐ์—…์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฐ”๋กœ ํŒ”๋ž€ํ‹ฐ์–ด(Palantir Technologies)์ž…๋‹ˆ๋‹ค.

a16z๊ฐ€ ์ฃผ์ตœํ•œ โ€˜์•„๋ฉ”๋ฆฌ์นธ ๋‹ค์ด๋‚ด๋ฏธ์ฆ˜ ์„œ๋ฐ‹โ€™์— ์ฐธ์„ํ•œ ํŒ”๋ž€ํ‹ฐ์–ด์˜ CEO ์•Œ๋ ‰์Šค ์นดํ”„(Alex Carp)๋Š” ํŠน์œ ์˜ ๊ฑฐ์นจ์—†๋Š” ํ™”๋ฒ•์œผ๋กœ ์ค‘๋™ ์ •์„ธ, AI ๋ฌด๊ธฐ์ฒด๊ณ„, ๊ทธ๋ฆฌ๊ณ  ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ๊ฐ€ ์ง๋ฉดํ•œ ์œ„๊ธฐ์— ๋Œ€ํ•ด ์‹ฌ๋„ ์žˆ๋Š” ํ†ต์ฐฐ์„ ๊ณต์œ ํ–ˆ์Šต๋‹ˆ๋‹ค. 20๋…„ ๋™์•ˆ โ€˜๊ดด์งœ๋“ค์˜ ์ง‘ํ•ฉ์†Œโ€™ ์ทจ๊ธ‰์„ ๋ฐ›์œผ๋ฉด์„œ๋„ ๋๋‚ด ๋ฏธ๊ตญ์˜ ์ „๋žต์  ์šฐ์œ„๋ฅผ ์ž…์ฆํ•ด๋‚ธ ๊ทธ์˜ ๋ฉ”์‹œ์ง€๋ฅผ ์ •๋ฆฌํ–ˆ์Šต๋‹ˆ๋‹ค.


1. ์žฌ๊ฑด๋œ ๋ฏธ๊ตญ์˜ ์–ต์ œ๋ ฅ: ๊ธฐ์ˆ ์ด ์ „์žฅ์„ ์ง€๋ฐฐํ•˜๋‹ค

์•Œ๋ ‰์Šค ์นดํ”„๋Š” ์ตœ๊ทผ ์ค‘๋™์—์„œ ๋ฒŒ์–ด์ง„ ์ผ๋ จ์˜ ๊ตฐ์‚ฌ ์ž‘์ „์„ ์–ธ๊ธ‰ํ•˜๋ฉฐ ๋Œ€ํ™”๋ฅผ ์‹œ์ž‘ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ํ•œ๋•Œ ์•ฝํ™”๋˜์—ˆ๋˜ ๋ฏธ๊ตญ์˜ **์–ต์ œ๋ ฅ(Deterrence)**์ด ๊ธฐ์ˆ ์˜ ํž˜์œผ๋กœ ๋‹ค์‹œ ์‚ด์•„๋‚˜๊ณ  ์žˆ๋‹ค๊ณ  ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค.

โ€œ์šฐ๋ฆฌ๋Š” ์ง€๊ธˆ ์–ด๋А ๊ตญ๊ฐ€๋„ ๋ณด์œ ํ•˜์ง€ ๋ชปํ•œ ๊ฐ•๋ ฅํ•œ ์–ต์ œ๋ ฅ์„ ๋ฐœํœ˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๋‹จ์ˆœํžˆ ๊ตฐ๋Œ€์˜ ๊ทœ๋ชจ ๋ฌธ์ œ๊ฐ€ ์•„๋‹™๋‹ˆ๋‹ค. ์ „์Ÿ์˜ ๋ณธ์งˆ์ด ๊ธฐ์ˆ ๋กœ ๋ณ€ํ™”ํ–ˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.โ€

๊ทธ๋Š” ์ œ2์ฐจ ์„ธ๊ณ„๋Œ€์ „ ๋‹น์‹œ ๋ฏธ๊ตญ์ด ์Šน๋ฆฌํ•  ์ˆ˜ ์žˆ์—ˆ๋˜ ์›๋™๋ ฅ์ด ๊ธฐ์ˆ ์  ์šฐ์œ„์˜€์Œ์„ ์ƒ๊ธฐ์‹œ์ผฐ์Šต๋‹ˆ๋‹ค. ์˜ค๋Š˜๋‚  โ€˜์˜คํผ๋ ˆ์ด์…˜ ๋ฏธ๋“œ๋‚˜์ž‡ ํ•ด๋จธ(Operation Midnight Hammer)โ€˜์™€ ๊ฐ™์€ ์ž‘์ „์—์„œ ๋ณด์—ฌ์ค€ ์••๋„์ ์ธ ์„ฑ๊ณผ๋Š” ์†Œํ”„ํŠธ์›จ์–ด, ํ•˜๋“œ์›จ์–ด, ๊ทธ๋ฆฌ๊ณ  AI๊ฐ€ ๊ฒฐํ•ฉ๋œ ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ์ฒด๊ณ„์˜ ์Šน๋ฆฌ๋ผ๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์นดํ”„๋Š” ์„œ๊ตฌ ์‚ฌํšŒ๊ฐ€ ๋ฒ•์น˜์ฃผ์˜์™€ ํ‰๋“ฑ์„ ์ง€ํ–ฅํ•ด์•ผ ํ•˜์ง€๋งŒ, ํ˜„์‹ค ์„ธ๊ณ„์—์„œ๋Š” ๊ฒฐ๊ตญ **๊ตฐ์‚ฌ์  ์šฐ์œ„(Military Superiority)**๋งŒ์ด ๊ฒฐ์ •์ ์ธ ํˆฌํ‘œ๊ถŒ์„ ํ–‰์‚ฌํ•œ๋‹ค๊ณ  ๋‹จ์–ธํ–ˆ์Šต๋‹ˆ๋‹ค.

2. ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ๋ฅผ ํ–ฅํ•œ ๊ฒฝ๊ณ : โ€œ๊ตญ์œ ํ™”์˜ ๋Š‘๋Œ€๊ฐ€ ๋ฌธ์•ž์— ์™€ ์žˆ๋‹คโ€

์นดํ”„๋Š” ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ์˜ ์—˜๋ฆฌํŠธ์ฃผ์˜์™€ ์ •์น˜์  ํŽธํ–ฅ์„ฑ์— ๋Œ€ํ•ด ๋‚  ์„  ๋น„ํŒ์„ ์„œ์Šด์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ๊ธฐ์ˆ  ๊ธฐ์—…๋“ค์ด ๊ตญ๊ฐ€ ์•ˆ๋ณด์™€ ๋…ธ๋™์ž์˜ ์‚ถ์„ ์™ธ๋ฉดํ•  ๊ฒฝ์šฐ, ๊ธฐ์ˆ  ์ž์ฒด๊ฐ€ **๊ตญ์œ ํ™”(Nationalization)**๋˜๋Š” ๋น„๊ทน์„ ๋งž์ดํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๊ฒฝ๊ณ ํ–ˆ์Šต๋‹ˆ๋‹ค.

  • ํ™”์ดํŠธ์นผ๋ผ ์‹ค์—… ๋ฌธ์ œ: AI๊ฐ€ ๊ณ ํ•™๋ ฅ ํ™”์ดํŠธ์นผ๋ผ์˜ ์ผ์ž๋ฆฌ๋ฅผ ๋บ๋Š” ๋ฐ๋งŒ ์ง‘์ค‘ํ•˜๊ณ , ์ •์ž‘ ๊ตญ๊ฐ€๋ฅผ ์ง€ํ‚ค๋Š” ๊ตฐ์ธ๋“ค์„ ์œ„ํ•œ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ์„ ์™ธ๋ฉดํ•œ๋‹ค๋ฉด ๋Œ€์ค‘์˜ ๋ถ„๋…ธ๋ฅผ ํ”ผํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค.
  • ์ •์น˜์  ๊ณ ๋ฆฝ: ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ๊ฐ€ ํŒฐ๋กœ์•จํ† (Palo Alto) ์•ˆ์—์„œ๋งŒ ์ธ๊ธฐ ์žˆ๋Š” ๊ธฐ์ˆ ์— ๋งค๋ชฐ๋œ๋‹ค๋ฉด, ์›Œ์‹ฑํ„ด D.C.์™€ ์ผ๋ฐ˜ ๋ฏธ๊ตญ ์‹œ๋ฏผ๋“ค๋กœ๋ถ€ํ„ฐ ์™ธ๋ฉด๋ฐ›๊ฒŒ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

๊ทธ๋Š” โ€œ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ๊ฐ€ 160์˜ ์ง€๋Šฅ์ง€์ˆ˜(IQ)๋ฅผ ๊ฐ€์กŒ์„์ง€๋Š” ๋ชฐ๋ผ๋„, ๋Œ€์ค‘์˜ ์ •์„œ๋ฅผ ์ฝ์ง€ ๋ชปํ•œ๋‹ค๋ฉด ๊ฒฐ๊ตญ ๊ธฐ์ˆ  ํ†ต์ œ๊ถŒ์„ ๊ตญ๊ฐ€์— ๋นผ์•—๊ธฐ๊ฒŒ ๋  ๊ฒƒโ€์ด๋ผ๋ฉฐ, ๊ธฐ์ˆ ๊ณผ ๊ตญ๊ฐ€ ์ด์ต์˜ ์ •๋ ฌ(Alignment)์„ ๊ฐ•๋ ฅํžˆ ์ด‰๊ตฌํ–ˆ์Šต๋‹ˆ๋‹ค.

3. AI ๊ฒฝ์Ÿ์€ โ€˜์ œ๋กœ์„ฌ ๊ฒŒ์ž„โ€™์ด๋‹ค

๋งŽ์€ ์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ ๊ธฐ์—…๋“ค์ด AI ๋ฐœ์ „์„ โ€˜๋ชจ๋‘์—๊ฒŒ ์ด๋กœ์šด ๊ฒŒ์ž„โ€™์œผ๋กœ ํฌ์žฅํ•˜์ง€๋งŒ, ์นดํ”„์˜ ์‹œ๊ฐ์€ ๋ƒ‰ํ˜นํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Š” ๊ตญ๊ฐ€ ๊ฐ„์˜ AI ๊ฒฝ์Ÿ์„ ์ฒ ์ €ํ•œ ์ œ๋กœ์„ฌ(Zero-sum) ๊ฒŒ์ž„์œผ๋กœ ๊ทœ์ •ํ•ฉ๋‹ˆ๋‹ค.

โ€œ์ด๋ก ์ ์œผ๋กœ๋Š” ๋ชจ๋‘๊ฐ€ ํ‰๋“ฑํ•œ ์‹œ์Šคํ…œ์ด ์ข‹๊ฒ ์ง€๋งŒ, ํ˜„์‹ค์€ ์šฐ๋ฆฌ ์•„๋‹ˆ๋ฉด ์ค‘๊ตญ, ํ˜น์€ ๋Ÿฌ์‹œ์•„์ž…๋‹ˆ๋‹ค. ์„œ๊ตฌ ๋ฌธ๋ช…์ด ์šฐ์œ„๋ฅผ ์ ํ•˜์ง€ ๋ชปํ•˜๋ฉด ์šฐ๋ฆฌ๊ฐ€ ๋ˆ„๋ฆฌ๋Š” ์ž์œ ์™€ ๊ถŒ๋ฆฌ๋„ ์‚ฌ๋ผ์ง‘๋‹ˆ๋‹ค.โ€

๊ทธ๋Š” ๊ฑฐ๋Œ€์–ธ์–ด๋ชจ๋ธ(LLM)์„ ํฌํ•จํ•œ ์ฒจ๋‹จ ๊ธฐ์ˆ ์ด ๋‹จ์ˆœํžˆ ๊ฒฝ์ œ์  ๋„๊ตฌ๋ฅผ ๋„˜์–ด, ์„œ๊ตฌ์˜ ๊ฐ€์น˜๊ด€์„ ์ˆ˜ํ˜ธํ•˜๋Š” ๋ฐฉํŒจ๊ฐ€ ๋˜์–ด์•ผ ํ•œ๋‹ค๊ณ  ์ฃผ์žฅํ•ฉ๋‹ˆ๋‹ค. ๋ฏธ๊ตญ์ด ๊ธฐ์ˆ ์  ๊ฒฉ์ฐจ๋ฅผ ๋ฒŒ๋ฆฌ์ง€ ๋ชปํ•  ๋•Œ ๋‹ฅ์ณ์˜ฌ ๋ฌธํ™”์ , ๊ตฐ์‚ฌ์  ์••๋ฐ•์€ ์ƒ์ƒ ์ด์ƒ์ผ ๊ฒƒ์ด๋ผ๋Š” ๋ถ„์„์ž…๋‹ˆ๋‹ค.

4. ํŒ”๋ž€ํ‹ฐ์–ด์˜ ๋น„๋ฐ€ ๋ณ‘๊ธฐ: โ€˜์‹ ๊ฒฝ๋‹ค์–‘์„ฑโ€™๊ณผ ๊ฐœ์„ฑ ์žˆ๋Š” ์ธ์žฌ

ํŒ”๋ž€ํ‹ฐ์–ด๊ฐ€ ์–ด๋–ป๊ฒŒ ๊ทธํ† ๋ก ๋…๋ณด์ ์ธ ์ œํ’ˆ๋“ค์„ ๋งŒ๋“ค์–ด๋‚ผ ์ˆ˜ ์žˆ์—ˆ์„๊นŒ์š”? ์นดํ”„๋Š” ๊ทธ ๋น„๊ฒฐ๋กœ **์‹ ๊ฒฝ๋‹ค์–‘์„ฑ(Neurodivergent)**์„ ๊ผฝ์•˜์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ๋‚œ๋…์ฆ(Dyslexia)์„ ์•“๊ณ  ์žˆ๋Š” ์ž์‹ ์˜ ๊ฒฝํ—˜์„ ๊ณต์œ ํ•˜๋ฉฐ, ๋‚จ๋“ค๊ณผ ๋‹ค๋ฅด๊ฒŒ ์ƒ๊ฐํ•˜๋Š” ์ธ์žฌ๋“ค์˜ ์ค‘์š”์„ฑ์„ ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค.

  • ๊ฐœ๋ณ„์ ์ธ ํ”Œ๋ ˆ์ด๋ถ: ํŒ”๋ž€ํ‹ฐ์–ด๋Š” ์ง์›๋“ค์—๊ฒŒ ํšŒ์‚ฌ์˜ ๋งค๋‰ด์–ผ์„ ๊ฐ•์š”ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹  ๊ฐ์ž๊ฐ€ ๊ฐ€์ง„ ๊ณ ์œ ํ•œ ์žฌ๋Šฅ์„ ๊ทน๋Œ€ํ™”ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋•์Šต๋‹ˆ๋‹ค.
  • ์ด์งˆ์ ์ธ ์ง‘๋‹จ: ๋ฏผ์ฃผ๋‹น์›, ๊ณตํ™”๋‹น์›, ์˜ˆ์ˆ ๊ฐ€, ๊ธฐ์ˆ ์ž๊ฐ€ ํ•œ๋ฐ ์„ž์—ฌ ํ† ๋ก ํ•˜๋Š” ๋ฌธํ™”๊ฐ€ ํŒ”๋ž€ํ‹ฐ์–ด์˜ ํž˜์ž…๋‹ˆ๋‹ค.

์นดํ”„๋Š” โ€œ๋ฏธ๊ตญ์˜ ์ง„์ •ํ•œ ๊ฐ•์ ์€ ์‚ฌ์ƒ, ์ข…๊ต, ๊ทธ๋ฆฌ๊ณ  ์‹ ๊ฒฝํ•™์ ์œผ๋กœ ๋‹ค๋ฅธ ๋ชจ๋“  ์ด๋“ค์ด ๋ชจ์—ฌ ์ž์œ ๋กญ๊ฒŒ ์ž์‹ ์„ ํ‘œํ˜„ํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ โ€์ด๋ผ๋ฉฐ, ์ด๋Ÿฌํ•œ โ€˜๋‹ค๋ฆ„โ€™์„ ๊ธฐ์ˆ ๋กœ ์ฆํญ์‹œํ‚ค๋Š” ๊ฒƒ์ด ํŒ”๋ž€ํ‹ฐ์–ด์˜ ์‚ฌ๋ช…์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.

5. ๊ฒฐ๋ก : โ€œ์šฐ๋ฆฌ ๊ตฐ์ธ์ด ๋ฌด์‚ฌํžˆ ๋Œ์•„์˜ค๊ฒŒ ํ•˜๋Š” ๊ฒƒโ€

๊ฐ•์—ฐ์˜ ๋งˆ์ง€๋ง‰์—์„œ ์นดํ”„๋Š” ํŒ”๋ž€ํ‹ฐ์–ด๊ฐ€ ํ•˜๋Š” ๊ฐ€์žฅ ์ค‘์š”ํ•œ ์ผ์€ ๊ฒฐ๊ตญ **โ€œ๋ฏธ๊ตญ ์ „์‚ฌ(War fighters)๋“ค์ด ์ง‘์œผ๋กœ ๋ฌด์‚ฌํžˆ ๋Œ์•„์˜ฌ ํ™•๋ฅ ์„ ๋†’์ด๋Š” ๊ฒƒโ€**์ด๋ผ๊ณ  ๋งํ–ˆ์Šต๋‹ˆ๋‹ค. ๋™์‹œ์— ์ ๋“ค์—๊ฒŒ๋Š” โ€œ๋ฏธ๊ตญ์„ ๊ณต๊ฒฉํ•˜๋ฉด ๊ฒฐ์ฝ” ๋ฌด์‚ฌํžˆ ๋Œ์•„๊ฐˆ ์ˆ˜ ์—†๋‹คโ€๋Š” ๋ช…ํ™•ํ•œ ๋ฉ”์‹œ์ง€๋ฅผ ์ฃผ๋Š” ๊ฒƒ์ด ์ง„์ •ํ•œ ํ‰ํ™”์˜ ๊ธธ์ด๋ผ๊ณ  ์—ญ์„คํ–ˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๋Š” ๊ธฐ์ˆ  ์ฐฝ์—…์ž๋“ค์—๊ฒŒ ์กฐ์–ธํ–ˆ์Šต๋‹ˆ๋‹ค. โ€œ๋˜‘๋˜‘ํ•˜๋‹ค๊ณ  ๋ชจ๋“  ๋ถ„์•ผ์—์„œ ์œ ๋Šฅํ•œ ๊ฒƒ์€ ์•„๋‹ˆ๋‹ค. ํ˜„์žฅ์˜ ๊ตฐ์ธ๋“ค์„ ๋งŒ๋‚˜ ๊ณต๊ฐํ•˜๊ณ , ๊ทธ๋“ค์ด ์ฒ˜ํ•œ ํ˜„์‹ค์„ ์ดํ•ดํ•˜๋ผ.โ€

์•Œ๋ ‰์Šค ์นดํ”„์˜ ๋ฉ”์‹œ์ง€๋Š” ๋ช…ํ™•ํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ์ˆ ์€ ์ง„๊ณต ์ƒํƒœ์—์„œ ์กด์žฌํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๊ธฐ์ˆ ์ด ์ž์œ  ๋ฏผ์ฃผ์ฃผ์˜๋ผ๋Š” ๊ฐ€์น˜์™€ ๊ฒฐํ•ฉํ•  ๋•Œ, ๋น„๋กœ์†Œ ์„ธ์ƒ์„ ์ง€ํ‚ค๋Š” ๊ฐ•๋ ฅํ•œ ํž˜์ด ๋œ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ํŒ”๋ž€ํ‹ฐ์–ด๋Š” ์ง€๋‚œ 20๋…„๊ฐ„ ์ด ์‹ ๋…์„ ์ฆ๋ช…ํ•ด ์™”์œผ๋ฉฐ, ์ด์ œ ๊ทธ โ€˜์ด๋‹จ์•„์˜ ์ฒ ํ•™โ€™์€ ๋ฏธ๊ตญ์˜ ์ƒˆ๋กœ์šด ํ‘œ์ค€์ด ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.


๐Ÿ’ก ํ•ต์‹ฌ ์šฉ์–ด ์ •๋ฆฌ

  • ์•„๋ฉ”๋ฆฌ์นธ ๋‹ค์ด๋‚ด๋ฏธ์ฆ˜(American Dynamism): ๊ตญ๊ฐ€ ์•ˆ๋ณด, ์—๋„ˆ์ง€, ์ œ์กฐ ๋“ฑ ๊ตญ๊ฐ€์˜ ๊ทผ๊ฐ„์ด ๋˜๋Š” ์‚ฐ์—…์— ๊ธฐ์ˆ  ํ˜์‹ ์„ ์ ‘๋ชฉํ•ด ๋ฏธ๊ตญ์˜ ์—ญ๋™์„ฑ์„ ๋˜์‚ด๋ฆฌ๋ ค๋Š” ์›€์ง์ž„.
  • ์–ต์ œ๋ ฅ(Deterrence): ์ƒ๋Œ€๋ฐฉ์ด ๊ณต๊ฒฉํ–ˆ์„ ๋•Œ ์–ป๋Š” ์ด์ต๋ณด๋‹ค ํ”ผํ•ด๊ฐ€ ํ›จ์”ฌ ํฌ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์คŒ์œผ๋กœ์จ ๊ณต๊ฒฉ ์˜์‚ฌ๋ฅผ ๊บพ๋Š” ํž˜.
  • ์‹ ๊ฒฝ๋‹ค์–‘์„ฑ(Neurodivergent): ์žํ, ๋‚œ๋…์ฆ, ADHD ๋“ฑ ๋‡Œ์˜ ๊ธฐ๋Šฅ์  ์ฐจ์ด๋ฅผ โ€˜์žฅ์• โ€™๊ฐ€ ์•„๋‹Œ ์ธ๊ฐ„์˜ โ€˜๋‹ค์–‘์„ฑโ€™์œผ๋กœ ๋ณด๋Š” ๊ด€์ .
  • ์ œ๋กœ์„ฌ(Zero-sum): ํ•œ์ชฝ์˜ ์ด๋“์ด ๋‹ค๋ฅธ ์ชฝ์˜ ์†์‹ค๋กœ ์ด์–ด์ ธ ์ „์ฒด ํ•ฉ์ด 0์ด ๋˜๋Š” ์ƒํƒœ. ์—ฌ๊ธฐ์„œ๋Š” ๊ธ€๋กœ๋ฒŒ ๊ธฐ์ˆ  ํŒจ๊ถŒ ๊ฒฝ์Ÿ์˜ ์น˜์—ดํ•จ์„ ์˜๋ฏธํ•จ.