YouTube Digest

May 21, 2026

English

Based on โ€œAnthropic Just Bought a Dev Tools Startup for $300M. Hereโ€™s What Its Founder Told Me.โ€ from Every Watch the original video

The Internetโ€™s Next Evolution: How AI Will Learn to Code Its Own Future

The internet, a marvel of interconnected systems, was fundamentally designed for human interaction. Its architecture, from websites to web services, caters to clicks, scrolls, and visual interfaces. But as artificial intelligence, particularly large language models (LLMs), becomes increasingly sophisticated, a new challenge emerges: how do we empower AI to truly use the internet, not just observe it?

This is the frontier where companies like Stainless, an API and SDK powerhouse behind some of the biggest names in tech, are operating. Founded and led by Alex Ratray, Stainless has long specialized in making computers talk to each other seamlessly. Now, theyโ€™re at the forefront of enabling AI to do the same, grappling with the complexities of whatโ€™s known as Model Context Protocol (MCP).

โ€œAPIs are the dendrites of the internet,โ€ explains Ratray, drawing a vivid analogy to the neural connections in our brains. Just as thoughts donโ€™t happen without neurons communicating, the internet ceases to function without servers and applications exchanging information. For decades, this communication has been between humans and computers via user interfaces (UIs), or between computers via APIs. Now, a โ€œnew computer has entered the chatโ€: LLMs.

The Promise and Pitfalls of Model Context Protocol (MCP)

Model Context Protocol (MCP) represents the internetโ€™s attempt to give LLMs their own native interface. Imagine every website or web service transforming into a set of โ€œtoolsโ€ an AI can use directly to get work done. For instance, a Gmail MCP might offer tools like โ€œsend mailโ€ or โ€œread inbox,โ€ allowing an LLM to compose and manage emails without a human intermediary. The promise is profound: AI agents capable of automating complex, multi-step tasks across various applications.

Ratray illustrates this with a common, yet surprisingly convoluted, business scenario. โ€œLetโ€™s say Dan walks into my store and buys a pair of stripey socks,โ€ he begins. โ€œThe next day, somethingโ€™s wrong, and I tell my team, โ€˜Hey, can we refund Dan for those stripey socks and send him a discount code with a thank you note?โ€™โ€

For a human operator, this seemingly simple task involves a laborious dance across multiple applications: logging into an internal admin, searching a Stripe dashboard for transactions, finding Dan, identifying the specific product, initiating a refund, creating a discount code, and then switching to another SaaS app to send an email. Each step involves multiple clicks, scrolls, and loading screens โ€“ a process that could easily rack up 15 different interactions across five different apps.

The vision of Agentic AI, powered by MCP, is to type that same request into an LLM and have it execute the entire sequence autonomously. โ€œThatโ€™s the big vision of MCP,โ€ Ratray emphasizes. โ€œTo take that prompt and just have the AI go off and do that.โ€

However, the reality of current MCP implementations falls short. Ratray admits, โ€œitโ€™s not working so great.โ€ The core problem lies in the sheer volume of information and the way LLMs process it. To truly empower an AI agent to do anything a human operator could, youโ€™d need to expose every single function of every single API and dashboard โ€“ hundreds of endpoints for a service like Stripe alone, let alone multiple integrated SaaS tools.

This leads to several critical challenges:

  • Context Window Limits: Describing every possible tool, its parameters, and documentation to an LLM quickly consumes its entire โ€œcontext budgetโ€โ€”hundreds of thousands of tokens. This is not only expensive but also confusing for the model.
  • Model Ergonomics: Designing tools that an LLM can understand and use effectively is a โ€œnew research problem.โ€ Humans can learn Python SDKs; LLMs struggle to โ€œthinkโ€ like a developer, often requiring handcrafted, specialized tools.
  • Chained Actions: LLMs have difficulty with sustained, repeated chains of actions, especially when dealing with large datasets (e.g., paginating through thousands of transactions to find a specific one).
  • Security and Permissions: Exposing a vast array of tools to an AI agent raises significant security concerns. โ€œYou donโ€™t want the AI to color outside the lines,โ€ Ratray warns, envisioning scenarios where an AI might accidentally (or maliciously) refund all customers or transfer funds to its own account.

Current solutions often involve drastically limiting the number of tools exposed or using โ€œdynamic mode,โ€ where the LLM has only three tools: list endpoints, get endpoint details, and execute endpoint. While this helps with context, it increases the number of turns the model needs to perform a single action, making it slower and more expensive.

Stainless in Action: AI for Business Insights

Despite these challenges, Ratray and his team at Stainless are actively leveraging MCP for internal operations. โ€œI use MCP to actually, funnily enough, not so much on the coding side, but I use it on the business side,โ€ he shares.

Stainless employs MCP servers for Notion, HubSpot, Gong, and even a read-only copy of their own PostgreSQL database. Ratray can prompt an LLM, โ€œHey, what are the interesting customers that signed up for Stainless last week?โ€ The AI then cross-references data from the database, HubSpot, Notion notes, and even Gong call transcripts to provide a comprehensive answer.

Beyond generating business reports, Stainless is experimenting with AI for knowledge management and even bug fixing. Ratray has a โ€œmaster company Git knowledge repositoryโ€ where he instructs Claude Code to store interesting customer quotes and their citations, SQL queries, and other valuable insights. This allows the AI to โ€œcacheโ€ information, avoiding repetitive searches and quickly retrieving relevant data for future queries or board preparations.

โ€œThatโ€™s [expletive] so cool,โ€ the interviewer exclaims, highlighting the transformative potential of such systems. While still experimental for tasks like bug fixing, the promise of AI-powered efficiency is clear.

The โ€œCyborgโ€ Future: AI That Writes Code

Looking ahead, Ratray has a bold vision for how AI will truly integrate with the internet, a future he provocatively calls โ€œcyborgs.โ€ This doesnโ€™t mean literal human-robot hybrids, but rather AI agents that are โ€œpart GPT neural net LLM, part AI, and part code.โ€ The โ€œcodeโ€ refers to traditional CPU-based software, not just GPU-driven neural networks.

Ratray envisions a future where LLMs donโ€™t just pick from a vast menu of pre-defined tools, but instead write the code themselves. He proposes a model where, instead of a โ€œbajillion tools,โ€ the LLM has just two:

  1. Execute Code: The LLM is given a text box to write TypeScript (or another language) using an APIโ€™s SDK. For example, it could write Stripe.transactions.list() or Stripe.refunds.create().
  2. Search Docs: A tool to query documentation and ask questions about the API.

This approach offers several compelling advantages:

  • Massive Context Reduction: Instead of loading hundreds of thousands of tokens for tool definitions, the initial context window is minimalโ€”perhaps a thousand tokens or less.
  • Efficient Execution: The AI writes the code, which is then executed by the API providerโ€™s server. This runs at CPU speed, eliminating the slow round trips to the LLM for every action. An AI agent could write complex loops to find a specific transaction, only returning a concise summary of its findings.
  • Type Safety and Error Handling: A crucial benefit of code execution is the ability to incorporate type checking. If an LLM hallucinates an API call (e.g., stripe.transactions.list, which doesnโ€™t exist), the code execution tool can immediately identify the error, suggest correct alternatives (like paymentIntents or orders), and provide relevant documentation. This โ€œsub-agentโ€ is well-trained and unburdened by the broader conversationโ€™s context.

โ€œLLMs are really good at writing code,โ€ Ratray asserts. The main hurdle today is their difficulty working with specific library versions and avoiding hallucinations when dealing with raw HTTP APIs without the guardrails of a typed SDK. By having the API provider execute the code, these issues can be managed within a controlled environment.

The Security Imperative: Back to the API Layer

The โ€œcyborgโ€ vision also offers a clearer path for security. Ratray argues that security should ultimately reside at the API layer itself, not by attempting to limit whatโ€™s exposed through MCP. โ€œRight now you see people trying to implement security by sort of limiting whatโ€™s exposed through MCP, and that kind of makes sense, but at the end of the day, you could do anything thatโ€™s in the API under the hood, right?โ€ he points out.

The solution, he believes, lies in robust authentication protocols like OAuth with granular, properly scoped permissions. This ensures that even if an AI agent generates code, its actions are constrained by the permissions granted at the API level, preventing unintended or unauthorized operations.

The journey to a truly intelligent, internet-fluent AI is complex, fraught with technical and security challenges. But pioneers like Alex Ratray and companies like Stainless are charting a course. By moving beyond simplistic tool definitions to an architecture where AI can dynamically write and execute code, we may soon unlock a future where AI agents seamlessly navigate and operate the digital world, transforming productivity and automation in ways we are only just beginning to imagine.


Based on โ€œWas anything good released at Google I/O 2026?โ€ from How I AI Watch the original video

Google I/O 2026: The AI Revolution, One Flash (and a Few Fumbles) at a Time

Google I/O, the tech giantโ€™s annual flagship event, is always a whirlwind of announcements, new features, and ambitious product visions. For Claraveville, a product leader and self-proclaimed AI obsessive behind โ€œHow I AI,โ€ the 2026 iteration was no different. This year, Google went โ€œfull bore into agents,โ€ launching a staggering array of AI-powered tools and models, promising to reshape everything from coding to creative production. Yet, as Claraveville meticulously explored these innovations, a recurring theme emerged: the tantalizing promise of AI often collided with the frustrating reality of limited access, confusing branding, and features that, well, just didnโ€™t quite work โ€“ at least not yet.

The Foundation: Gemini 3.5 Flash โ€” Speed Meets Intelligence

At the core of Googleโ€™s I/O announcements was the unveiling of the Gemini 3.5 family of models, prominently featuring Gemini 3.5 Flash. Claraveville immediately highlighted its significance: โ€œtheir fastest, smartest coding model.โ€ What truly sets Flash apart, according to Googleโ€™s benchmarks, is its dual prowess. It not only โ€œrivals the intelligence of some of our favorite coding modelsโ€ like Opus and GPT-4 but is also โ€œfour times as fast.โ€ This combination promises a โ€œsuper smart modelโ€ฆ at the speed of something much more like their 3.1 flash model.โ€

The emphasis on agentic capabilities was unmistakable. Google is betting big on AI agents, and Gemini 3.5 Flash is designed to be their engine. Claraveville noted that while some of these agentic features might feel like Google playing โ€œcatch-upโ€ to competitors like Anthropic and OpenAI, they come with two distinct advantages: the sheer speed of the Flash models and a pronounced โ€œcreative consumer bent.โ€

Beyond coding, Geminiโ€™s multimodal strength continues to be a standout. Claraveville has consistently championed Gemini for its ability to handle diverse data types: โ€œI have always told people if you are working with files, videos, any sort of transformative work where you have to go from one modalityโ€ฆ Gemini models are really, really good at handling files. I love it for handling videos.โ€ The benchmarks presented at I/O seemed to confirm this, with Gemini 3.5 Flash reportedly exceeding its peers in multimodal and agentic capabilities, all while being incredibly fast.

Developersโ€™ New Toolkit: Anti-gravity and Google AI Studio

For developers, Google introduced updates to its agentic IDE, Anti-gravity, now in version 2.0, and the Anti-gravity CLI. Claraveville observed that many of these features felt familiar, clearly positioning Google in a competitive space with tools like CodeX. Anti-gravity 2.0 brings โ€œprojectsโ€ (folder-constrained workspaces) and โ€œscheduled tasksโ€ โ€“ a UI reminiscent of existing agentic coding apps.

However, Anti-gravity also introduced some intriguing new capabilities. The core agent features now include sub-agents, allowing the main agent to delegate specific tasks, with the browser sub-agent being a particularly popular use case. โ€œHooksโ€ allow developers to tap into various lifecycle events of their agent, offering on-demand customization. Claraveville was particularly keen on the new โ€œslash commands,โ€ especially /goal, which enables an agent to pursue a long-running task until a defined objective is met.

But the real curiosity lay with /grill me. This command promises a more โ€œaggressiveโ€ clarification process, where the agent will โ€œask clarification questions back and really get to the heart of what your requirements are.โ€ Claraveville mused, โ€œNow the question is, is this actually as hardcore as the slash command communicates or is this just a cute way to differentiate against the question and answer tool? Weโ€™ll see.โ€

Putting Anti-gravity to the test, Claraveville attempted to convert a blog generator UI into an API for an AI agent. While the outcome was successful โ€“ generating a programmatic API endpoint with authentication and visual workflow triggers โ€“ the speed didnโ€™t quite match the โ€œFlashโ€ hype in this specific instance. โ€œIโ€™m not noticing it is particularly fasterโ€ฆ it doesnโ€™t feel that much faster to me,โ€ she noted, concluding that while it was โ€œvery nice, fast, kind of what I would expect from a coding agent,โ€ it still felt like โ€œplaying catch-up.โ€

Beyond the IDE, Google AI Studio received updates, leveraging Gemini 3.5 Flash for low-code/no-code app development. The most significant announcement here was the ability to build apps โ€œconnected to your Google Workspace appsโ€ โ€“ reading Sheets, drafting Gmails, organizing Drive, and managing calendars. This is Googleโ€™s strategic move to โ€œown that experience themselves,โ€ carving off internal enterprise and personal assistant use cases. The studio also now supports creating Android apps.

However, Claravevilleโ€™s attempt to build a simple calendar management app for her kidsโ€™ sporting events quickly hit a wall. Despite being a paying customer, the promised Workspace integration was nowhere to be found. โ€œIt didnโ€™t do anything magical,โ€ she reported, highlighting a recurring frustration: โ€œtheyโ€™ve announced a bunch of stuff and then at the very, very bottom theyโ€™re like itโ€™s available to some subset of people or itโ€™s available but later this summer.โ€

The Consumer & Creative Frontier: Gemini AI, Omni, and Flow

Before diving deeper into consumer-facing AI, Claraveville paused to voice a significant critique: Googleโ€™s product naming strategy. โ€œI cannot keep the product and brand name straight,โ€ she lamented, listing the myriad of names already encountered: Gemini 3.5 Flash, Anti-gravity, Google AI Studio, and now โ€œGemini, Google AI, the consumerish AI product.โ€ Her plea to the Google team was clear: โ€œa comprehensive brand analysis and singular brand that I can work with.โ€

The consumer-facing Gemini AI product received a redesign, dubbed โ€œsomething ridiculousโ€ by Claraveville but intended to โ€œmake you feel goodโ€ with its โ€œglowโ€ and inspiring prompts. More substantially, Gemini now integrates Nano Banana, an acclaimed image generation model, with a focus on templates to inspire users. Claraveville put it to the test, asking it to โ€œupscale and beautify the image of this podcast host, change the background to a professional podcasting studio.โ€ The result was a mix of the promising and the peculiar: โ€œThis is not my face. This is horrifying in every way possible.โ€ Yet, she noted improvements in text generation and photorealism (even if not face-realism), all generated โ€œpretty fast.โ€

The real excitement in the creative space revolved around Omni, Googleโ€™s new video generation model, positioned as โ€œNano Banana for video.โ€ Omni promises โ€œlonger, more consistent, more photorealistic videos using AI,โ€ capable of leveraging โ€œreference materials.โ€ Claraveville emphasized Geminiโ€™s strong multimodal capabilities, noting that Omni allows users to โ€œcreate video from anythingโ€ and โ€œground Gemini in real-world knowledge.โ€ Key features include conversational video editing (e.g., changing a structure into bubbles), descriptive video editing, and the ability to refine videos while maintaining character consistency โ€“ a crucial advancement for production-level video.

Claraveville demonstrated Omniโ€™s power by animating a drawing her child made: โ€œAnimate this superhero breaking a kid out of class to go have fun.โ€ The result was a delightful success, a 10-second video that her โ€œkid is really going to like,โ€ significantly longer than previous models like Sora.

For more prescriptive video editing, Google introduced Flow, a tool embedding Omni with a focus on โ€œcinematic quality production.โ€ Flow allows users to blend multimodal references, edit conversationally, and even define characters (e.g., โ€œCaptain Escapo Schoolโ€) for consistent use across videos. Most intriguingly, Flow promised the ability to create an AI avatar of oneself. Claraveville eagerly attempted this, scanning a QR code, reading numbers aloud, and turning her head for the camera: โ€œPlease donโ€™t steal my identity!โ€ she quipped. The outcome? Another I/O disappointment. โ€œNo. No sir, I couldnโ€™tโ€ฆ it didnโ€™t even create the avatar.โ€ This led to a stark conclusion: โ€œThe promise is really good for some of these things, but the reality is if youโ€™re not able to use them or theyโ€™re broken on the dayโ€ฆ people are going to lose patience for some of this.โ€

Design & Marketing Tools: Pamelli and Stitch

The product naming confusion continued with Pamelli, Googleโ€™s brand and marketing content generator, and Stitch, their design tool. Pamelli introduced an agent that can create a โ€œcore brand identity from a website,โ€ including brand books, campaign assets, and photoshoots โ€“ a feature reminiscent of Claude Design. Claraveville highlighted Googleโ€™s underlying design.md standard, which allows product designs to be encoded in markdown for AI agents. Her own website, chatprd.ai, was used to demonstrate Pamelliโ€™s ability to generate a brand book, which later yielded โ€œdefinitely my brand colors. This is my tagline. And now I can createโ€ฆ a new website.โ€ Her final assessment of the generated website: โ€œitโ€™s fine.โ€

Stitch, described as an โ€œin-browser Figma,โ€ showcased features like streaming into a design canvas, starting with existing designs, and inline AI edits. Claraveville noted, โ€œThis is where I think I feel the speed of the Flash models is actually in design and this sort of experience versus the coding experience.โ€ She demonstrated Stitch by asking it to design an app in the style of the Google I/O site, showcasing the streaming design process. Her observation on AI design tools was candid: โ€œEvery AI tool ships their own version of slop. Googleโ€™s slop looks like Google.โ€

The Verdict: A Glimpse of the Future, Still Under Construction

Google I/O 2026 presented a comprehensive vision of an AI-powered future, driven by the speed of Gemini 3.5 Flash and the ambition of agentic AI. From developer tools like Anti-gravity and Google AI Studio to creative powerhouses like Omni and Flow, and specialized design/marketing tools like Pamelli and Stitch, Googleโ€™s portfolio is undeniably expansive.

Claravevilleโ€™s journey through these announcements was a microcosm of the current state of AI: breathtaking potential mixed with frustrating limitations. While the video generation capabilities of Omni and the speed felt by designers using Stitch were genuinely exciting, the struggles with accessing promised features, the broken avatar creation, and the pervasive product naming confusion painted a picture of a revolution still very much under construction.

Ultimately, Claravevilleโ€™s key takeaways were clear: โ€œThe video piece is the part that Iโ€™m most excited about. probably second is just the speed of coding that comes from these Flash models.โ€ Yet, she concluded with a tempered optimism: โ€œthere are some sharp edges, some things that definitely need to be improved and some things that actually need to be released.โ€ Google I/O 2026 offered a tantalizing glimpse into a more intelligent, agentic, and creative future โ€“ one that, with a bit more polish and accessibility, might just live up to its ambitious promise.


Based on โ€œTrumpโ€™s Taxpayer-Funded Revenge Planโ€ from New York Times Podcasts Watch the original video

Taxpayer-Funded Retribution: Unpacking Trumpโ€™s $1.7 Billion Slush Fund

In a move that has sent shockwaves through Washington and ignited bipartisan fury, President Trumpโ€™s administration has unveiled an unprecedented $1.776 billion fund of taxpayer money. Ostensibly designed to compensate โ€œvictims of weaponization and warfareโ€ by the federal government, critics across the political spectrum are decrying it as a thinly veiled political slush fund, potentially funneling taxpayer dollars to the Presidentโ€™s loyalists, including those convicted for their roles in the January 6th Capitol insurrection.

The announcement came after President Trump abruptly dropped a contentious $10 million lawsuit he and his family had filed against the Internal Revenue Service (IRS). This sudden withdrawal, coupled with the creation of such a massive and vaguely defined compensation fund, has raised profound questions about conflicts of interest, the politicization of federal agencies, and the very fabric of American governance.

The Roots of a Peculiar Lawsuit

To understand the genesis of this controversial fund, one must rewind to President Trumpโ€™s first run for office in 2016. Breaking with decades of presidential tradition, Trump famously refused to release his tax returns, a decision that became a recurring point of contention throughout his presidency.

The Unreleased Tax Returns and the IRS Leak

The mystery surrounding Trumpโ€™s finances deepened in 2020 when The New York Times published a series of bombshell stories based on leaked tax returns, revealing that for many years, Trump had paid little to no federal income taxes. This was followed by ProPublica, another news organization, releasing detailed tax strategies of some of the worldโ€™s wealthiest individuals, including Jeff Bezos and Elon Musk.

It soon became clear that the source of this highly sensitive financial information was an internal IRS leaker. In the fall of 2023, Charles Littlejohn, a former IRS contractor, was caught. He pleaded guilty to providing this information to the Times and ProPublica and was subsequently sentenced to prison in early 2024. Leaking someoneโ€™s tax information is a federal crime, and civil lawsuits against the IRS are permitted for individuals whose tax data has been improperly released.

It was under this premise that President Trump, his two sons, and the Trump Organization filed a lawsuit in January, demanding at least $10 million from the IRS. Their argument: the IRS should have prevented Littlejohn from accessing and leaking their information.

A President Suing His Own Government

What made this lawsuit uniquely problematic was not just the claim, but the plaintiff himself. As President, Donald Trump controlled the very agencies he was suing. He appointed the leadership of the IRS and the Treasury Department, both named in the suit. Furthermore, the Justice Department (DOJ), tasked with defending the IRS, had been significantly reshaped under his influence. Adding another layer of complexity, the acting Attorney General, Todd Blanche, was also President Trumpโ€™s personal lawyer before returning to the White House.

This created an unprecedented web of conflicts. The President, as a private citizen, hired private lawyers to sue the government. Simultaneously, as the head of the executive branch, he effectively controlled the government lawyers who were supposed to defend against that very suit. As one commentator aptly put it, โ€œYou almost need a whiteboard to like map out all of these conflicts.โ€ Donald Trump, in essence, sat atop both parties in the lawsuit, raising profound questions about the legitimacy of any potential outcome.

The Judgeโ€™s Dilemma and the Sudden Deal

The inherent conflict did not go unnoticed by the judiciary. The presiding judge began to express serious concerns, questioning whether a genuine legal conflict even existed. Under legal principles, both sides in a lawsuit must be genuinely adversarial; one cannot sue oneself or a friend to manufacture a favorable ruling. The judge indicated she was considering dismissing the case outright, viewing it potentially as a โ€œcharadeโ€ filed under false pretenses. She ordered both the Justice Department and Trumpโ€™s private lawyers to submit briefs explaining why the lawsuit should be allowed to proceed.

Facing a potentially embarrassing and unfavorable ruling, the administration reportedly worked โ€œfeverishlyโ€ to avoid this scenario. The outcome was a dramatic pivot: President Trump would drop his lawsuit entirely. In exchange, the Justice Department would create an โ€œunprecedented gigantic pot of taxpayer moneyโ€ to compensate โ€œvictims of weaponization.โ€

The $1.776 Billion โ€œVictimโ€ Fund Takes Shape

The fund, totaling $1,776,000,000 โ€“ a figure many observers note is no coincidence, referencing the year of Americaโ€™s founding โ€“ will be drawn from the Judgment Fund, a standing account the Justice Department uses to settle legal claims against the United States. This sum, however, is far larger than typical withdrawals.

The money will be deposited into a separate account controlled by five individuals appointed by the Attorney General. These appointees will then consider claims from people who assert they have been victimized by the federal government.

Who Qualifies as a โ€œVictimโ€?

While the Justice Departmentโ€™s official language refers to โ€œvictims of weaponization and lawfare,โ€ the practical interpretation of this definition has sparked intense alarm. Among the primary groups expressing excitement about the fund are individuals convicted for their roles in the January 6th, 2021, Capitol insurrection โ€“ many of whom President Trump had previously pardoned. Others include Trump supporters and political allies who claim they were improperly investigated by the Justice Department or FBI under previous administrations simply for their political leanings.

The possibility that โ€œsomebody who stormed the Capitol on January 6th could get paid by the government would be shocking, to put it mildly,โ€ noted one observer. Indeed, the scenario of convicted criminals receiving taxpayer money from a government they attacked has fueled much of the public outrage. It remains unclear whether payments would be limited to covering legal fees or if individuals could โ€œprofit from the experience overall.โ€

A Veil of Secrecy

Crucially, many details about the fund remain shrouded in secrecy. The exact criteria for evaluating claims, the precise definition of โ€œweaponization,โ€ and how payment amounts will be determined are largely unknown. The identities of the five individuals who will control this enormous sum and make these critical decisions have also not been disclosed. This lack of transparency has led to concerns that the entire operation โ€œmight be very hidden from public view,โ€ allowing for discretionary payouts with little public accountability.

Outrage on Capitol Hill and Beyond

The creation of the fund immediately triggered a firestorm on Capitol Hill. Acting Attorney General Todd Blanche found himself defending the arrangement during a scheduled testimony before Congress. He argued that while โ€œunusual,โ€ it was โ€œnot unprecedentedโ€ and claimed the fund was โ€œnot limited to Republicans,โ€ โ€œnot limited to January 6th,โ€ and open to โ€œanybody in this countryโ€ฆ if they believe they are a victim of weaponization.โ€

Democrats Cry Foul, Republicans Get โ€œSqueamishโ€

However, Blancheโ€™s defense did little to quell the outrage. Democratic lawmakers lambasted the fund as an โ€œoutrageous unprecedented slush fundโ€ designed to pay the presidentโ€™s political allies. Senator Chris Van Hollen accused Blanche of โ€œpure theft of public fundsโ€ and called the idea of rewarding convicted criminals โ€œobscene.โ€ He further asserted that it was part of President Trumpโ€™s โ€œbroader revenge tourโ€ and bluntly told Blanche, โ€œYou are still acting as the presidentโ€™s personal lawyer, not as acting Attorney General.โ€

Even Republicans, traditionally staunch allies of the President, showed signs of discomfort. Senate Majority Leader John Thune, while measured, stated that โ€œthere are and will be continue to be a lot of questions around that that the administration is going to have to answer.โ€ This โ€œmild discontentโ€ from top Republican lawmakers signaled that even within Trumpโ€™s own party, there was a distinct lack of enthusiasm or pride in the arrangement.

Internal Dissent

The controversy also reportedly spilled into the administration itself. The top lawyer at the Treasury Department, an agency involved in the fundโ€™s creation, suddenly resigned on Monday, just hours after the announcement. Initial reports suggest this resignation was, at least in part, due to the creation of the fund, indicating significant internal dissent over the move.

Trumpโ€™s Personal Windfall: Beyond the Fund

While the President, his sons, and the Trump Organization are explicitly barred from receiving payouts directly from the $1.776 billion fund, the agreement that led to its creation includes another extraordinary concession that could directly benefit Trump financially. The IRS will drop any audits it has of Trump, his family members, and any related business entities.

This is an โ€œextremely unusual and extraordinary concession,โ€ as audits can result in significant tax liabilities. Previous reports have indicated that the President has faced audits in the past where, if the IRS prevailed, he could owe more than a hundred million dollars in additional taxes. While the specifics of current audits remain unknown, this provision effectively โ€œrobbed the IRS of the opportunity to demand that the Trumps pay the taxes that the IRS says that they may owe under the law.โ€ This could directly deliver โ€œtens of millions of dollarsโ€ or more in financial benefit to the President, representing a substantial personal windfall.

The Unanswered Question: Can It Be Stopped?

The question now looms: can anything be done to stop this controversial fund? Attempts by House Democrats to intervene in Trumpโ€™s original lawsuit against the IRS were ultimately unsuccessful. Now, the focus shifts to Congressโ€™s โ€œpower of the purseโ€ โ€“ its constitutional authority to control federal spending. Theoretically, Congress could pass laws to address how money can be spent on legal settlements or specifically target this fund. However, this power has been โ€œerodedโ€ under the current presidency, and it remains โ€œunclear if that will actually happen.โ€

A New Chapter in Political Retribution

For many ethics experts and watchdogs, this fund marks a dangerous escalation in the politicization of government. It represents a significant advancement in President Trumpโ€™s stated goal of โ€œretribution against people and agencies that he saw as having been opposed to him.โ€ What began as a campaign to punish perceived enemies has now, according to critics, evolved into direct โ€œcompensation for President Trumpโ€™s alliesโ€ with taxpayer dollars.

Ethicists are calling this โ€œthe most corrupt act in American history,โ€ highlighting the almost complete lack of limits or controls on how this money will be used. Five individuals, appointed by the Presidentโ€™s former personal lawyer, will have โ€œseemingly total freedom to give money to whoever they deem is worthy of it and to give them as much money as they want.โ€

This โ€œradical and striking situationโ€ means that those deemed โ€œloyal to the presidentโ€ could be โ€œcashing out to the tunes of thousands or tens of thousands, hundreds of thousands, or even millions of dollars.โ€ It signifies a profound shift, not just in punishing enemies, but in directly rewarding allies using the publicโ€™s money, fundamentally challenging the principles of impartial governance and public trust.


ํ•œ๊ตญ์–ด

โ€œAnthropic Just Bought a Dev Tools Startup for $300M. Hereโ€™s What Its Founder Told Me.โ€ โ€” Every ๊ธฐ๋ฐ˜ ๊ธฐ์‚ฌ ์›๋ณธ ์˜์ƒ ๋ณด๊ธฐ

AI ์‹œ๋Œ€, ์ปดํ“จํ„ฐ ์†Œํ†ต์˜ ๋ฏธ๋ž˜๋ฅผ ๋ฌป๋‹ค: Stainless CEO ์•Œ๋ ‰์Šค ๋žซ๋ ˆ์ด์™€์˜ ๋Œ€ํ™”

์ธํ„ฐ๋„ท์€ ์ปดํ“จํ„ฐ๋“ค์ด ์„œ๋กœ ๋Œ€ํ™”ํ•˜๋ฉฐ ์ž‘๋™ํ•˜์ง€๋งŒ, ๊ทธ ์ „์ฒด์ ์ธ ์•„ํ‚คํ…์ฒ˜(architecture)๋Š” ์ธ๊ณต์ง€๋Šฅ(AI)์ด ๋“ฑ์žฅํ•˜๊ธฐ ์ „์˜ ์„ธ์ƒ์— ๋งž์ถฐ ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด์ œ ์šฐ๋ฆฌ๋Š” ๋ชจ๋ธ ์ปจํ…์ŠคํŠธ ํ”„๋กœํ† ์ฝœ(Model Context Protocol, MCP)์ด๋ผ๋Š” ์ƒˆ๋กœ์šด ๋ฐฉ์‹์œผ๋กœ AI๋ฅผ ์ธํ„ฐ๋„ท์— ์—ฐ๊ฒฐํ•˜๋ ค ํ•ฉ๋‹ˆ๋‹ค. MCP๋Š” ๋ชจ๋“  ์›น์‚ฌ์ดํŠธ๋‚˜ ์›น ์„œ๋น„์Šค๋ฅผ AI๊ฐ€ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๋ฐ ๊ธฐ๋ณธ์ ์œผ๋กœ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๋„๊ตฌ ์„ธํŠธ๋กœ ์ „ํ™˜ํ•ฉ๋‹ˆ๋‹ค. ์•ž์œผ๋กœ 10๋…„ ๋™์•ˆ MCP๋ฅผ ์ž˜ ํ™œ์šฉํ•˜๋Š” ์†Œํ”„ํŠธ์›จ์–ด ๊ธฐ์—…๋“ค์ด ์‹œ์žฅ์„ ์„ ๋„ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์ด๋Ÿฌํ•œ ๋ณ€ํ™”์˜ ์ค‘์‹ฌ์—์„œ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•˜๋Š” ๊ธฐ์—…์ด ๋ฐ”๋กœ Stainless์ž…๋‹ˆ๋‹ค. Stainless๋Š” ์ปดํ“จํ„ฐ๋“ค์ด ์„œ๋กœ ์›ํ™œํ•˜๊ฒŒ ์†Œํ†ตํ•˜๋„๋ก ๋•๋Š” ํšŒ์‚ฌ๋กœ, OpenAI, Anthropic๊ณผ ๊ฐ™์€ ๋Œ€๊ธฐ์—…๋“ค์„ ์œ„ํ•œ API(Application Programming Interface) ๋ฐ SDK(Software Development Kit)๋ฅผ ๊ฐœ๋ฐœํ•ฉ๋‹ˆ๋‹ค. ์ตœ๊ทผ์—๋Š” MCP ์„œ๋ฒ„ ๊ตฌ์ถ•์—๋„ ์ ๊ทน์ ์œผ๋กœ ๋‚˜์„œ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

Every ์ฑ„๋„์€ Stainless์˜ ์ฐฝ๋ฆฝ์ž์ด์ž CEO์ธ ์•Œ๋ ‰์Šค ๋žซ๋ ˆ์ด(Alex Ratray)๋ฅผ ์ดˆ๋Œ€ํ•˜์—ฌ MCP์˜ ๋ฏธ๋ž˜, ํšจ๊ณผ์ ์ธ MCP ์„ค๊ณ„ ๋ฐฉ๋ฒ•, ํ™•์žฅ์„ฑ ๋ฐ ๋ณด์•ˆ ๋ฌธ์ œ, ๊ทธ๋ฆฌ๊ณ  AI๊ฐ€ ์ธํ„ฐ๋„ท์„ ๋” ์ž˜ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๋ชจ๋ธ์— ๋Œ€ํ•œ ์‹ฌ์ธต์ ์ธ ๋Œ€ํ™”๋ฅผ ๋‚˜๋ˆด์Šต๋‹ˆ๋‹ค.


์•Œ๋ ‰์Šค ๋žซ๋ ˆ์ด์™€ Stainless: ์ปดํ“จํ„ฐ ์†Œํ†ต์˜ ์ „๋ฌธ๊ฐ€

์•Œ๋ ‰์Šค ๋žซ๋ ˆ์ด๋Š” Stainless์˜ ์ฐฝ๋ฆฝ์ž์ด์ž CEO์ž…๋‹ˆ๋‹ค. ๊ทธ๋Š” ์ด์ „์— Stripe์—์„œ API ๊ฐœ๋ฐœ ์—…๋ฌด๋ฅผ ๋‹ด๋‹นํ–ˆ์œผ๋ฉฐ, ํ˜„์žฌ Stainless๋ฅผ ํ†ตํ•ด ์ปดํ“จํ„ฐ ๊ฐ„์˜ ์†Œํ†ต์„ ์œ„ํ•œ API๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ OpenAI, Anthropic ๋“ฑ ์ฃผ์š” ๊ธฐ์—…๋“ค์˜ ๊ธฐ์ˆ  ๊ธฐ๋ฐ˜์„ ๋‹ค์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์•Œ๋ ‰์Šค๋Š” ์ง„ํ–‰์ž ๋Œ„(Dan)๊ณผ ๋Œ€ํ•™ ์‹œ์ ˆ๋ถ€ํ„ฐ ์นœ๊ตฌ์˜€์œผ๋ฉฐ, ์„œ๋กœ ์Šคํƒ€ํŠธ์—…์„ ์‹œ์ž‘ํ•˜๋ฉฐ ํ•จ๊ป˜ ์„ฑ์žฅํ•ด์˜จ ํŠน๋ณ„ํ•œ ์ธ์—ฐ์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์•Œ๋ ‰์Šค๋Š” Stainless์˜ ๋ฏธ์…˜์ด โ€œ์ปดํ“จํ„ฐ๋“ค์ด ์„œ๋กœ ์‰ฝ๊ฒŒ ๋Œ€ํ™”ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋งŒ๋“œ๋Š” ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด ๋จผ์ € API์˜ ๊ฐœ๋…๋ถ€ํ„ฐ ์งš๊ณ  ๋„˜์–ด๊ฐ‘๋‹ˆ๋‹ค.

API: ์ธํ„ฐ๋„ท์˜ ์‹ ๊ฒฝ๋ง

API๋Š” ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์ธํ„ฐํŽ˜์ด์Šค(Application Programming Interface)์˜ ์•ฝ์ž๋กœ, ํ•œ ์ปดํ“จํ„ฐ ํ”„๋กœ๊ทธ๋žจ์ด ๋‹ค๋ฅธ ์ปดํ“จํ„ฐ ํ”„๋กœ๊ทธ๋žจ๊ณผ ๋Œ€ํ™”ํ•˜๋Š” ๋ฐฉ์‹, ์ฆ‰ ์•ฑ๊ณผ ์•ฑ์ด ์†Œํ†ตํ•˜๋Š” ํ†ต๋กœ๋ฅผ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์•Œ๋ ‰์Šค๋Š” API๋ฅผ โ€œ์ธํ„ฐ๋„ท์˜ ์ˆ˜์ƒ๋Œ๊ธฐ(dendrite)โ€œ์— ๋น„์œ ํ•ฉ๋‹ˆ๋‹ค. ๋‡Œ์˜ ๋‰ด๋Ÿฐ๋“ค์ด ์ˆ˜์ƒ๋Œ๊ธฐ๋ฅผ ํ†ตํ•ด ์ •๋ณด๋ฅผ ๊ตํ™˜ํ•ด์•ผ ์‚ฌ๊ณ ๊ฐ€ ๊ฐ€๋Šฅํ•˜๋“ฏ์ด, ํด๋ผ์šฐ๋“œ์˜ ๋ชจ๋“  ์„œ๋ฒ„๋“ค์ด ์„œ๋กœ ๋Œ€ํ™”ํ•˜์ง€ ์•Š๋Š”๋‹ค๋ฉด ์ธํ„ฐ๋„ท์€ ์กด์žฌํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ํ˜„๋Œ€ ์†Œํ”„ํŠธ์›จ์–ด์˜ ๊ฑฐ์˜ ๋ชจ๋“  ๋ถ€๋ถ„์ด API๋ฅผ ํ†ตํ•ด ์—ฐ๊ฒฐ๋˜์–ด ์žˆ์œผ๋ฉฐ, ์ด๋Š” ๊ธฐ์ˆ  ์ž๋™ํ™”์˜ ํ•ต์‹ฌ ๋™๋ ฅ์ž…๋‹ˆ๋‹ค.

ํ•˜์ง€๋งŒ AI์˜ ๋“ฑ์žฅ์œผ๋กœ ์ƒˆ๋กœ์šด ์œ ํ˜•์˜ โ€œ์ปดํ“จํ„ฐโ€๊ฐ€ ์ด ๋Œ€ํ™”์— ์ฐธ์—ฌํ•˜๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ณผ๊ฑฐ์—๋Š” ์ธ๊ฐ„์ด ์‚ฌ์šฉ์ž ์ธํ„ฐํŽ˜์ด์Šค(UI)๋ฅผ ํ†ตํ•ด ์ปดํ“จํ„ฐ์™€ ์ƒํ˜ธ์ž‘์šฉํ•˜๊ฑฐ๋‚˜, ์ปดํ“จํ„ฐ๊ฐ€ API๋ฅผ ํ†ตํ•ด ๋‹ค๋ฅธ ์ปดํ“จํ„ฐ์™€ ๋Œ€ํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด์ œ๋Š” ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(Large Language Model, LLM)์ด ์ปดํ“จํ„ฐ์™€ ์ƒํ˜ธ์ž‘์šฉํ•ด์•ผ ํ•˜๋Š” ์‹œ๋Œ€๊ฐ€ ์˜จ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋ฐ”๋กœ ์ด ์ง€์ ์—์„œ MCP, ์ฆ‰ ๋ชจ๋ธ ์ปจํ…์ŠคํŠธ ํ”„๋กœํ† ์ฝœ์ด ์ค‘์š”ํ•ด์ง‘๋‹ˆ๋‹ค. MCP๋Š” LLM์„ ์ปดํ“จํ„ฐ์— ์—ฐ๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ ์‹œ์Šคํ…œ์œผ๋กœ, Stainless๊ฐ€ ์ง‘์ค‘์ ์œผ๋กœ ํˆฌ์žํ•˜๊ณ  ์žˆ๋Š” ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค.

SDK์™€ MCP: ์ธ๊ฐ„๊ณผ AI๋ฅผ ์œ„ํ•œ ์ธํ„ฐํŽ˜์ด์Šค

Stainless์˜ ํ•ต์‹ฌ ์ œํ’ˆ์€ ์†Œํ”„ํŠธ์›จ์–ด ๊ฐœ๋ฐœ ํ‚คํŠธ(Software Development Kit, SDK)์ž…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, Stripe์˜ REST API๋ฅผ ํŒŒ์ด์ฌ(Python) ๊ฐœ๋ฐœ์ž๊ฐ€ ํŽธ๋ฆฌํ•˜๊ฒŒ ์‚ฌ์šฉํ•˜๋ ค๋ฉด pip install stripe ๋ช…๋ น์œผ๋กœ Stripe ํŒŒ์ด์ฌ SDK๋ฅผ ์„ค์น˜ํ•œ ํ›„ Stripe.customers.create์™€ ๊ฐ™์€ ์ฝ”๋“œ๋ฅผ ํ†ตํ•ด ์†์‰ฝ๊ฒŒ ๊ณ ๊ฐ ๊ฐ์ฒด๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. SDK๋Š” ๊ฐœ๋ฐœ์ž๋“ค์ด API์™€ ์‰ฝ๊ฒŒ ์ƒํ˜ธ์ž‘์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ๋•๋Š” ๋„๊ตฌ์ž…๋‹ˆ๋‹ค.

๊ทธ๋ ‡๋‹ค๋ฉด LLM์ด API์™€ ์‰ฝ๊ฒŒ ์ƒํ˜ธ์ž‘์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์€ ๋ฌด์—‡์ผ๊นŒ์š”? ๋ฐ”๋กœ MCP์ž…๋‹ˆ๋‹ค. ์ด๋ก ์ ์œผ๋กœ MCP๋Š” ์›น์‚ฌ์ดํŠธ๊ฐ€ ์ธ๊ฐ„์„ ์œ„ํ•ด ์„ค๊ณ„๋œ ๊ฒƒ์ฒ˜๋Ÿผ, LLM์„ ์œ„ํ•ด ๋„๊ตฌ ์„ธํŠธ๋ฅผ ๋…ธ์ถœํ•˜๋Š” ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, Gmail MCP๋Š” โ€˜๋ฉ”์ผ ๋ณด๋‚ด๊ธฐโ€™, โ€˜๋ฉ”์ผ ์ž‘์„ฑํ•˜๊ธฐโ€™, โ€˜๋ฐ›์€ ํŽธ์ง€ํ•จ ์ฝ๊ธฐโ€™์™€ ๊ฐ™์€ ๋„๊ตฌ๋“ค์„ LLM์— ์ œ๊ณตํ•˜์—ฌ, LLM์ด ์ง์ ‘ ๋กœ๊ทธ์ธํ•˜์—ฌ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค.

ํ•˜์ง€๋งŒ ์•Œ๋ ‰์Šค๋Š” ํ˜„์žฌ MCP์˜ ๊ตฌํ˜„์ด ์•„์ง ์ด์ƒ์ ์ด์ง€ ์•Š๋‹ค๊ณ  ์ง€์ ํ•ฉ๋‹ˆ๋‹ค. โ€œ์‚ฌ๋žŒ๋“ค์ด MCP๋ฅผ ์‹คํ—˜ํ•˜๊ณ  ์žˆ์ง€๋งŒ, ๊ธฐ๋Œ€๋งŒํผ ์ž˜ ์ž‘๋™ํ•˜์ง€ ์•Š๋Š”๋‹คโ€๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.


MCP์˜ ์•ฝ์†๊ณผ ํ˜„์‹ค: โ€˜์—์ด์ „ํŠธ AIโ€™์˜ ๋น„์ „

์•Œ๋ ‰์Šค๋Š” MCP์™€ ์—์ด์ „ํŠธ AI(Agentic AI)์˜ ๊ถ๊ทน์ ์ธ ๋น„์ „์„ ์„ค๋ช…ํ•˜๊ธฐ ์œ„ํ•ด ๋งค์šฐ ์ผ์ƒ์ ์ธ ์˜ˆ์‹œ๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.

์ผ์ƒ ์—…๋ฌด์˜ ์ž๋™ํ™”: ์ค„๋ฌด๋Šฌ ์–‘๋ง ํ™˜๋ถˆ ์‚ฌ๋ก€

์–ด๋–ค ๊ณ ๊ฐ(๋Œ„)์ด ๊ฐ€๊ฒŒ์—์„œ ์ค„๋ฌด๋Šฌ ์–‘๋ง์„ ํฌํ•จํ•œ ๋ช‡ ๊ฐ€์ง€ ๋ฌผ๊ฑด์„ ๊ตฌ๋งคํ–ˆ๋Š”๋ฐ, ๋‹ค์Œ ๋‚  ๋ฌธ์ œ๊ฐ€ ๋ฐœ์ƒํ•˜์—ฌ ํ™˜๋ถˆ์„ ์š”์ฒญํ–ˆ๋‹ค๊ณ  ๊ฐ€์ •ํ•ด ๋ด…์‹œ๋‹ค. ๋งค์žฅ ์ง์›์ด โ€œ๋Œ„์ด ์–ด์ œ ์‚ฐ ์ค„๋ฌด๋Šฌ ์–‘๋ง์„ ํ™˜๋ถˆํ•ด์ฃผ๊ณ , ๋‹ค์Œ ๋ฐฉ๋ฌธ ์‹œ ์‚ฌ์šฉํ•  ํ• ์ธ ์ฝ”๋“œ์™€ ๊ฐ์‚ฌ ํŽธ์ง€๋ฅผ ๋ณด๋‚ด์ค˜โ€๋ผ๊ณ  ์ง€์‹œํ•ฉ๋‹ˆ๋‹ค.

ํ˜„์žฌ์˜ ๋ฐฉ์‹์œผ๋กœ๋Š” ์ง์›์ด ์—ฌ๋Ÿฌ ๋‹จ๊ณ„๋ฅผ ๊ฑฐ์ณ์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋‚ด๋ถ€ ๊ด€๋ฆฌ ์‹œ์Šคํ…œ์„ ์—ด๊ณ , Stripe ๋Œ€์‹œ๋ณด๋“œ์—์„œ ๊ฒฐ์ œ ๋‚ด์—ญ์„ ํ™•์ธํ•˜๋ฉฐ โ€˜๋Œ„โ€™์ด๋ผ๋Š” ์ด๋ฆ„๊ณผ โ€˜์ค„๋ฌด๋Šฌ ์–‘๋งโ€™์ด ํฌํ•จ๋œ ์ฃผ๋ฌธ์„ ์ฐพ์Šต๋‹ˆ๋‹ค. ์—ฌ๋Ÿฌ ๋ฒˆ์˜ ํด๋ฆญ๊ณผ ์Šคํฌ๋กค์„ ๊ฑฐ์ณ ํ•ด๋‹น ์ฃผ๋ฌธ์„ ์ฐพ์€ ํ›„ ํ™˜๋ถˆ ํ™”๋ฉด์œผ๋กœ ์ด๋™ํ•˜์—ฌ ์ •ํ™•ํ•œ ๊ธˆ์•ก์„ ํ™˜๋ถˆ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค. ์ด์–ด์„œ ํ• ์ธ ์ฝ”๋“œ๋ฅผ ์ƒ์„ฑํ•˜๊ณ , ๋‹ค๋ฅธ SaaS(์„œ๋น„์Šคํ˜• ์†Œํ”„ํŠธ์›จ์–ด) ์•ฑ์— ๋กœ๊ทธ์ธํ•˜์—ฌ ๊ฐ์‚ฌ ํŽธ์ง€์™€ ํ•จ๊ป˜ ํ• ์ธ ์ฝ”๋“œ๋ฅผ ๋ณด๋ƒ…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ•˜๋‚˜์˜ ๊ฐ„๋‹จํ•œ ์ž‘์—…์„ ์œ„ํ•ด 5๊ฐœ์˜ ๋‹ค๋ฅธ ์•ฑ์„ ์˜ค๊ฐ€๋ฉฐ 15๋ฒˆ์˜ ํด๋ฆญ๊ณผ ๋กœ๋”ฉ์„ ๊ธฐ๋‹ค๋ ค์•ผ ํ•˜๋Š” ๊ฒƒ์ด ํ˜„์‹ค์ž…๋‹ˆ๋‹ค.

์—์ด์ „ํŠธ AI์˜ ์•ฝ์†์€ ๋ฐ”๋กœ ์ด ์ง€์ ์—์„œ ๋น›์„ ๋ฐœํ•ฉ๋‹ˆ๋‹ค. ์ง์›์ด ChatGPT๋‚˜ Claude ๊ฐ™์€ AI์—๊ฒŒ โ€œ์ฑ—๋ด‡ ์นœ๊ตฌ, ๋‚ด ์นœ๊ตฌ ๋Œ„ D.์˜ ํ™˜๋ถˆ์„ ๋„์™€์ค„ ์ˆ˜ ์žˆ์„๊นŒ?โ€๋ผ๊ณ  ์ž…๋ ฅํ•˜๋ฉด, AI๊ฐ€ ์Šค์Šค๋กœ 5๊ฐœ์˜ ์•ฑ๊ณผ 15๊ฐœ์˜ ํ™”๋ฉด์„ ์˜ค๊ฐ€๋ฉฐ ํ•„์š”ํ•œ ์ž‘์—…์„ ์™„๋ฃŒํ•˜๊ณ  โ€œ์™„๋ฃŒํ–ˆ์Šต๋‹ˆ๋‹คโ€๋ผ๊ณ  ๋ณด๊ณ ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

๊ด‘๋ฒ”์œ„ํ•œ ๋„๊ตฌ ๋…ธ์ถœ์˜ ๋‚œ์ œ: ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ ํ•œ๊ณ„

์—์ด์ „ํŠธ AI๊ฐ€ ์ด๋Ÿฌํ•œ ๋น„์ „์„ ์‹คํ˜„ํ•˜๋ ค๋ฉด, ์ธ๊ฐ„ ์šด์˜์ž๊ฐ€ ํ•  ์ˆ˜ ์žˆ๋Š” ๋ชจ๋“  ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋‹จ์ˆœํžˆ โ€˜ํ™˜๋ถˆ ์ƒ์„ฑโ€™์ด๋‚˜ โ€˜๊ฑฐ๋ž˜ ๋‚ด์—ญ ์กฐํšŒโ€™ ๊ฐ™์€ ๋ช‡ ๊ฐ€์ง€ ๋„๊ตฌ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ, Stripe ๋Œ€์‹œ๋ณด๋“œ์—์„œ ๊ฐ€๋Šฅํ•œ ๋ชจ๋“  ๊ธฐ๋Šฅ, ์ฆ‰ Stripe API์—์„œ ์ œ๊ณตํ•˜๋Š” ์ˆ˜๋ฐฑ ๊ฐœ์˜ ์—”๋“œํฌ์ธํŠธ(endpoint)๋ฅผ ๋ชจ๋‘ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ์–ด์•ผ ํ•จ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.

๋ฌธ์ œ๋Š” ์—ฌ๊ธฐ์— ์žˆ์Šต๋‹ˆ๋‹ค. ํ˜„์žฌ LLM์— ์ด๋Ÿฌํ•œ ๋ฐฉ๋Œ€ํ•œ ๋„๊ตฌ ๋ชฉ๋ก๊ณผ ๊ฐ ๋„๊ตฌ์˜ ์„ค๋ช…, ์š”์ฒญ/์‘๋‹ต ์†์„ฑ, ์ƒ์„ธ ๋ฌธ์„œ ๋“ฑ์„ MCP ์ •์˜๋กœ ์ œ๊ณตํ•˜๋ฉด, ์ˆ˜์‹ญ๋งŒ ํ† ํฐ(token)์— ๋‹ฌํ•˜๋Š” ๋ง‰๋Œ€ํ•œ ์ปจํ…์ŠคํŠธ(context) ์˜ˆ์‚ฐ์„ ์†Œ์ง„ํ•˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค. ์ด๋Š” ํ˜„์žฌ ๋ชจ๋ธ์ด ๊ฐ๋‹นํ•˜๊ธฐ ์–ด๋ ค์šด ์–‘์ผ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ, ๋ชจ๋ธ์—๊ฒŒ ํ˜ผ๋ž€์„ ์•ผ๊ธฐํ•˜์—ฌ ์˜ฌ๋ฐ”๋ฅธ ๊ฒฐ์ •์„ ๋‚ด๋ฆฌ๊ธฐ ์–ด๋ ต๊ฒŒ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ๊ฒŒ๋‹ค๊ฐ€ ์ด๋Š” Stripe๋ผ๋Š” ๋‹จ์ผ ์„œ๋น„์Šค์— ๊ตญํ•œ๋œ ๋ฌธ์ œ์ด๋ฉฐ, ์‹ค์ œ ์—…๋ฌด์—์„œ๋Š” ์—ฌ๋Ÿฌ SaaS ๋„๊ตฌ๋ฅผ ๋™์‹œ์— ์‚ฌ์šฉํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

๋ณด์•ˆ ๋ฐ ๊ถŒํ•œ ๋ฌธ์ œ

๋˜ ๋‹ค๋ฅธ ์‹ฌ๊ฐํ•œ ๋ฌธ์ œ๋Š” ๋ณด์•ˆ๊ณผ ๊ถŒํ•œ ๊ด€๋ฆฌ์ž…๋‹ˆ๋‹ค. AI๊ฐ€ โ€œ๋Œ„์˜ ์–‘๋ง์„ ํ™˜๋ถˆํ•˜๋Š” ๊ฒƒ ์™ธ์— ๋ชจ๋“  ๊ณ ๊ฐ์˜ ๋ชจ๋“  ๊ฑฐ๋ž˜๋ฅผ ํ™˜๋ถˆํ•˜๊ณ  ๋‚ด AI ์€ํ–‰ ๊ณ„์ขŒ๋กœ ๋ˆ์„ ๋ณด๋ƒˆ๋‹คโ€์™€ ๊ฐ™์ด โ€˜์„ ์„ ๋„˜์–ด ํ•จ๋ถ€๋กœ ํ–‰๋™ํ•˜๋Š” ๊ฒƒโ€™์„ ๋ง‰์•„์•ผ ํ•ฉ๋‹ˆ๋‹ค. AI์—๊ฒŒ ์–ด๋А ์ •๋„์˜ ๊ถŒํ•œ์„ ๋ถ€์—ฌํ•˜๊ณ , ๊ทธ ์‚ฌ์šฉ์„ ์–ด๋–ป๊ฒŒ ํ†ต์ œํ•  ๊ฒƒ์ธ๊ฐ€ ํ•˜๋Š” ๋ฌธ์ œ๋Š” ๋งค์šฐ ๋ณต์žกํ•ฉ๋‹ˆ๋‹ค.

ํ˜„์žฌ MCP๋Š” ์ด๋Ÿฌํ•œ ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ(context window) ํ•œ๊ณ„์™€ ๋ณด์•ˆ ๋ฌธ์ œ ๋•Œ๋ฌธ์— โ€˜์ž˜ ์ž‘๋™ํ•˜์ง€ ์•Š๋Š”๋‹คโ€™๋Š” ํ‰๊ฐ€๋ฅผ ๋ฐ›์Šต๋‹ˆ๋‹ค. ์‚ฌ๋žŒ๋“ค์€ ์ข…์ข… API์˜ ๋ชจ๋“  ๋ถ€๋ถ„์„ ๋…ธ์ถœํ•˜๊ธฐ๋ณด๋‹ค, ํŠน์ • LLM์— ๋งž์ถฐ ์ˆ˜์ž‘์—…์œผ๋กœ ์ตœ์ ํ™”๋œ ์†Œ์ˆ˜์˜ ํŠนํ™”๋œ ๋„๊ตฌ๋ฅผ ๋งŒ๋“ค๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์˜ โ€˜์ธ์ฒด ๊ณตํ•™(ergonomics)โ€˜์„ ๊ณ ๋ คํ•ด์•ผ ํ•˜๋Š” ์ƒˆ๋กœ์šด ์—ฐ๊ตฌ ๊ณผ์ œ์™€ ๊ฐ™์Šต๋‹ˆ๋‹ค.


์ข‹์€ MCP ์„œ๋ฒ„ ์„ค๊ณ„์˜ ์›์น™

์•Œ๋ ‰์Šค๋Š” ํ˜„์žฌ์˜ ๊ธฐ์ˆ  ์ˆ˜์ค€์—์„œ ํšจ๊ณผ์ ์ธ MCP ์„œ๋ฒ„๋ฅผ ๊ตฌ์ถ•ํ•˜๊ธฐ ์œ„ํ•œ ์›์น™๋“ค์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.

  1. ์ œํ’ˆ ๊ด€๋ฆฌ ์ค‘์‹ฌ ์ ‘๊ทผ: ์‹œ์žฅ์— ๋‚˜๊ฐ€ ๊ณ ๊ฐ์˜ ์‹ค์ œ ๋‹ˆ์ฆˆ๋ฅผ ํŒŒ์•…ํ•˜๊ณ , ์†Œํ”„ํŠธ์›จ์–ด๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ชจ์Šต์„ ๊ด€์ฐฐํ•˜๋ฉฐ AI๋ฅผ ํ†ตํ•ด ์–ด๋–ค ์ƒˆ๋กœ์šด ๊ฐ€์น˜๋ฅผ ์ฐฝ์ถœํ•  ์ˆ˜ ์žˆ์„์ง€ ๊ณ ๋ฏผํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
  2. ์ •๊ตํ•œ ์—”์ง€๋‹ˆ์–ด๋ง ์ž‘์—…: ๋ชจ๋ธ์— ๋งž์ถฐ ๋„๊ตฌ๋ฅผ ์ •๊ตํ•˜๊ฒŒ ๋ž˜ํ•‘(wrapping)ํ•˜๊ณ , ๋‹ค์–‘ํ•œ ํด๋ผ์ด์–ธํŠธ(Cursor, Cloud Code ๋“ฑ)์™€ ๊ธฐ๋ณธ ๋ชจ๋ธ์„ ๊ณ ๋ คํ•œ ํ‰๊ฐ€ ์‹œ์Šคํ…œ(eval system)์„ ๊ตฌ์ถ•ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
  3. ํ”ผ๋“œ๋ฐฑ ๋ฉ”์ปค๋‹ˆ์ฆ˜: ํ˜„์žฌ๋Š” AI๊ฐ€ ๋„๊ตฌ ํ˜ธ์ถœ(tool call)์„ ์„ฑ๊ณต์ ์œผ๋กœ ์ˆ˜ํ–‰ํ–ˆ๋Š”์ง€, ์‚ฌ์šฉ์ž์—๊ฒŒ ์œ ์šฉํ–ˆ๋Š”์ง€ ํ”ผ๋“œ๋ฐฑ์„ ๋ฐ›๊ธฐ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. โ€˜ํ”ผ๋“œ๋ฐฑ ๋ณด๋‚ด๊ธฐ(send feedback)โ€˜์™€ ๊ฐ™์€ ๋„๊ตฌ๋ฅผ MCP์— ํฌํ•จํ•˜์—ฌ, ์‚ฌ์šฉ์ž๊ฐ€ โ€œ์ด๊ฑฐ ์ •๋ง ์“ธ๋ชจ์—†์—ˆ์–ดโ€๋ผ๊ณ  ๋งํ•˜๋ฉด MCP ์„œ๋ฒ„๊ฐ€ ์ด๋ฅผ ์ธ์ง€ํ•˜๊ณ  ๊ฐœ์„ ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
  4. ๊ตฌ์ฒด์ ์ธ ์„ค๊ณ„ ์›์น™:
    • ๋„๊ตฌ ์ˆ˜ ์ตœ์†Œํ™”: ๋„๊ตฌ์˜ ๊ฐœ์ˆ˜๋ฅผ ์ƒ๋Œ€์ ์œผ๋กœ ์ ๊ฒŒ ์œ ์ง€ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
    • ์ •ํ™•ํ•œ ๋„๊ตฌ ์ด๋ฆ„ ๋ฐ ์„ค๋ช…: ๋„๊ตฌ ์ด๋ฆ„๊ณผ ์„ค๋ช…์„ ๋งค์šฐ ์ •ํ™•ํ•˜๊ณ  ๊ตฌ์ฒด์ ์œผ๋กœ ์ž‘์„ฑํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. (์ด๊ฒƒ์ด ๋งค์šฐ ์–ด๋ ต์Šต๋‹ˆ๋‹ค.)
    • ๊ฐ„๊ฒฐํ•œ ์ž…๋ ฅ ์Šคํ‚ค๋งˆ: ์ž…๋ ฅ ํŒŒ๋ผ๋ฏธํ„ฐ(parameter)์˜ ์ˆ˜๋ฅผ ์ ๊ฒŒ ํ•˜๊ณ , ๊ฐ„๊ฒฐํ•˜๋ฉด์„œ๋„ ์ถฉ๋ถ„ํžˆ ์„ค๋ช…๋˜์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
    • ์ •ํ™•ํ•œ ์‘๋‹ต ๋ฐ์ดํ„ฐ: ๋ชจ๋ธ์ด ํ•„์š”๋กœ ํ•˜๋Š” ์ตœ์†Œํ•œ์˜ ๋ฐ์ดํ„ฐ๋งŒ ์‘๋‹ต์œผ๋กœ ๋ฐ˜ํ™˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. (์–ด๋–ค ๋ฐ์ดํ„ฐ๊ฐ€ ํ•„์š”ํ• ์ง€ ๋ฏธ๋ฆฌ ์•Œ๊ธฐ ์–ด๋ ต๊ธฐ ๋•Œ๋ฌธ์— ์ด ์—ญ์‹œ ์–ด๋ ต์Šต๋‹ˆ๋‹ค.) Stainless๋Š” ๋ชจ๋ธ์— JSON ๋ฐ์ดํ„ฐ๋ฅผ ํ•„ํ„ฐ๋งํ•˜๋Š” JQ ํ•„ํ„ฐ(JQ filter)๋ฅผ ์ œ๊ณตํ•˜๋Š” ๊ธฐ์ˆ ์„ ์‚ฌ์šฉํ•˜๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค.

๋„๊ตฌ ํƒ์ƒ‰(Tool Browsing) ๋ฌธ์ œ์™€ ๋™์  ๋ชจ๋“œ(Dynamic Mode)

๋„๊ตฌ ํƒ์ƒ‰ ๋ฌธ์ œ๋Š” ๋ฐฉ๋Œ€ํ•œ ๋„๊ตฌ ๋ชฉ๋ก ์ค‘์—์„œ AI๊ฐ€ ์ ์ ˆํ•œ ๋„๊ตฌ๋ฅผ ์ฐพ์•„๋‚ด๋Š” ์–ด๋ ค์›€์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์ด์— ๋Œ€ํ•œ ํ•œ ๊ฐ€์ง€ ์ ‘๊ทผ ๋ฐฉ์‹์€ Stainless๊ฐ€ ์ œ๊ณตํ•˜๋Š” โ€˜๋™์  ๋ชจ๋“œ(dynamic mode)โ€˜์ž…๋‹ˆ๋‹ค. API์˜ ํฌ๊ธฐ์™€ ์ƒ๊ด€์—†์ด ์„ธ ๊ฐ€์ง€ ๊ธฐ๋ณธ ๋„๊ตฌ๋งŒ ๋…ธ์ถœํ•ฉ๋‹ˆ๋‹ค.

  1. list endpoints: ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ์—”๋“œํฌ์ธํŠธ ๋ชฉ๋ก์„ ์กฐํšŒํ•ฉ๋‹ˆ๋‹ค.
  2. get endpoint: ํŠน์ • ์—”๋“œํฌ์ธํŠธ์— ๋Œ€ํ•œ ์ •๋ณด๋ฅผ ํ•™์Šตํ•ฉ๋‹ˆ๋‹ค.
  3. execute endpoint: ์—”๋“œํฌ์ธํŠธ๋ฅผ ์‹คํ–‰ํ•ฉ๋‹ˆ๋‹ค.

์ด ๋ฐฉ์‹์€ ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ ๋ฌธ์ œ๋ฅผ ํšจ๊ณผ์ ์œผ๋กœ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์ง€๋งŒ, ํ•˜๋‚˜์˜ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•ด ๋ชจ๋ธ์ด ์„ธ ๋ฒˆ์˜ โ€˜ํ„ด(turn)โ€˜์„ ๊ฑฐ์ณ์•ผ ํ•˜๋ฏ€๋กœ ์†๋„๊ฐ€ ๋А๋ ค์ง€๊ณ  ๋น„์šฉ์ด ๋” ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ๋„๊ตฌ๋“ค์ด ์ง์ ‘ ๋กœ๋“œ๋˜๋Š” ๋ฐฉ์‹์ด ์•„๋‹ˆ๋ฏ€๋กœ ์„ฑ๋Šฅ ์ €ํ•˜๊ฐ€ ๋ฐœ์ƒํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.


์•Œ๋ ‰์Šค ๋žซ๋ ˆ์ด์˜ AI ํ™œ์šฉ๋ฒ•: ๊ฐœ์ธ์ ์ธ MCP ๊ฒฝํ—˜

์•Œ๋ ‰์Šค๋Š” Stainless์˜ MCP ์„œ๋ฒ„๋ฅผ ์ฝ”๋”ฉ ์ž‘์—…๋ณด๋‹ค๋Š” ๋น„์ฆˆ๋‹ˆ์Šค ์ธก๋ฉด์—์„œ ํ™œ์šฉํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ๋งํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Š” Notion, HubSpot, Gong ๋“ฑ์˜ MCP ์„œ๋ฒ„์™€ ์ฝ๊ธฐ ์ „์šฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค MCP ์„œ๋ฒ„๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์งˆ๋ฌธ์„ ๋˜์ง‘๋‹ˆ๋‹ค.

โ€œ์ง€๋‚œ์ฃผ Stainless์— ๊ฐ€์ž…ํ•œ ํฅ๋ฏธ๋กœ์šด ๊ณ ๊ฐ์€ ๋ˆ„๊ตฌ์ธ๊ฐ€?โ€

AI๋Š” PostgreSQL ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ์ฟผ๋ฆฌํ•˜๊ณ , HubSpot์—์„œ ๊ต์ฐจ ์ฐธ์กฐํ•˜๋ฉฐ, Notion์—์„œ ๋…ธํŠธ๋ฅผ ์ฐพ๊ณ , Gong์—์„œ ๋…น์ทจ๋ก์„ ํ™•์ธํ•˜์—ฌ ์ด์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์ •๋ณด๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์•„์ง์€ โ€˜์‹คํ—˜ ๋ชจ๋“œโ€™์— ๊ฐ€๊น๊ณ  ๊ฐ€๋” ์—ฐ๊ฒฐ์ด ๋Š๊ธฐ๋Š” ๋“ฑ์˜ โ€˜์‚ฌ์†Œํ•œ ๋ถˆํŽธํ•จ(paper cuts)โ€˜์ด ์žˆ์ง€๋งŒ, ๊ทธ ์ž ์žฌ๋ ฅ์€ ์—„์ฒญ๋‚˜๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค.

์•Œ๋ ‰์Šค๋Š” ํŠนํžˆ AI๋ฅผ ํ™œ์šฉํ•˜์—ฌ โ€˜AI๋ฅผ ์œ„ํ•œ ๋…ธํŠธ ์ˆ˜์ง‘โ€™ ๊ด€ํ–‰์„ ๋งค์šฐ ์œ ์šฉํ•˜๋‹ค๊ณ  ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Š” ํŠน๋ณ„ํ•œ Git ๋ ˆํฌ์ง€ํ† ๋ฆฌ(repository)์— โ€˜๋…ธํŠธ ํด๋”โ€™๋‚˜ โ€˜์—ฐ๊ตฌ ํด๋”โ€™๋ฅผ ๋งŒ๋“ค์–ด, AI์—๊ฒŒ ํฅ๋ฏธ๋กœ์šด ๊ณ ๊ฐ ์ธ์šฉ๋ฌธ์ด ๋ฐœ๊ฒฌ๋˜๋ฉด ์ „์ฒด ์ถœ์ฒ˜์™€ ํ•จ๊ป˜ ์ด ํด๋”์— ์ €์žฅํ•˜๋„๋ก ์ง€์‹œํ•ฉ๋‹ˆ๋‹ค. ์ด๋ ‡๊ฒŒ ํ•˜๋ฉด ๋‹ค์Œ๋ฒˆ์— ์งˆ๋ฌธํ•  ๋•Œ๋งˆ๋‹ค MCP ์„œ๋ฒ„๋ฅผ ๋‹ค์‹œ ๊ฒ€์ƒ‰ํ•  ํ•„์š” ์—†์ด AI๊ฐ€ ์บ์‹œ๋œ ์ •๋ณด๋ฅผ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. Claude Code์™€ ๊ฐ™์€ ๋„๊ตฌ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ด๋Ÿฌํ•œ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๋ฉฐ, ํ˜„์žฌ๋Š” ๊ตฌ์กฐ๊ฐ€ ๋‹ค์†Œ ํ˜ผ๋ž€์Šค๋Ÿฝ๋”๋ผ๋„ AI๊ฐ€ ๋น„์ •ํ˜• ๋ฐ์ดํ„ฐ๋ฅผ ์ž˜ ์ฒ˜๋ฆฌํ•˜๊ธฐ ๋•Œ๋ฌธ์— ํฐ ๋ฌธ์ œ๋Š” ์—†๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋Š” ์ด ๋ ˆํฌ์ง€ํ† ๋ฆฌ์— ๊ณ ๊ฐ ์ธ์šฉ๋ฌธ ์™ธ์—๋„ SQL ์ฟผ๋ฆฌ(SQL queries)๋ฅผ ์ €์žฅํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ด์‚ฌํšŒ(board) ์ค€๋น„๋ฅผ ์œ„ํ•ด โ€œ์ง€๋‚œ 3๊ฐœ์›”๊ฐ„ XYZ ์ง€ํ‘œ์˜ ์›”๋ณ„ ์„ฑ์žฅ๋ฅ ์€ ์–ด๋• ๋‚˜์š”?โ€๋ผ๊ณ  ์งˆ๋ฌธํ•˜๊ณ , AI์™€ ๋ฐ˜๋ณต์ ์ธ ๋Œ€ํ™”๋ฅผ ํ†ตํ•ด ํŠน์ • ๋น„์ฆˆ๋‹ˆ์Šค ๋งฅ๋ฝ์„ ๋ฐ˜์˜ํ•œ SQL ์ฟผ๋ฆฌ๋ฅผ ์™„์„ฑํ•œ ํ›„, ์ด๋ฅผ โ€˜๋ถ„์„ ํด๋”โ€™์— ์ €์žฅํ•˜์—ฌ ๋‹ค์Œ์— ์žฌ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.

๋˜ํ•œ, Stainless ํŒ€์€ ๊ณ ๊ฐ ์ง€์›์—๋„ AI๋ฅผ ํ™œ์šฉํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ณ ๊ฐ ๋ฌธ์˜๊ฐ€ ์ ‘์ˆ˜๋˜๋ฉด, AI(Claude Code)์—๊ฒŒ ๋จผ์ € ํ•ด๊ฒฐ์„ ์‹œ๋„ํ•˜๋„๋ก ํ•ฉ๋‹ˆ๋‹ค. 100% ๋˜๋Š” 50%์˜ ์„ฑ๊ณต๋ฅ ์€ ์•„๋‹ˆ์ง€๋งŒ, ์ „๋ฐ˜์ ์ธ ํšจ์œจ์„ฑ์„ ๊ฐœ์„ ํ•  ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ๋‹ค๊ณ  ๋ณด๊ณ  ์‹คํ—˜์„ ์ง„ํ–‰ ์ค‘์ž…๋‹ˆ๋‹ค.


AI์˜ ๋ฏธ๋ž˜: ์‚ฌ์ด๋ณด๊ทธ์™€ ์ฝ”๋“œ ์‹คํ–‰

์•Œ๋ ‰์Šค๋Š” ์—์ด์ „ํŠธ AI๋ฅผ ๋„˜์–ด AI์˜ ๋ฏธ๋ž˜์— ๋Œ€ํ•œ ํฐ ๋น„์ „์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Š” โ€œAI์˜ ๋ฏธ๋ž˜๋Š” ์‚ฌ์ด๋ณด๊ทธ(cyborg)๊ฐ€ ๋  ๊ฒƒโ€์ด๋ผ๊ณ  ๋งํ•ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ โ€˜์‚ฌ์ด๋ณด๊ทธโ€™๋ž€ ๋ถ€๋ถ„์ ์œผ๋กœ๋Š” GPT์™€ ๊ฐ™์€ ์‹ ๊ฒฝ๋ง LLM AI์ด๊ณ , ๋ถ€๋ถ„์ ์œผ๋กœ๋Š” ์ „ํ†ต์ ์ธ CPU ๊ธฐ๋ฐ˜ ์†Œํ”„ํŠธ์›จ์–ด์ธ ๊ธฐ๊ณ„๋ฅผ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์‚ฌ์ด๋ณด๊ทธ AI๋Š” ํฌ๊ฒŒ ๋‘ ๊ฐ€์ง€ ๋ฐฉ์‹์œผ๋กœ ์ „๊ฐœ๋  ๊ฒƒ์ด๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ํ•˜๋‚˜๋Š” ๋ฐฉ๊ธˆ ์–ธ๊ธ‰ํ•œ ์ผํšŒ์„ฑ ์šด์˜ ์‚ฌ์šฉ ์‚ฌ๋ก€์ด๊ณ , ๋‹ค๋ฅธ ํ•˜๋‚˜๋Š” ํ”„๋กœ๋•์…˜ ์†Œํ”„ํŠธ์›จ์–ด(production software)์ž…๋‹ˆ๋‹ค.

์ฝ”๋“œ ์‹คํ–‰ ๋„๊ตฌ(Code Execution Tool)์˜ ๋ถ€์ƒ

์•Œ๋ ‰์Šค๋Š” ๋ชจ๋ธ์ด ์ˆ˜๋งŽ์€ ๋„๊ตฌ๋ฅผ ๊ฐ€์ง€๋Š” ๋Œ€์‹ , ๋‹จ ๋‘ ๊ฐ€์ง€ ๋„๊ตฌ๋งŒ ๊ฐ€์ง€๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ๋‚˜์•„๊ฐˆ ๊ฒƒ์ด๋ผ๊ณ  ์˜ˆ์ธกํ•ฉ๋‹ˆ๋‹ค.

  1. ์ฝ”๋“œ ์‹คํ–‰ ๋„๊ตฌ (Execute Code Tool): ๋ชจ๋ธ์€ ํ…์ŠคํŠธ ์ƒ์ž์— ํƒ€์ž…์Šคํฌ๋ฆฝํŠธ(TypeScript)์™€ ๊ฐ™์€ ์ฝ”๋“œ๋ฅผ ์ž‘์„ฑํ•˜๊ณ , ์ด ์ฝ”๋“œ๋Š” API์˜ ํƒ€์ž…์Šคํฌ๋ฆฝํŠธ SDK๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Stripe.transactions.list ๋˜๋Š” Stripe.refunds.create์™€ ๊ฐ™์€ API ํ˜ธ์ถœ์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ์€ ์ฝ”๋“œ ์ž‘์„ฑ์— ๋งค์šฐ ๋Šฅ์ˆ™ํ•˜๋ฉฐ, SDK์™€ API๊ฐ€ ์ž˜ ๊ตฌ์„ฑ๋˜์–ด ์žˆ๋‹ค๋ฉด ํŒจํ„ด์„ ํ†ตํ•ด ์‰ฝ๊ฒŒ ํ™•์žฅํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  2. ๋ฌธ์„œ ๊ฒ€์ƒ‰ ๋„๊ตฌ (Search Docs Tool): ๋ชจ๋ธ์ด ํ™•์‹คํ•˜์ง€ ์•Š๊ฑฐ๋‚˜ ์ฒซ ์‹œ๋„์—์„œ ์˜ค๋ฅ˜๊ฐ€ ๋ฐœ์ƒํ•˜๋ฉด, ์ด ๋„๊ตฌ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์„œ๋ฅผ ๊ฒ€์ƒ‰ํ•˜๊ณ  ์งˆ๋ฌธํ•ฉ๋‹ˆ๋‹ค.

์ด๋Ÿฌํ•œ ๋ฐฉ์‹์€ ์•ž์„œ ์–ธ๊ธ‰๋œ ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ ๋ฌธ์ œ๋ฅผ ๊ทน์ ์œผ๋กœ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค. ์ดˆ๊ธฐ ์ปจํ…์ŠคํŠธ ์˜ํ–ฅ์€ 1,000 ํ† ํฐ ๋ฏธ๋งŒ์œผ๋กœ ๋งค์šฐ ์ œํ•œ์ ์ด๋ฉฐ, ํŽ˜์ด์ง€๊ฐ€ ๋งค๊ฒจ์ง„(paginated) ๋ชฉ๋ก ์š”์ฒญ๊ณผ ๊ฐ™์€ ์ž‘์—…์˜ ์ปจํ…์ŠคํŠธ ์˜ํ–ฅ์€ ๊ฑฐ์˜ 0์— ๊ฐ€๊น์Šต๋‹ˆ๋‹ค. ๋ชจ๋ธ์€ ์—ฌ๋Ÿฌ ๋ฐ˜๋ณต ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•œ ํ›„ ์ตœ์ข…์ ์œผ๋กœ ํ•„์š”ํ•œ ์ •๋ณด(์˜ˆ: ๊ณ ๊ฐ ID, ๊ฑฐ๋ž˜ ID, ํ™˜๋ถˆ ID)๋งŒ ์ฝ˜์†”์— ๊ธฐ๋ก(console.log)ํ•  ๊ฒƒ์ด๊ณ , ์ด๋กœ ์ธํ•ด ๋Œ์•„์˜ค๋Š” ์ปจํ…์ŠคํŠธ๋Š” ๋‹จ 10์ค„ ์ •๋„์˜ ํ…์ŠคํŠธ๋กœ ๋งค์šฐ ์ตœ์†Œํ™”๋ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด๋Ÿฌํ•œ ์ฝ”๋“œ๋Š” ๋ชจ๋ธ๊ณผ์˜ ์™•๋ณต(round trip) ์—†์ด CPU ๊ธฐ๋ฐ˜ ์„œ๋ฒ„์—์„œ Stripe API ์˜†์—์„œ ์ดˆ๊ณ ์†์œผ๋กœ ์‹คํ–‰๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์™œ ์ง์ ‘ ์ฝ”๋“œ๋ฅผ ์‹คํ–‰ํ•˜์ง€ ์•Š๋Š”๊ฐ€?

์ผ๊ฐ์—์„œ๋Š” ๋ชจ๋ธ์ด ์ฝ”๋“œ๋ฅผ ์ž‘์„ฑํ•˜๋ฉด ์‚ฌ์šฉ์ž๊ฐ€ ์ง์ ‘ ์‹คํ–‰ํ•˜๋ฉด ๋˜๋Š” ๊ฒƒ ์•„๋‹ˆ๋ƒ๋Š” ์˜๋ฌธ์„ ์ œ๊ธฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์•Œ๋ ‰์Šค๋Š” ์ด์— ๋Œ€ํ•ด ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์„ค๋ช…ํ•ฉ๋‹ˆ๋‹ค.

  • ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ๋ฌธ์ œ: LLM์€ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ์™€ ์ž‘์—…ํ•˜๊ณ , ํŠน์ • ๋ฒ„์ „์˜ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋ฅผ ์ •ํ™•ํžˆ ์‚ฌ์šฉํ•˜๋ฉฐ, API์˜ ํ™˜๊ฐ(hallucination)์„ ํ”ผํ•˜๊ณ , ์˜ค๋ฅ˜ ๋ฐœ์ƒ ์‹œ ๋ฐ˜๋ณต์ ์œผ๋กœ ์ˆ˜์ •ํ•˜๋Š” ๋ฐ ์–ด๋ ค์›€์„ ๊ฒช์Šต๋‹ˆ๋‹ค. npm์ด๋‚˜ ํŒŒ์ด์ฌ ํŒจํ‚ค์ง€ ์ธ๋ฑ์Šค์˜ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋ฅผ ์™„๋ฒฝํ•˜๊ฒŒ ์‚ฌ์šฉํ•  ์ˆ˜ ์—†๋‹ค๋ฉด, ๊ฒฐ๊ตญ ์›์‹œ HTTP API๋ฅผ ์ง์ ‘ ํ˜ธ์ถœํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
  • OpenAPI ์ŠคํŽ™์˜ ๋ฐฉ๋Œ€ํ•จ: ์›์‹œ HTTP API๋ฅผ ์‚ฌ์šฉํ•˜๋ ค๋ฉด ๋ฐฉ๋Œ€ํ•œ OpenAPI ์ŠคํŽ™ ๋ฌธ์„œ๋ฅผ ์ดํ•ดํ•ด์•ผ ํ•˜๋Š”๋ฐ, ์ด๋Š” ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ ๋ฌธ์ œ๋ฅผ ๋‹ค์‹œ ์•ผ๊ธฐํ•ฉ๋‹ˆ๋‹ค.
  • ํƒ€์ž… ๊ฒ€์‚ฌ(Type Checking) ๋ฐ ์˜ค๋ฅ˜ ์ฒ˜๋ฆฌ: ์ฝ”๋“œ ์‹คํ–‰ ๋„๊ตฌ๋Š” ํƒ€์ž… ์ฒด์ปค(type checker)๋ฅผ ์‹คํ–‰ํ•˜์—ฌ โ€œstripe.transactions.list๋Š” ์กด์žฌํ•˜์ง€ ์•Š๋Š”๋‹ค. payment intents, orders, balance transactions ์ค‘ ์–ด๋А ๊ฒƒ์„ ์›ํ•˜๋Š”๊ฐ€?โ€์™€ ๊ฐ™์ด ๋ชจ๋ธ์˜ ์˜ค๋ฅ˜๋ฅผ ์ฆ‰์‹œ ๊ฐ์ง€ํ•˜๊ณ , ๊ด€๋ จ ๋ฌธ์„œ์™€ AI ๊ธฐ๋ฐ˜ ์ œ์•ˆ์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ธฐ๋Šฅ์€ API ์ œ๊ณต์ž๊ฐ€ ์ž˜ ๊ตฌ์ถ•ํ•ด์•ผ ํ•˜๋ฉฐ, ์ด๋Š” ์ „์ฒด ๋Œ€ํ™”์˜ ์ปจํ…์ŠคํŠธ์— ์–ฝ๋งค์ด์ง€ ์•Š๋Š” ์„œ๋ธŒ ์—์ด์ „ํŠธ(sub agent)์— ์˜ํ•ด ์ˆ˜ํ–‰๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋ณด์•ˆ ๋ชจ๋ธ: API ๋ ˆ์ด์–ด์—์„œ ํ•ด๊ฒฐ

๋ณด์•ˆ ๋ฌธ์ œ์— ๋Œ€ํ•ด์„œ๋Š” ๊ถ๊ทน์ ์œผ๋กœ API ๋ ˆ์ด์–ด ์ž์ฒด์—์„œ ํ•ด๊ฒฐ๋˜์–ด์•ผ ํ•œ๋‹ค๊ณ  ์•Œ๋ ‰์Šค๋Š” ์ฃผ์žฅํ•ฉ๋‹ˆ๋‹ค. ํ˜„์žฌ๋Š” MCP๋ฅผ ํ†ตํ•ด ๋…ธ์ถœ๋˜๋Š” ๊ธฐ๋Šฅ์„ ์ œํ•œํ•จ์œผ๋กœ์จ ๋ณด์•ˆ์„ ๊ตฌํ˜„ํ•˜๋ ค ํ•˜์ง€๋งŒ, ๊ฒฐ๊ตญ ๋‚ด๋ถ€์ ์œผ๋กœ๋Š” API์˜ ๋ชจ๋“  ๊ธฐ๋Šฅ์„ ํ˜ธ์ถœํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ฌ๋ฐ”๋ฅธ ์ ‘๊ทผ ๋ฐฉ์‹์€ granular permission(์„ธ๋ถ„ํ™”๋œ ๊ถŒํ•œ)๊ณผ ์ ์ ˆํ•œ ์Šค์ฝ”ํ”„(scope)๋ฅผ ๊ฐ€์ง„ OAuth(์˜ค์–ด์“ฐ)๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๊ตฌ์ถ•ํ•˜๊ธฐ ์–ด๋ ต์ง€๋งŒ, ๋ณด์•ˆ์ด ์ œ๋Œ€๋กœ ์ž‘๋™ํ•ด์•ผ ํ•˜๋Š” ์˜ฌ๋ฐ”๋ฅธ ๊ณ„์ธต์ž…๋‹ˆ๋‹ค.


๋ธŒ๋ ˆ์ธ ํŠธ๋Ÿฌ์ŠคํŠธ(Brain Trust): AI ์˜ต์ €๋ฒ„๋นŒ๋ฆฌํ‹ฐ ํ”Œ๋žซํผ

๋งŽ์€ ๊ธฐ์—…๋“ค์ด AI๋ฅผ ํ”„๋กœ๋•์…˜ ํ™˜๊ฒฝ์— ๋ฐฐํฌํ•˜๋ฉฐ ์ƒ์‚ฐ์„ฑ์„ ๋†’์ด๊ณ  ์žˆ์ง€๋งŒ, ๋™์‹œ์— ๋ถˆ์•ˆ๊ฐ๋„ ์ปค์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํ”„๋กฌํ”„ํŠธ(prompt)๋ฅผ ์ˆ˜์ •ํ•˜๊ณ , ๋ชจ๋ธ์„ ๊ต์ฒดํ•˜๊ณ , ํŒŒ๋ผ๋ฏธํ„ฐ(parameter)๋ฅผ ์กฐ์ •ํ•˜๋ฉด ํ…Œ์ŠคํŠธ์—์„œ๋Š” ๋ฌธ์ œ๊ฐ€ ์—†์–ด ๋ณด์ด์ง€๋งŒ, ๋ฉฐ์น  ํ›„ ๊ณ ๊ฐ ์ง€์› ํ‹ฐ์ผ“์ด ์Ÿ์•„์ง€๊ธฐ ์‹œ์ž‘ํ•ฉ๋‹ˆ๋‹ค. AI๊ฐ€ ์˜ˆ์ƒ์น˜ ๋ชปํ•œ ๋‹ต๋ณ€์„ ์ œ๊ณตํ•˜๊ณ , ์–ธ์ œ ์™œ ์ด๋Ÿฐ ๋ฌธ์ œ๊ฐ€ ๋ฐœ์ƒํ–ˆ๋Š”์ง€ ์•Œ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค.

๋ธŒ๋ ˆ์ธ ํŠธ๋Ÿฌ์ŠคํŠธ(Brain Trust)๋Š” ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” AI ์˜ต์ €๋ฒ„๋นŒ๋ฆฌํ‹ฐ(observability) ํ”Œ๋žซํผ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ํ‰๊ฐ€(eval)์™€ ๊ด€์ฐฐ ๊ฐ€๋Šฅ์„ฑ์„ ํ•˜๋‚˜์˜ ์›Œํฌํ”Œ๋กœ์šฐ(workflow)๋กœ ํ†ตํ•ฉํ•˜์—ฌ, ํ”„๋กœ๋•์…˜ ํ™˜๊ฒฝ์—์„œ ์‹ค์ œ๋กœ ๋ฐœ์ƒํ•œ ์ผ์„ ํ™•์ธํ•˜๊ณ  ๋ณ€๊ฒฝ ์‚ฌํ•ญ์ด ๊ฐœ์„ ๋˜์—ˆ๋Š”์ง€ ์•…ํ™”๋˜์—ˆ๋Š”์ง€ ์ธก์ •ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋•์Šต๋‹ˆ๋‹ค. ํŠธ๋ ˆ์ด์Šค(trace)๋Š” ์ „์ฒด ์‹คํ–‰ ๊ฒฝ๋กœ๋ฅผ ๋ณด์—ฌ์ฃผ๊ณ , ํ‰๊ฐ€๋Š” โ€˜์ข‹์€ ๊ฒฐ๊ณผโ€™๊ฐ€ ๋ฌด์—‡์ธ์ง€ ์ •์˜ํ•˜๋ฉฐ, ์‹คํ—˜์€ ๋ฐฐํฌ ์ „์— ํ”„๋กฌํ”„ํŠธ์™€ ๋ชจ๋ธ์„ ๋น„๊ตํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.

ํ”„๋กœ๋•์…˜ ํŠธ๋ ˆ์ด์Šค๋Š” ํ‰๊ฐ€ ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ์ง์ ‘ ๊ณต๊ธ‰๋˜์–ด ๋ชจ๋“  ์‹คํŒจ๋ฅผ ํ…Œ์ŠคํŠธ ์‚ฌ๋ก€๋กœ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ์‚ฌ์šฉ์ž์—๊ฒŒ ๋„๋‹ฌํ•˜๊ธฐ ์ „์— CI(Continuous Integration) ๋‹จ๊ณ„์—์„œ ํšŒ๊ท€(regression)๋ฅผ ์žก์•„๋‚ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. Notion, Stripe, Zapier, Vercel, RAMP์™€ ๊ฐ™์€ ๊ธฐ์—…๋“ค์ด ๋ธŒ๋ ˆ์ธ ํŠธ๋Ÿฌ์ŠคํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ณ ํ’ˆ์งˆ AI๋ฅผ ๋Œ€๊ทœ๋ชจ๋กœ ๋ฐฐํฌํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ธŒ๋ ˆ์ธ ํŠธ๋Ÿฌ์ŠคํŠธ๋Š” ํŒŒ์ด์ฌ, ํƒ€์ž…์Šคํฌ๋ฆฝํŠธ, ๊ณ , ๋ฃจ๋น„, C์šฉ SDK๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ, ํŠน์ • ํ”„๋ ˆ์ž„์›Œํฌ๋‚˜ ๋ฒค๋”์— ์ข…์†๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. SOCK 2 Type 2 ์ธ์ฆ์„ ๋ฐ›์•˜๊ณ  GDPR ๋ฐ HIPAA๋ฅผ ์ค€์ˆ˜ํ•ฉ๋‹ˆ๋‹ค. ๋” ์ž์„ธํ•œ ์ •๋ณด๋Š” brain.dev์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.


๊ฒฐ๋ก : AI์™€ ์ธ๊ฐ„, ์ฝ”๋“œ์˜ ๊ณต์กด

์•Œ๋ ‰์Šค ๋žซ๋ ˆ์ด์™€์˜ ๋Œ€ํ™”๋Š” API๋ฅผ ํ†ตํ•œ ์ปดํ“จํ„ฐ ๊ฐ„์˜ ์†Œํ†ต์ด ์–ด๋–ป๊ฒŒ AI ์‹œ๋Œ€์— ์ง„ํ™”ํ•˜๊ณ  ์žˆ๋Š”์ง€์— ๋Œ€ํ•œ ๊นŠ์€ ํ†ต์ฐฐ์„ ์ œ๊ณตํ–ˆ์Šต๋‹ˆ๋‹ค. ์ธํ„ฐ๋„ท์˜ ๊ธฐ์กด ์•„ํ‚คํ…์ฒ˜๊ฐ€ AI ์‹œ๋Œ€์— ๋งž์ถฐ ์žฌํŽธ๋˜์–ด์•ผ ํ•˜๋ฉฐ, MCP๊ฐ€ ๊ทธ ํ•ต์‹ฌ์ ์ธ ์—ญํ• ์„ ํ•  ๊ฒƒ์ด๋ผ๋Š” ์ ์€ ๋ถ„๋ช…ํ•ฉ๋‹ˆ๋‹ค.

ํ˜„์žฌ MCP๋Š” ์ปจํ…์ŠคํŠธ ์œˆ๋„์šฐ์˜ ํ•œ๊ณ„์™€ ๋ณต์žกํ•œ ๋„๊ตฌ ์„ค๊ณ„, ๋ณด์•ˆ ๋ฌธ์ œ ๋“ฑ ์—ฌ๋Ÿฌ ๋‚œ๊ด€์— ๋ถ€๋”ชํ˜€ ์žˆ์ง€๋งŒ, ์•Œ๋ ‰์Šค๊ฐ€ ์ œ์‹œํ•˜๋Š” โ€˜์‚ฌ์ด๋ณด๊ทธโ€™ ๋น„์ „๊ณผ โ€˜์ฝ”๋“œ ์‹คํ–‰ ๋„๊ตฌโ€™๋Š” ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•  ํ˜์‹ ์ ์ธ ๋Œ€์•ˆ์ด ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. LLM์ด ์ง์ ‘ ์ฝ”๋“œ๋ฅผ ์ž‘์„ฑํ•˜๊ณ  ์ด๋ฅผ API ์ œ๊ณต์ž๊ฐ€ ์•ˆ์ „ํ•˜๊ณ  ํšจ์œจ์ ์œผ๋กœ ์‹คํ–‰ํ•˜๋Š” ๋ฐฉ์‹์€ AI๊ฐ€ ์ธํ„ฐ๋„ท์˜ ๋ฐฉ๋Œ€ํ•œ ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•˜๋Š” ๋ฐ ์žˆ์–ด ๊ฒŒ์ž„ ์ฒด์ธ์ €๊ฐ€ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

Stainless๋Š” ์ด๋Ÿฌํ•œ ๋ฏธ๋ž˜๋ฅผ ๊ตฌ์ถ•ํ•˜๋Š” ์ตœ์ „์„ ์— ์„œ ์žˆ์Šต๋‹ˆ๋‹ค. AI์™€ ์ธ๊ฐ„, ๊ทธ๋ฆฌ๊ณ  ์ฝ”๋“œ๊ฐ€ ์œ ๊ธฐ์ ์œผ๋กœ ํ˜‘๋ ฅํ•˜๋ฉฐ ์ƒ์ƒํ•  ์ˆ˜ ์—†์—ˆ๋˜ ์ž๋™ํ™”์™€ ํšจ์œจ์„ฑ์„ ์‹คํ˜„ํ•  โ€˜์‚ฌ์ด๋ณด๊ทธโ€™ ์‹œ๋Œ€๋Š” ์ƒ๊ฐ๋ณด๋‹ค ๋” ๋น ๋ฅด๊ฒŒ ๋‹ค๊ฐ€์˜ฌ์ง€ ๋ชจ๋ฆ…๋‹ˆ๋‹ค. ์ด ์—ฌ์ •์—์„œ Stainless์™€ ๊ฐ™์€ ๊ธฐ์—…๋“ค์˜ ์—ญํ• ์€ ๋”์šฑ ์ค‘์š”ํ•ด์งˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค.


โ€œWas anything good released at Google I/O 2026?โ€ โ€” How I AI ๊ธฐ๋ฐ˜ ๊ธฐ์‚ฌ ์›๋ณธ ์˜์ƒ ๋ณด๊ธฐ

Google I/O 2026, ๊ณผ์—ฐ ๊ธฐ๋Œ€๋งŒํผ ํ˜์‹ ์ ์ด์—ˆ๋‚˜? AI ์‹œ๋Œ€์˜ ๋ช…๊ณผ ์•”

์„œ๋ก : AI ํ˜์‹ ์„ ํ–ฅํ•œ ๊ตฌ๊ธ€์˜ ์•ผ์‹ฌ

โ€˜How I AIโ€™ ์ฑ„๋„์˜ ํ”„๋กœ๋•ํŠธ ๋ฆฌ๋”์ด์ž AI ์ „๋ฌธ๊ฐ€์ธ ํด๋ผ๋ผ๋นŒ(Claraveville)์€ ์ด๋ฒˆ Google I/O ํ–‰์‚ฌ๋ฅผ ํ†ตํ•ด ๊ณต๊ฐœ๋œ ๊ตฌ๊ธ€์˜ ์ˆ˜๋งŽ์€ AI ์‹ ์ œํ’ˆ๊ณผ ๊ธฐ๋Šฅ๋“ค์„ ์‹ฌ์ธต์ ์œผ๋กœ ๋ถ„์„ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋งค๋…„ ๊ฐœ๋ฐœ์ž๋“ค์˜ ๊ธฐ๋Œ€๋ฅผ ํ•œ๋ชธ์— ๋ฐ›๋Š” ๊ตฌ๊ธ€์˜ ์ฃผ๋ ฅ ์ด๋ฒคํŠธ์ธ Google I/O๋Š” ์˜ฌํ•ด๋„ ์–ด๊น€์—†์ด ๋‹ค์–‘ํ•œ ๊ธฐ์ˆ ์  ๋ฐœํ‘œ์™€ ํ•จ๊ป˜ ์†Œ๋น„์ž์šฉ ํฌ๋ฆฌ์—์ดํ‹ฐ๋ธŒ ๊ธฐ๋Šฅ์˜ ์ž ์žฌ๋ ฅ์„ ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ๊ณผ์—ฐ ๊ตฌ๊ธ€์ด ์ œ์‹œํ•œ ์ƒˆ๋กœ์šด ๊ธฐ๋Šฅ๋“ค์ด ๋‹จ์ˆœํ•œ ๊ณผ๋Œ€๊ด‘๊ณ ๋ฅผ ๋„˜์–ด ์‹ค์งˆ์ ์ธ ํ˜์‹ ์„ ๊ฐ€์ ธ์˜ฌ ์ˆ˜ ์žˆ์„์ง€, ํด๋ผ๋ผ๋นŒ์˜ ์‹œ์„ ์œผ๋กœ ๊ธฐ์ˆ ์  ์ธก๋ฉด๋ถ€ํ„ฐ ํฅ๋ฏธ๋กœ์šด ์†Œ๋น„์ž ๊ธฐ๋Šฅ๊นŒ์ง€ ํ•จ๊ป˜ ์‚ดํŽด๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค.

์ด๋ฒˆ Google I/O์˜ ํ•ต์‹ฌ์€ ํฌ๊ฒŒ ์„ธ ๊ฐ€์ง€์˜€์Šต๋‹ˆ๋‹ค. ์ฒซ์งธ, ๊ฐœ๋ฐœ์ž๋ฅผ ์œ„ํ•œ ๋” ๋น ๋ฅด๊ณ  ๊ฐ•๋ ฅํ•œ AI ๋ชจ๋ธ๊ณผ ๊ฐœ๋ฐœ ๋„๊ตฌ์˜ ๋ฐœ์ „. ๋‘˜์งธ, ๊ตฌ๊ธ€ ์›Œํฌ์ŠคํŽ˜์ด์Šค(Google Workspace)์™€์˜ ์—ฐ๋™์„ ํ†ตํ•œ ์ผ๋ฐ˜ ์‚ฌ์šฉ์ž์™€ ๊ธฐ์—…์˜ ์ƒ์‚ฐ์„ฑ ํ–ฅ์ƒ. ์…‹์งธ, ์ด๋ฏธ์ง€์™€ ๋น„๋””์˜ค ์ƒ์„ฑ, ๋””์ž์ธ ๋„๊ตฌ ๋“ฑ ์ฐฝ์˜์  ์˜์—ญ์—์„œ์˜ AI ํ™œ์šฉ ํ™•๋Œ€์ž…๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ํด๋ผ๋ผ๋นŒ์€ ์ด๋Ÿฌํ•œ ์•ผ์‹ฌ ์ฐฌ ๋ฐœํ‘œ๋“ค ์†์— ์•„์ง์€ โ€˜๋ฏธ์™„์˜ ์กฐ๊ฐโ€™๋“ค์ด ์กด์žฌํ•˜๋ฉฐ, ๋ช‡๋ช‡ ๊ธฐ๋Šฅ๋“ค์€ ์‹ค์ œ ๊ตฌํ˜„์— ์žˆ์–ด ์•„์‰ฌ์›€์„ ๋‚จ๊ฒผ๋‹ค๊ณ  ์ง€์ ํ•ฉ๋‹ˆ๋‹ค.

1๋ถ€: ๊ฐœ๋ฐœ์ž๋ฅผ ์œ„ํ•œ ์‹ฌ์ธต ๊ธฐ์ˆ  ํ˜์‹  โ€“ ๋ชจ๋ธ ๋ฐ ๊ฐœ๋ฐœ ๋„๊ตฌ

Gemini 3.5: ์†๋„์™€ ์ง€๋Šฅ์˜ ๊ฒฐํ•ฉ

๊ตฌ๊ธ€์€ ์ด๋ฒˆ I/O์—์„œ ์ƒˆ๋กœ์šด ์ œ๋ฏธ๋‹ˆ 3.5(Gemini 3.5) ๋ชจ๋ธ๊ตฐ์„ ๊ณต๊ฐœํ–ˆ์œผ๋ฉฐ, ํŠนํžˆ ๊ฐ€์žฅ ๋น ๋ฅด๊ณ  ์Šค๋งˆํŠธํ•œ ์ฝ”๋”ฉ ๋ชจ๋ธ์ธ ์ œ๋ฏธ๋‹ˆ 3.5 ํ”Œ๋ž˜์‹œ(Gemini 3.5 Flash)๋ฅผ ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ ์ด ๋ชจ๋ธ์ด ๊ฒฝ์Ÿ ๋ชจ๋ธ์ธ ํด๋กœ๋“œ 5.5 ์˜คํ‘ธ์Šค(Claude 5.5 Opus)๋‚˜ GPT-4์™€ ๊ฐ™์€ ์ตœ์ƒ์œ„ ์ฝ”๋”ฉ ๋ชจ๋ธ์˜ ์ง€๋Šฅ์— ํ•„์ ํ•˜๋ฉด์„œ๋„, ์ฒ˜๋ฆฌ ์†๋„๋Š” ๋ฌด๋ ค 4๋ฐฐ๋‚˜ ๋น ๋ฅด๋‹ค๋Š” ์ ์— ์ฃผ๋ชฉํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ๊ธ€์˜ ๋ฒค์น˜๋งˆํฌ์— ๋”ฐ๋ฅด๋ฉด, ์ œ๋ฏธ๋‹ˆ 3.5 ํ”Œ๋ž˜์‹œ๋Š” ์ฝ”๋ฑ์Šค(CodeX) 47๊ณผ ์œ ์‚ฌํ•œ ์Šค๋งˆํŠธํ•จ์„ 3.1 ํ”Œ๋ž˜์‹œ ๋ชจ๋ธ ์ˆ˜์ค€์˜ ์†๋„๋กœ ์ œ๊ณตํ•˜์—ฌ, โ€˜์ดˆ๊ณ ์†-์ดˆ๊ณ ์ง€๋Šฅโ€™์ด๋ผ๋Š” ๋‘ ๋งˆ๋ฆฌ ํ† ๋ผ๋ฅผ ์žก์•˜๋‹ค๋Š” ํ‰๊ฐ€์ž…๋‹ˆ๋‹ค.

ํŠนํžˆ ๊ตฌ๊ธ€์€ ์ œ๋ฏธ๋‹ˆ 3.5 ๋ชจ๋ธ์˜ โ€˜์—์ด์ „ํŠธ(Agent) ๊ธฐ๋Šฅโ€™์— ์ง‘์ค‘ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” AI๊ฐ€ ๋ณต์žกํ•œ ์ž‘์—…์„ ์Šค์Šค๋กœ ๊ณ„ํšํ•˜๊ณ  ์‹คํ–‰ํ•˜๋ฉฐ, ํ•„์š”์— ๋”ฐ๋ผ ๋‹ค๋ฅธ AI๋ฅผ ํ˜ธ์ถœํ•˜๋Š” ๋Šฅ๋ ฅ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ ๊ตฌ๊ธ€์ด AI ์—์ด์ „ํŠธ ๋ถ„์•ผ์— ์ „๋ ฅ์„ ๋‹คํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ด๋Š” ๊ฒฝ์Ÿ์‚ฌ(์•คํŠธ๋กœํ”ฝ(Anthropic), ์˜คํ”ˆAI(OpenAI))์˜ ๋ฐœ์ž์ทจ๋ฅผ ๋”ฐ๋ผ์žก๋Š”(Catch-up) ํ–‰๋ณด๋กœ ๋ณด์ผ ์ˆ˜ ์žˆ์ง€๋งŒ, ํ”Œ๋ž˜์‹œ ๋ชจ๋ธ์˜ ์†๋„์™€ ์†Œ๋น„์ž ์นœํ™”์ ์ธ ์ฐฝ์˜์„ฑ์ด๋ผ๋Š” ๋‘ ๊ฐ€์ง€ ์ฐจ๋ณ„์ ์„ ๊ฐ€์ง€๊ณ  ์žˆ๋‹ค๊ณ  ๋ถ„์„ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ์ œ๋ฏธ๋‹ˆ ๋ชจ๋ธ์˜ ๊ฐ•์ ์ธ โ€˜๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ(Multimodal)โ€™ ๋Šฅ๋ ฅ, ์ฆ‰ ํŒŒ์ผ, ๋น„๋””์˜ค ๋“ฑ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•˜๊ณ  ๋ณ€ํ™˜ํ•˜๋Š” ๋Šฅ๋ ฅ์€ ์—ฌ์ „ํžˆ ๋…๋ณด์ ์ด๋ผ๊ณ  ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค.

Anti-gravity 2.0: ๊ฐœ๋ฐœ ์ƒ์‚ฐ์„ฑ์„ ์œ„ํ•œ ์—์ด์ „ํŠธ IDE

๊ฐœ๋ฐœ์ž๋ฅผ ์œ„ํ•œ ๋„๊ตฌ๋กœ ๊ตฌ๊ธ€์€ ์—์ด์ „ํŠธ ๊ธฐ๋ฐ˜ IDE(ํ†ตํ•ฉ ๊ฐœ๋ฐœ ํ™˜๊ฒฝ)์ธ โ€˜์•ˆํ‹ฐ๊ทธ๋ž˜๋น„ํ‹ฐ(Anti-gravity)โ€˜์˜ 2.0 ๋ฒ„์ „์„ ๊ณต๊ฐœํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ํด๋กœ๋“œ ์ฝ”๋“œ(Claude Code)๋‚˜ ์ฝ”๋ฑ์Šค(CodeX)์™€ ๊ฐ™์€ ๊ฒฝ์Ÿ ๋„๊ตฌ์—์„œ ๋ณผ ์ˆ˜ ์žˆ์—ˆ๋˜ ๊ฐœ๋…๋“ค์„ ๊ตฌ๊ธ€๋งŒ์˜ ๋ฐฉ์‹์œผ๋กœ ๊ตฌํ˜„ํ•œ ๊ฒƒ์œผ๋กœ ํ‰๊ฐ€๋ฉ๋‹ˆ๋‹ค.

์ฃผ์š” ๊ธฐ๋Šฅ:

  • ํ”„๋กœ์ ํŠธ(Projects): ํด๋” ๋‹จ์œ„๋กœ ์ œํ•œ๋œ ์ž‘์—… ํ™˜๊ฒฝ์„ ์ œ๊ณตํ•˜์—ฌ ํšจ์œจ์ ์ธ ์ฝ”๋“œ ๊ด€๋ฆฌ๊ฐ€ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.
  • ์˜ˆ์•ฝ ์ž‘์—…(Scheduled Tasks): ํŠน์ • ํ”„๋กฌํ”„ํŠธ(Prompt)๋ฅผ ์ •ํ•ด์ง„ ์Šค์ผ€์ค„์— ๋”ฐ๋ผ ์ •๊ธฐ์ ์œผ๋กœ ์‹คํ–‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • ์„œ๋ธŒ ์—์ด์ „ํŠธ(Sub-agents): ๋ฉ”์ธ ์—์ด์ „ํŠธ๊ฐ€ ํŠน์ • ์ž‘์—…์„ ์œ„ํ•ด ๋ณด์กฐ ์—์ด์ „ํŠธ๋ฅผ ์ƒ์„ฑํ•˜๊ณ  ํ™œ์šฉํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ํŠนํžˆ ๋ธŒ๋ผ์šฐ์ € ์„œ๋ธŒ ์—์ด์ „ํŠธ๋Š” ์›น ๊ธฐ๋ฐ˜ ์ž‘์—…์— ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค.
  • ํ›…(Hooks): ์—์ด์ „ํŠธ์˜ ์ƒ์•  ์ฃผ๊ธฐ(์˜ˆ: ๋„๊ตฌ ์‹คํ–‰, ์ž‘์—… ์™„๋ฃŒ, ์ƒˆ ์„ธ์…˜ ์‹œ์ž‘)์˜ ํŠน์ • ์ด๋ฒคํŠธ์— ๊ฐœ์ž…ํ•˜์—ฌ ๋งž์ถคํ˜• ๋™์ž‘์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • ๋„ค์ดํ‹ฐ๋ธŒ ๊นƒ(Native Git) ๋ฐ ๋กœ์ปฌ ๊ฐœ๋ฐœ ํ™˜๊ฒฝ: ์ฝ”๋“œ ๋ฒ„์ „ ๊ด€๋ฆฌ ๋ฐ ๋กœ์ปฌ ๊ฐœ๋ฐœ ํ™˜๊ฒฝ๊ณผ์˜ ํ†ตํ•ฉ์„ ๊ฐ•ํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค.
  • ์Šฌ๋ž˜์‹œ ๋ช…๋ น์–ด(Slash Commands):
    • /goal: ์—์ด์ „ํŠธ์—๊ฒŒ ํŠน์ • ๋ชฉํ‘œ๋ฅผ ๋ถ€์—ฌํ•˜๊ณ , ๋ชฉํ‘œ ๋‹ฌ์„ฑ๊นŒ์ง€ ๋ฐ˜๋ณต์ ์œผ๋กœ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.
    • /grill me: ํด๋กœ๋“œ ์ฝ”๋“œ์˜ ์ •์ค‘ํ•œ ์งˆ๋ฌธ๊ณผ ๋‹ฌ๋ฆฌ, ์ด ๋ช…๋ น์–ด๋Š” ์—์ด์ „ํŠธ๊ฐ€ ์‚ฌ์šฉ์ž์˜ ์š”๊ตฌ์‚ฌํ•ญ์„ ๋”์šฑ ์ฒ ์ €ํ•˜๊ฒŒ ๊ฒ€์ฆํ•˜๊ณ  ํ•ต์‹ฌ์„ ํŒŒ๊ณ ๋“œ๋Š” โ€˜๊ฐ•๋„ ๋†’์€ ์งˆ๋ฌธโ€™์„ ๋˜์ง€๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.
    • /schedule, /browser: ์˜ˆ์•ฝ ์ž‘์—… ๋ฐ ๋ธŒ๋ผ์šฐ์ € ๊ธฐ๋ฐ˜ ์ž‘์—…์„ ์œ„ํ•œ ๋ช…๋ น์–ด์ž…๋‹ˆ๋‹ค.

ํด๋ผ๋ผ๋นŒ์€ ์•ˆํ‹ฐ๊ทธ๋ž˜๋น„ํ‹ฐ CLI(Command Line Interface)๋ฅผ ํ†ตํ•ด ๋ธ”๋กœ๊ทธ ์ƒ์„ฑ๊ธฐ API ๊ตฌ์ถ•์„ ์‹œ์—ฐํ–ˆ์Šต๋‹ˆ๋‹ค. ์š”์ฒญ์— ๋”ฐ๋ผ API ์—”๋“œํฌ์ธํŠธ(Endpoint)์™€ ์ธ์ฆ, ์‹œ๊ฐ์  ์›Œํฌํ”Œ๋กœ์šฐ(Workflow) ํŠธ๋ฆฌ๊ฑฐ ๊ธฐ๋Šฅ ๋“ฑ์„ ๋น ๋ฅด๊ฒŒ ์ƒ์„ฑํ•˜๋Š” ๋ชจ์Šต์„ ๋ณด์—ฌ์ฃผ์—ˆ์ง€๋งŒ, ํด๋ผ๋ผ๋นŒ์€ โ€œ๊ธฐ๋Œ€ํ–ˆ๋˜ ๋งŒํผ ํŠน๋ณ„ํžˆ ๋น ๋ฅด๋‹ค๋Š” ๋А๋‚Œ์€ ๋“ค์ง€ ์•Š์•˜๋‹คโ€๊ณ  ์–ธ๊ธ‰ํ•˜๋ฉฐ, โ€œ๊ฒฝ์Ÿ์‚ฌ ๊ธฐ๋Šฅ์„ ๋”ฐ๋ผ์žก๋Š” ์ˆ˜์ค€โ€์ด๋ผ๊ณ  ํ‰๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿผ์—๋„ ์ œ๋ฏธ๋‹ˆ 3.5 ํ”Œ๋ž˜์‹œ ๋ชจ๋ธ์˜ ๋น ๋ฅธ ์†๋„๊ฐ€ ๊ฒฐํ•ฉ๋œ๋‹ค๋ฉด, ๊ฐœ๋ฐœ์ž๋“ค์ด ๊ฐ„๋‹จํ•œ ์ž‘์—…์„ ๋น ๋ฅด๊ฒŒ ์ฒ˜๋ฆฌํ•˜๋Š” ๋ฐ ์œ ์šฉํ•˜๊ฒŒ ํ™œ์šฉ๋  ์ˆ˜ ์žˆ์„ ๊ฒƒ์ด๋ผ๊ณ  ์ „๋งํ–ˆ์Šต๋‹ˆ๋‹ค.

2๋ถ€: ์ผ๋ฐ˜ ์‚ฌ์šฉ์ž ๋ฐ ๊ธฐ์—…์„ ์œ„ํ•œ AI ๊ฒฝํ—˜ ํ™•์žฅ

Google AI Studio: ์›Œํฌ์ŠคํŽ˜์ด์Šค ์—ฐ๋™ ๋ฐ Android ์•ฑ ๊ฐœ๋ฐœ

๊ตฌ๊ธ€์€ ๋กœ์šฐ์ฝ”๋“œ/๋…ธ์ฝ”๋“œ(Low-code/No-code) ๊ฐœ๋ฐœ ํ”Œ๋žซํผ์ธ โ€˜Google AI Studioโ€™์— ์ œ๋ฏธ๋‹ˆ 3.5 ํ”Œ๋ž˜์‹œ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์ƒˆ๋กœ์šด ๊ธฐ๋Šฅ๋“ค์„ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ€์žฅ ์ฃผ๋ชฉํ•  ๋งŒํ•œ ์ ์€ ๊ตฌ๊ธ€ ์›Œํฌ์ŠคํŽ˜์ด์Šค ์•ฑ(Google Workspace Apps)๊ณผ์˜ ํ†ตํ•ฉ์ž…๋‹ˆ๋‹ค. ์ด์ œ ์‚ฌ์šฉ์ž๋Š” ์Šคํ”„๋ ˆ๋“œ์‹œํŠธ(Sheets)๋ฅผ ์ฝ๊ณ , Gmail ์ดˆ์•ˆ์„ ์ž‘์„ฑํ•˜๋ฉฐ, ๋“œ๋ผ์ด๋ธŒ(Drive)๋ฅผ ์ •๋ฆฌํ•˜๊ณ , ์บ˜๋ฆฐ๋”๋ฅผ ํ™•์ธํ•˜๋Š” ๋“ฑ ์›Œํฌ์ŠคํŽ˜์ด์Šค ๋ฐ์ดํ„ฐ์— ์—ฐ๊ฒฐ๋œ AI ์•ฑ์„ ์ง์ ‘ ๊ตฌ์ถ•ํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๊ธฐ์—… ๋‚ด๋ถ€ ์ƒ์‚ฐ์„ฑ ๋„๊ตฌ๋‚˜ ๊ฐœ์ธ ๋น„์„œ ์•ฑ ๋“ฑ ๋‹ค์–‘ํ•œ ํ™œ์šฉ ์‚ฌ๋ก€๋ฅผ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•  ์ž ์žฌ๋ ฅ์„ ๊ฐ€์ง‘๋‹ˆ๋‹ค. ๋˜ํ•œ, Google AI Studio ๋‚ด์—์„œ ์•ˆ๋“œ๋กœ์ด๋“œ ์•ฑ์„ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ๋Šฅ๋„ ์ถ”๊ฐ€๋˜์–ด, ๋ชจ๋ฐ”์ผ ์•ฑ ๊ฐœ๋ฐœ์˜ ์ง„์ž… ์žฅ๋ฒฝ์„ ๋‚ฎ์ถ”์—ˆ์Šต๋‹ˆ๋‹ค.

ํ•˜์ง€๋งŒ ํด๋ผ๋ผ๋นŒ์€ ์บ˜๋ฆฐ๋” ์—ฐ๋™ ์•ฑ์„ ์‹œ๋„ํ•˜๋Š” ๊ณผ์ •์—์„œ ๊ธฐ๋Šฅ์— ๋Œ€ํ•œ ์ ‘๊ทผ์„ฑ ๋ฌธ์ œ๋ฅผ ๊ฒช์—ˆ์Šต๋‹ˆ๋‹ค. ์œ ๋ฃŒ ๊ณ ๊ฐ์ž„์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์›Œํฌ์ŠคํŽ˜์ด์Šค ํ†ตํ•ฉ ๊ธฐ๋Šฅ์„ ์ฐพ๊ฑฐ๋‚˜ ํ™œ์„ฑํ™”ํ•  ์ˆ˜ ์—†์—ˆ์œผ๋ฉฐ, ์ด๋Š” ๊ตฌ๊ธ€์ด ๋ฐœํ‘œํ•œ ๊ธฐ๋Šฅ๋“ค์ด ์•„์ง ๋ชจ๋“  ์‚ฌ์šฉ์ž์—๊ฒŒ ์ฆ‰์‹œ ์ œ๊ณต๋˜์ง€ ์•Š๊ฑฐ๋‚˜, ์‚ฌ์šฉ ๋ฐฉ๋ฒ•์ด ๋ช…ํ™•ํ•˜์ง€ ์•Š๋‹ค๋Š” ์ ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ โ€œ์•„์ง์€ ์•ฝ์†๋œ ๋น„์ „์ผ ๋ฟ, ์‹ค์ œ ๊ตฌํ˜„๊นŒ์ง€๋Š” ์‹œ๊ฐ„์ด ๊ฑธ๋ฆด ๊ฒƒโ€์ด๋ผ๊ณ  ์•„์‰ฌ์›€์„ ํ‘œํ–ˆ์Šต๋‹ˆ๋‹ค.

๋ณต์žกํ•œ ๋ธŒ๋žœ๋“œ๋ช…: ๊ตฌ๊ธ€์˜ ์ˆ™์ œ

ํด๋ผ๋ผ๋นŒ์€ ์ด๋ฒˆ Google I/O ๋ฐœํ‘œ๋ฅผ ํ†ตํ‹€์–ด ๊ฐ€์žฅ ํฐ ๋ถˆ๋งŒ ์ค‘ ํ•˜๋‚˜๋กœ โ€˜๋ณต์žกํ•˜๊ณ  ์ผ๊ด€์„ฑ ์—†๋Š” ์ œํ’ˆ ๋ฐ ๋ธŒ๋žœ๋“œ๋ช…โ€™์„ ๊ผฝ์•˜์Šต๋‹ˆ๋‹ค. ์ œ๋ฏธ๋‹ˆ 3.5 ํ”Œ๋ž˜์‹œ, ์•ˆํ‹ฐ๊ทธ๋ž˜๋น„ํ‹ฐ IDE, ์•ˆํ‹ฐ๊ทธ๋ž˜๋น„ํ‹ฐ CLI, Google AI Studio, ์†Œ๋น„์ž์šฉ ์ œ๋ฏธ๋‹ˆ(Gemini/Google AI), ์˜ด๋‹ˆ(Omni), ํ”Œ๋กœ์šฐ(Flow), ํŒŒ๋ฉœ๋ฆฌ(Pomelli), ์Šคํ‹ฐ์น˜(Stitch) ๋“ฑ ๋„ˆ๋ฌด๋‚˜ ๋งŽ์€ ์ด๋ฆ„๋“ค์ด ํ˜ผ์žฌ๋˜์–ด ์žˆ์–ด ์–ด๋–ค ์ œํ’ˆ์ด ์–ด๋–ค ๊ธฐ๋Šฅ์„ ํ•˜๋Š”์ง€ ํ˜ผ๋ž€์Šค๋Ÿฝ๋‹ค๋Š” ์ง€์ ์ž…๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ ๊ตฌ๊ธ€ ํŒ€์—๊ฒŒ โ€œํฌ๊ด„์ ์ธ ๋ธŒ๋žœ๋“œ ๋ถ„์„๊ณผ ๋‹จ์ผํ™”๋œ ๋ธŒ๋žœ๋“œ ์ „๋žต์ด ์ ˆ์‹คํ•˜๋‹คโ€๊ณ  ๊ฐ•๋ ฅํžˆ ์š”์ฒญํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์‚ฌ์šฉ์ž ๊ฒฝํ—˜๊ณผ ์ œํ’ˆ ์ธ์ง€์— ์žˆ์–ด ์ค‘์š”ํ•œ ๋ถ€๋ถ„์ด๋ฉฐ, ๊ตฌ๊ธ€์ด ํ•ด๊ฒฐํ•ด์•ผ ํ•  ์ˆ™์ œ๋กœ ๋‚จ์•˜์Šต๋‹ˆ๋‹ค.

3๋ถ€: ์ฐฝ์˜์„ฑ์„ ์œ„ํ•œ AI โ€“ ์ด๋ฏธ์ง€, ๋น„๋””์˜ค, ๋””์ž์ธ

Gemini (์†Œ๋น„์ž์šฉ): ์žฌ์„ค๊ณ„์™€ ์ด๋ฏธ์ง€ ์ƒ์„ฑ์˜ ์ง„ํ™”

์†Œ๋น„์ž์šฉ AI ์ œํ’ˆ์ธ ์ œ๋ฏธ๋‹ˆ(Gemini)๋Š” ์ƒˆ๋กœ์šด UI ๋””์ž์ธ๊ณผ ํ•จ๊ป˜ ์‚ฌ์šฉ์ž ๊ฒฝํ—˜์„ ๋Œ€ํญ ๊ฐœ์„ ํ–ˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ์ด๋ฏธ์ง€ ์ƒ์„ฑ ๊ธฐ๋Šฅ์€ ๋‚˜๋…ธ ๋ฐ”๋‚˜๋‚˜(Nano Banana) ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๋‹ค์–‘ํ•œ ํ…œํ”Œ๋ฆฟ๊ณผ ๊ธฐ๋Šฅ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ ์ž์‹ ์˜ ์‚ฌ์ง„์„ ํ™œ์šฉํ•ด โ€˜ํŒŸ์บ์ŠคํŠธ ์ง„ํ–‰์ž์˜ ์ด๋ฏธ์ง€๋ฅผ ์ „๋ฌธ ์ŠคํŠœ๋””์˜ค ๋ฐฐ๊ฒฝ์œผ๋กœ ์—…์Šค์ผ€์ผ ๋ฐ ๋ฏธํ™”โ€™ํ•˜๋Š” ํ…Œ์ŠคํŠธ๋ฅผ ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฒฐ๊ณผ๋ฌผ์€ โ€œ๋‚ด ์–ผ๊ตด์ด ์•„๋‹ˆ๋ฉฐ ๋ชจ๋“  ๋ฉด์—์„œ ๋”์ฐํ•˜๋‹คโ€๊ณ  ํ‰๊ฐ€ํ•  ์ •๋„๋กœ ์–ผ๊ตด ์ธ์‹์—๋Š” ์‹คํŒจํ–ˆ์ง€๋งŒ, ์‚ฌ์ง„๊ณผ ๊ฐ™์€ ์‚ฌ์‹ค์ ์ธ(Photorealistic) ์งˆ๊ฐ๊ณผ ๋น ๋ฅธ ์ƒ์„ฑ ์†๋„๋Š” ์ธ์ƒ์ ์ด์—ˆ๋‹ค๊ณ  ์–ธ๊ธ‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” AI ์ด๋ฏธ์ง€ ์ƒ์„ฑ ๋ชจ๋ธ๋“ค์ด ํ…์ŠคํŠธ ์ƒ์„ฑ ๋Šฅ๋ ฅ๊ณผ ํ•จ๊ป˜ ์ด๋ฏธ์ง€์˜ ์‚ฌ์‹ค์„ฑ์„ ๋†’์ด๋Š” ๋ฐ ์ฃผ๋ ฅํ•˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

Omni & Flow: ๋น„๋””์˜ค ์ƒ์„ฑ ๋ฐ ํŽธ์ง‘์˜ ํ˜์‹ 

์ด๋ฒˆ I/O์—์„œ ๊ฐ€์žฅ ํฅ๋ฏธ๋กœ์šด ๋ฐœํ‘œ ์ค‘ ํ•˜๋‚˜๋Š” ์ƒˆ๋กœ์šด ๋น„๋””์˜ค ์ƒ์„ฑ ๋ชจ๋ธ์ธ โ€˜์˜ด๋‹ˆ(Omni)โ€˜์˜€์Šต๋‹ˆ๋‹ค. ์˜ด๋‹ˆ๋Š” ๊ธฐ์กด ๋ชจ๋ธ๋ณด๋‹ค ๋” ๊ธธ๊ณ , ์ผ๊ด€์ ์ด๋ฉฐ, ํฌํ† ๋ฆฌ์–ผ๋ฆฌ์Šคํ‹ฑํ•œ ๋น„๋””์˜ค๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ฐธ๊ณ  ์ž๋ฃŒ๋ฅผ ํ™œ์šฉํ•ด ๋น„๋””์˜ค๋ฅผ ๋งŒ๋“œ๋Š” ๋Šฅ๋ ฅ๋„ ๊ฐ–์ท„์Šต๋‹ˆ๋‹ค. ๊ตฌ๊ธ€์€ ์˜ด๋‹ˆ๋ฅผ โ€œ๋น„๋””์˜ค๋ฅผ ์œ„ํ•œ ๋‚˜๋…ธ ๋ฐ”๋‚˜๋‚˜โ€๋ผ๊ณ  ์นญํ•˜๋ฉฐ, ์ถ”๋ก  ๋ชจ๋ธ๊ณผ ๋น„๋””์˜ค ์ƒ์„ฑ ๋ชจ๋ธ์„ ๊ฒฐํ•ฉํ•˜์—ฌ โ€˜๋ฌด์—‡์ด๋“  ๋น„๋””์˜ค๋กœ ๋งŒ๋“ค ์ˆ˜ ์žˆ๋‹คโ€™๊ณ  ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค.

ํด๋ผ๋ผ๋นŒ์€ ์ž์‹ ์˜ ์•„์ด๊ฐ€ ๊ทธ๋ฆฐ ๊ทธ๋ฆผ์„ ํ™œ์šฉํ•ด ์Šˆํผํžˆ์–ด๋กœ๊ฐ€ ์•„์ด๋ฅผ ์ˆ˜์—…์—์„œ ํƒˆ์ถœ์‹œ์ผœ ์ฆ๊ฒ๊ฒŒ ๋…ธ๋Š” ๋น„๋””์˜ค๋ฅผ ์ƒ์„ฑํ•˜๋Š” ํ…Œ์ŠคํŠธ๋ฅผ ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ด๋‹ˆ๋Š” ๊ณ„ํš์„ ์„ธ์šฐ๊ณ  ์•ฝ 10์ดˆ ๊ธธ์ด์˜ ๋น„๋””์˜ค๋ฅผ ์„ฑ๊ณต์ ์œผ๋กœ ์ƒ์„ฑํ–ˆ์œผ๋ฉฐ, ํด๋ผ๋ผ๋นŒ์€ โ€œ์•„์ด๊ฐ€ ์ •๋ง ์ข‹์•„ํ•  ๊ฒƒโ€์ด๋ผ๋ฉฐ ๊ฒฐ๊ณผ์— ๋งŒ์กฑํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์†Œ๋ผ(Sora)์™€ ๊ฐ™์€ ๊ธฐ์กด ๋น„๋””์˜ค ๋ชจ๋ธ๋ณด๋‹ค ๋” ๊ธด ๊ธธ์ด์™€ ์ผ๊ด€์„ฑ์„ ๋ณด์—ฌ์ฃผ๋ฉฐ, ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ๊ธฐ๋Šฅ์˜ ๊ฐ•๋ ฅํ•จ์„ ์ž…์ฆํ–ˆ์Šต๋‹ˆ๋‹ค.

์˜ด๋‹ˆ์˜ ์ฃผ์š” ๋น„๋””์˜ค ํŽธ์ง‘ ๊ธฐ๋Šฅ:

  • ๋Œ€ํ™”ํ˜• ๋น„๋””์˜ค ํŽธ์ง‘: ํ”„๋กฌํ”„ํŠธ๋งŒ์œผ๋กœ ๋น„๋””์˜ค์˜ ํŠน์ • ์š”์†Œ๋ฅผ ๋ณ€๊ฒฝ(์˜ˆ: ๊ฑด๋ฌผ์„ ๊ฑฐํ’ˆ์œผ๋กœ ๋ฐ”๊พธ๊ธฐ)ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • ๋น„๋””์˜ค ์„ค๋ช… ๋ฐ ์žฌํŽธ์ง‘: ์˜ด๋‹ˆ๊ฐ€ ๋น„๋””์˜ค๋ฅผ ์„ค๋ช…ํ•˜๊ณ , ์ด ์„ค๋ช…์„ ํŽธ์ง‘ํ•˜์—ฌ ์ƒˆ๋กœ์šด ๋ฒ„์ „์˜ ๋น„๋””์˜ค๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • ์ผ๊ด€์„ฑ ์œ ์ง€: ํ™˜๊ฒฝ, ๊ฐ๋„ ๋“ฑ์„ ๋ณ€๊ฒฝํ•˜๋ฉด์„œ๋„ ์บ๋ฆญํ„ฐ์˜ ์ผ๊ด€์„ฑ์„ ์œ ์ง€ํ•˜์—ฌ ํ”„๋กœ๋•์…˜ ์ˆ˜์ค€์˜ ๋น„๋””์˜ค ์ œ์ž‘์— ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค.

๋”์šฑ ์‹ฌ์ธต์ ์ธ ๋น„๋””์˜ค ํŽธ์ง‘์„ ์œ„ํ•ด ๊ตฌ๊ธ€์€ โ€˜ํ”Œ๋กœ์šฐ(Flow)โ€˜๋ผ๋Š” ์ƒˆ๋กœ์šด ๋„๊ตฌ๋ฅผ ๊ณต๊ฐœํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ด๋‹ˆ ๋ชจ๋ธ์ด ๋‚ด์žฅ๋œ ํ”Œ๋กœ์šฐ๋Š” ์‹œ๋„ค๋งˆํ‹ฑ(Cinematic) ํ’ˆ์งˆ์˜ ํŽธ์ง‘์— ์ค‘์ ์„ ๋‘๋ฉฐ, ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์ฐธ์กฐ ์ž๋ฃŒ๋ฅผ ํ˜ผํ•ฉํ•˜๊ณ  ๋Œ€ํ™”ํ˜•์œผ๋กœ ๋น„๋””์˜ค๋ฅผ ํŽธ์ง‘ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ์‚ฌ์šฉ์ž๊ฐ€ ์บ๋ฆญํ„ฐ๋ฅผ ์ •์˜ํ•˜๊ณ  ํ–ฅํ›„ ๋น„๋””์˜ค์—์„œ ์ด ์บ๋ฆญํ„ฐ๋ฅผ ๊ณ„์† ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ๋Šฅ(์˜ˆ: โ€˜ํƒˆ์ถœ๋Œ€์žฅ ์Šค์ฟจโ€™์ด๋ผ๋Š” ์บ๋ฆญํ„ฐ ์ƒ์„ฑ)๊ณผ ์ž์‹ ์˜ ์•„๋ฐ”ํƒ€๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๊ธฐ๋Šฅ์ด ์ฃผ๋ชฉ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค.

ํ•˜์ง€๋งŒ ํด๋ผ๋ผ๋นŒ์€ ์ž์‹ ์˜ ์•„๋ฐ”ํƒ€๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๊ณผ์ •์—์„œ ๋‹ค์‹œ ํ•œ๋ฒˆ ์ขŒ์ ˆ์„ ๋ง›๋ด์•ผ ํ–ˆ์Šต๋‹ˆ๋‹ค. QR ์ฝ”๋“œ๋ฅผ ์Šค์บ”ํ•˜๊ณ  ์–ผ๊ตด์„ ์ดฌ์˜ํ•˜๋Š” ์ ˆ์ฐจ๋ฅผ ๊ฑฐ์ณค์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์•„๋ฐ”ํƒ€๋Š” ์ƒ์„ฑ๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ โ€œ๋‚ด ์‹ ๋ถ„์„ ๋„์šฉํ•˜์ง€ ๋ง์•„ ๋‹ฌ๋ผโ€๊ณ  ๋†๋‹ดํ•˜๋ฉฐ, โ€œ๋งŽ์€ ๊ฒƒ์„ ๋ฐœํ‘œํ–ˆ์ง€๋งŒ, ์•„์ง ์ž‘๋™ํ•˜์ง€ ์•Š๊ฑฐ๋‚˜ ๊ธฐ๋Šฅ์ด ๋ถˆ์™„์ „ํ•œ ์ ์ด ์‹ค๋ง์Šค๋Ÿฝ๋‹คโ€๊ณ  ์†”์งํ•œ ์‹ฌ์ •์„ ํ† ๋กœํ–ˆ์Šต๋‹ˆ๋‹ค.

Pamelli & Stitch: ๋งˆ์ผ€ํŒ… ๋ฐ ๋””์ž์ธ์˜ AI ํ˜๋ช…

๋งˆ์ผ€ํ„ฐ์™€ ๋””์ž์ด๋„ˆ๋ฅผ ์œ„ํ•œ ๋‘ ๊ฐ€์ง€ ํฅ๋ฏธ๋กœ์šด ๋„๊ตฌ๋„ ์†Œ๊ฐœ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

  • ํŒŒ๋ฉœ๋ฆฌ(Pamelli): ๋ธŒ๋žœ๋“œ ๋ฐ ๋งˆ์ผ€ํŒ… ์ฝ˜ํ…์ธ  ์ƒ์„ฑ ๋„๊ตฌ์ž…๋‹ˆ๋‹ค. ์›น์‚ฌ์ดํŠธ URL์„ ์ž…๋ ฅํ•˜๋ฉด ํ•ด๋‹น ์›น์‚ฌ์ดํŠธ์˜ ๋ธŒ๋žœ๋“œ ์•„์ด๋ดํ‹ฐํ‹ฐ๋ฅผ ๋ถ„์„ํ•˜์—ฌ ๋ธŒ๋žœ๋“œ ๋ถ, ์บ ํŽ˜์ธ ์ž์‚ฐ, ํฌํ† ์ŠˆํŠธ ๋“ฑ์„ ์ƒ์„ฑํ•ด ์ค๋‹ˆ๋‹ค. ๊ตฌ๊ธ€์ด ๋ฐœํ‘œํ•œ design.md ํ‘œ์ค€์„ ํ™œ์šฉํ•˜์—ฌ AI๊ฐ€ ๋””์ž์ธ ์‹œ์Šคํ…œ์„ ์ดํ•ดํ•˜๊ณ  ์ ์šฉํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ ์ž‘๋™ํ•ฉ๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ ์ž์‹ ์˜ ์›น์‚ฌ์ดํŠธ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๋ธŒ๋žœ๋“œ ๋ถ๊ณผ ์›น์‚ฌ์ดํŠธ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๊ณผ์ •์„ ์‹œ์—ฐํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฒฐ๊ณผ๋Š” โ€œ๊ทธ๋ƒฅ ๊ดœ์ฐฎ์€(fine)โ€ ์ˆ˜์ค€์ด์—ˆ์ง€๋งŒ, ๋ธŒ๋žœ๋“œ ์ •์ฒด์„ฑ์„ ์ž๋™์œผ๋กœ ์ƒ์„ฑํ•˜๊ณ  ์ด๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ๋‹ค์–‘ํ•œ ๋งˆ์ผ€ํŒ… ์ž์‚ฐ์„ ๋งŒ๋“œ๋Š” ์ž ์žฌ๋ ฅ์€ ๋†’์ด ํ‰๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.

  • ์Šคํ‹ฐ์น˜(Stitch): ํ”ผ๊ทธ๋งˆ(Figma) ์Šคํƒ€์ผ์˜ ์ธ ๋ธŒ๋ผ์šฐ์ €(in-browser) ๋””์ž์ธ ๋„๊ตฌ์ž…๋‹ˆ๋‹ค. ๊ธฐ์กด ๋””์ž์ธ์„ ํ™œ์šฉํ•˜๊ฑฐ๋‚˜ ์ธ๋ผ์ธ(inline) AI ํŽธ์ง‘์„ ํ†ตํ•ด ๊ตฌ์„ฑ ์š”์†Œ๋ฅผ ํ…์ŠคํŠธ๋กœ ์ˆ˜์ •ํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ๋””์ž์ธ ์บ”๋ฒ„์Šค์— ์‹ค์‹œ๊ฐ„์œผ๋กœ ๋””์ž์ธ์„ ์ŠคํŠธ๋ฆฌ๋ฐํ•˜๋Š” ๊ธฐ๋Šฅ์ด ํŠน์ง•์ž…๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ Google I/O ์‚ฌ์ดํŠธ ์Šคํƒ€์ผ๋กœ ์Šคํ‹ฐ์น˜ ์•ฑ ๋””์ž์ธ์„ ์š”์ฒญํ–ˆ๊ณ , ์—์ด์ „ํŠธ๊ฐ€ ๋””์ž์ธ ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•˜๊ณ  ์ฝ˜ํ…์ธ ๋ฅผ ๊ฐ€์ ธ์™€ ๋ชจ๋ฐ”์ผ ์•ฑ ํ™”๋ฉด์„ ์‹ค์‹œ๊ฐ„์œผ๋กœ ์ŠคํŠธ๋ฆฌ๋ฐํ•˜๋Š” ๋ชจ์Šต์„ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค. ํด๋ผ๋ผ๋นŒ์€ โ€œ๋””์ž์ธ ๋ถ„์•ผ์—์„œ๋Š” ํ”Œ๋ž˜์‹œ ๋ชจ๋ธ์˜ ์†๋„๋ฅผ ์‹ค์ œ๋กœ ์ฒด๊ฐํ•  ์ˆ˜ ์žˆ์—ˆ๋‹คโ€๋ฉฐ, โ€œ๋” ๋งŽ์€ ๋””์ž์ด๋„ˆ๋“ค์ด ์ด ๋„๊ตฌ๋ฅผ ์‚ฌ์šฉํ•ด๋ณด๊ณ  ์‹ถ์–ด ํ•  ๊ฒƒโ€์ด๋ผ๊ณ  ํ˜ธํ‰ํ–ˆ์Šต๋‹ˆ๋‹ค.

๊ฒฐ๋ก : ๊ธฐ๋Œ€์™€ ํ˜„์‹ค ์‚ฌ์ด์˜ Google I/O

์ด๋ฒˆ Google I/O 2026์€ ๊ตฌ๊ธ€์˜ AI์— ๋Œ€ํ•œ ๊นŠ์€ ์•ผ์‹ฌ๊ณผ ๊ธฐ์ˆ ์  ์ง„๋ณด๋ฅผ ๋ช…ํ™•ํžˆ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค. ์ƒˆ๋กœ์šด ์ œ๋ฏธ๋‹ˆ 3.5 ๋ชจ๋ธ๊ตฐ์˜ ์†๋„์™€ ์ง€๋Šฅ, ๊ฐœ๋ฐœ์ž๋ฅผ ์œ„ํ•œ ์•ˆํ‹ฐ๊ทธ๋ž˜๋น„ํ‹ฐ IDE์˜ ๋‹ค์–‘ํ•œ ๊ธฐ๋Šฅ, ์›Œํฌ์ŠคํŽ˜์ด์Šค ์—ฐ๋™์„ ํ†ตํ•œ ์ƒ์‚ฐ์„ฑ ํ–ฅ์ƒ, ๊ทธ๋ฆฌ๊ณ  ์˜ด๋‹ˆ์™€ ํ”Œ๋กœ์šฐ๋ฅผ ํ†ตํ•œ ํ˜์‹ ์ ์ธ ๋น„๋””์˜ค ์ƒ์„ฑ ๋ฐ ํŽธ์ง‘ ๊ธฐ๋Šฅ, ํŒŒ๋ฉœ๋ฆฌ์™€ ์Šคํ‹ฐ์น˜ ๊ฐ™์€ ๋งˆ์ผ€ํŒ…/๋””์ž์ธ ๋„๊ตฌ๊นŒ์ง€, ํฅ๋ฏธ๋กœ์šด ๋ฐœํ‘œ๋“ค์ด ์ค„์„ ์ด์—ˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ๋น„๋””์˜ค ์ƒ์„ฑ ๊ธฐ๋Šฅ๊ณผ ํ”Œ๋ž˜์‹œ ๋ชจ๋ธ ๊ธฐ๋ฐ˜์˜ ์ฝ”๋”ฉ ์†๋„ ํ–ฅ์ƒ์€ ํด๋ผ๋ผ๋นŒ์ด ๊ฐ€์žฅ ๊ธฐ๋Œ€ํ•˜๋Š” ๋ถ€๋ถ„์ด์—ˆ์Šต๋‹ˆ๋‹ค.

ํ•˜์ง€๋งŒ ํด๋ผ๋ผ๋นŒ์€ ์ด๋Ÿฌํ•œ ๋ฐœํ‘œ๋“ค ์†์—์„œ ๋ช‡ ๊ฐ€์ง€ โ€˜๋‚ ์นด๋กœ์šด ๋ชจ์„œ๋ฆฌ(sharp edges)โ€˜๋ฅผ ์ง€์ ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ผ๋ถ€ ํ•ต์‹ฌ ๊ธฐ๋Šฅ๋“ค(์˜ˆ: Google AI Studio์˜ ์›Œํฌ์ŠคํŽ˜์ด์Šค ์—ฐ๋™, Flow์˜ ์•„๋ฐ”ํƒ€ ์ƒ์„ฑ)์€ ์•„์ง ์ œ๋Œ€๋กœ ์ž‘๋™ํ•˜์ง€ ์•Š๊ฑฐ๋‚˜, ์ผ๋ฐ˜ ์‚ฌ์šฉ์ž์—๊ฒŒ๋Š” ์ ‘๊ทผ์„ฑ์ด ๋–จ์–ด์ง€๋Š” ๋ฌธ์ œ๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋„ˆ๋ฌด๋‚˜ ๋งŽ์€ ์ œํ’ˆ๋ช…๊ณผ ๋ธŒ๋žœ๋“œ๊ฐ€ ํ˜ผ์žฌ๋˜์–ด ์‚ฌ์šฉ์ž๋“ค์—๊ฒŒ ํ˜ผ๋ž€์„ ์•ผ๊ธฐํ•œ๋‹ค๋Š” ์ ๋„ ํฐ ๊ณผ์ œ๋กœ ๋‚จ์•˜์Šต๋‹ˆ๋‹ค.

๊ตฌ๊ธ€์˜ AI ํ˜์‹ ์€ ๋ถ„๋ช… ํ˜„์žฌ ์ง„ํ–‰ํ˜•์ด๋ฉฐ, ์—„์ฒญ๋‚œ ์ž ์žฌ๋ ฅ์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ฐœํ‘œ๋œ ๊ธฐ๋Šฅ๋“ค์ด ์‹ค์ œ ์‚ฌ์šฉ์ž๋“ค์—๊ฒŒ ์˜จ์ „ํžˆ ์ „๋‹ฌ๋˜๊ณ  ๋งค๋„๋Ÿฌ์šด ๊ฒฝํ—˜์„ ์ œ๊ณตํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์ œํ’ˆ์˜ ์™„์„ฑ๋„์™€ ์ ‘๊ทผ์„ฑ, ๊ทธ๋ฆฌ๊ณ  ์ผ๊ด€๋œ ๋ธŒ๋žœ๋“œ ์ „๋žต์— ๋Œ€ํ•œ ๊ตฌ๊ธ€์˜ ๋…ธ๋ ฅ์ด ๋”์šฑ ํ•„์š”ํ•  ๊ฒƒ์œผ๋กœ ๋ณด์ž…๋‹ˆ๋‹ค. ์ด๋ฒˆ Google I/O๋Š” AI ์‹œ๋Œ€์˜ ๋ฐ์€ ๋น„์ „์„ ์ œ์‹œํ•จ๊ณผ ๋™์‹œ์—, ๊ทธ ๋น„์ „์„ ํ˜„์‹ค๋กœ ๋งŒ๋“ค๊ธฐ ์œ„ํ•œ ์ง€๋‚œํ•œ ๊ณผ์ • ๋˜ํ•œ ํ•จ๊ป˜ ๋ณด์—ฌ์ค€ ํ–‰์‚ฌ์˜€์Šต๋‹ˆ๋‹ค.


โ€œTrumpโ€™s Taxpayer-Funded Revenge Planโ€ โ€” New York Times Podcasts ๊ธฐ๋ฐ˜ ๊ธฐ์‚ฌ ์›๋ณธ ์˜์ƒ ๋ณด๊ธฐ

ํŠธ๋Ÿผํ”„์˜ โ€˜๋‚ฉ์„ธ์ž ๋ณด๋ณต ๊ธฐ๊ธˆโ€™ ๋…ผ๋ž€: 17์–ต ๋‹ฌ๋Ÿฌ์˜ ์ง„์‹ค๊ณผ ๋ฏธ๊ตญ์˜ ๋ถ„์—ด

์ง€๋‚œ์ฃผ, ์›Œ์‹ฑํ„ด ์ •๊ฐ€๋ฅผ ๋’คํ”๋“  ์ถฉ๊ฒฉ์ ์ธ ์†Œ์‹์ด ์ „ํ•ด์กŒ์Šต๋‹ˆ๋‹ค. ๋„๋„๋“œ ํŠธ๋Ÿผํ”„ ์ „ ๋Œ€ํ†ต๋ น์ด ๊ตญ์„ธ์ฒญ(IRS)์„ ์ƒ๋Œ€๋กœ ์ œ๊ธฐํ–ˆ๋˜ 100์–ต ๋‹ฌ๋Ÿฌ ๊ทœ๋ชจ์˜ ์†Œ์†ก์„ ์ „๊ฒฉ ์ฒ ํšŒํ•˜๋Š” ๋Œ€์‹ , ๋ฒ•๋ฌด๋ถ€(Justice Department)๊ฐ€ 17์–ต 7,600๋งŒ ๋‹ฌ๋Ÿฌ์— ๋‹ฌํ•˜๋Š” ๊ฑฐ๋Œ€ํ•œ ๊ธฐ๊ธˆ์„ ์กฐ์„ฑํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐœํ‘œํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด ๊ธฐ๊ธˆ์€ ํ–‰์ •๋ถ€๊ฐ€ ์ฃผ์žฅํ•˜๋Š” โ€œ๋ฌด๊ธฐํ™”(weaponization) ๋ฐ ๋ฒ•๋ฅ ์ „(lawfare)โ€œ์˜ ํ”ผํ•ด์ž๋“ค์„ ๋ณด์ƒํ•˜๊ธฐ ์œ„ํ•œ ๊ฒƒ์œผ๋กœ, ํŠนํžˆ 1์›” 6์ผ ๋ฏธ๊ตญ ๊ตญํšŒ์˜์‚ฌ๋‹น ๋‚œ์ž… ์‚ฌ๊ฑด์— ์—ฐ๋ฃจ๋œ ์•ฝ 1,600๋ช…์˜ ํ”ผ๊ณ ์ธ๋“ค๊นŒ์ง€ ์ˆ˜ํ˜œ ๋Œ€์ƒ์ด ๋  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์—์„œ ์—„์ฒญ๋‚œ ๋…ผ๋ž€์„ ๋ถˆ๋Ÿฌ์ผ์œผํ‚ค๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

๋‰ด์š•ํƒ€์ž„์Šค ํŒŸ์บ์ŠคํŠธ โ€˜๋” ๋ฐ์ผ๋ฆฌ(The Daily)โ€˜๋Š” ์ด ์ „๋ก€ ์—†๋Š” ๊ธฐ๊ธˆ์˜ ๋ฐฐ๊ฒฝ, ์ž‘๋™ ๋ฐฉ์‹, ๊ทธ๋ฆฌ๊ณ  ์ด๋กœ ์ธํ•ด ์ด‰๋ฐœ๋œ ์ดˆ๋‹น์  ๋ถ„๋…ธ์— ๋Œ€ํ•ด ์‹ฌ์ธต์ ์œผ๋กœ ๋‹ค๋ฃจ์—ˆ์Šต๋‹ˆ๋‹ค. ๋น„ํ‰๊ฐ€๋“ค์€ ์ด ๊ธฐ๊ธˆ์„ โ€œํ”„๋ผ์šฐ๋“œ ๋ณด์ด์Šค(Proud Boys)์™€ ์˜ค์Šค ํ‚คํผ์Šค(Oath Keepers) ๋“ฑ ํŠธ๋Ÿผํ”„์˜ ์ •์น˜์  ์ถฉ์‹ ๋“ค์„ ์œ„ํ•œ 17์–ต ๋‹ฌ๋Ÿฌ ๊ทœ๋ชจ์˜ ์ •์น˜์  ์Œˆ์ง“๋ˆ(political slush fund)โ€œ์ด๋ผ๊ณ  ๋งน๋น„๋‚œํ•˜๋ฉฐ, ํ–‰์ •๋ถ€๊ฐ€ ๋‹ตํ•ด์•ผ ํ•  ์ˆ˜๋งŽ์€ ์งˆ๋ฌธ์ด ์žˆ๋‹ค๊ณ  ์ง€์ ํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ธฐ์‚ฌ๋Š” โ€˜๋” ๋ฐ์ผ๋ฆฌโ€™์˜ ๋ณด๋„๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ, ํŠธ๋Ÿผํ”„์˜ โ€˜๋‚ฉ์„ธ์ž ๋ณด๋ณต ๊ธฐ๊ธˆโ€™์„ ๋‘˜๋Ÿฌ์‹ผ ๋ณต์žกํ•œ ์ง„์‹ค์„ ํŒŒํ—ค์นฉ๋‹ˆ๋‹ค.

๋…ผ๋ž€์˜ ์‹œ์ž‘: ํŠธ๋Ÿผํ”„์˜ IRS ์†Œ์†ก๊ณผ ์„ธ๊ธˆ ์ •๋ณด ์œ ์ถœ

์ด ์ด์•ผ๊ธฐ๋Š” 2016๋…„ ํŠธ๋Ÿผํ”„๊ฐ€ ์ฒ˜์Œ ๋Œ€์„ ์— ์ถœ๋งˆํ–ˆ์„ ๋•Œ๋กœ ๊ฑฐ์Šฌ๋Ÿฌ ์˜ฌ๋ผ๊ฐ‘๋‹ˆ๋‹ค. ๋‹น์‹œ ๊ทธ๋Š” ์—ญ๋Œ€ ๋Œ€์„  ํ›„๋ณด๋“ค๊ณผ ๋‹ฌ๋ฆฌ ์ž์‹ ์˜ ๋‚ฉ์„ธ ์ž๋ฃŒ๋ฅผ ๊ณต๊ฐœํ•˜์ง€ ์•Š์•„ ํฐ ์Šค์บ”๋“ค์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๋Œ€ํ†ต๋ น ํ›„๋ณด๊ฐ€ ์ž์‹ ์˜ ์žฌ์ • ์ƒํƒœ๋ฅผ ํˆฌ๋ช…ํ•˜๊ฒŒ ๊ณต๊ฐœํ•˜๋Š” ๊ฒƒ์€ ์˜ค๋žซ๋™์•ˆ ๊ด€๋ก€์ฒ˜๋Ÿผ ์—ฌ๊ฒจ์กŒ์ง€๋งŒ, ํŠธ๋Ÿผํ”„๋Š” ์ด๋ฅผ ๊ฑฐ๋ถ€ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ์˜ ์ฒซ ์ž„๊ธฐ ๋‚ด๋‚ด ๋‚ฉ์„ธ ์ž๋ฃŒ ๊ณต๊ฐœ ์š”๊ตฌ๋Š” ๊ณ„์†๋˜์—ˆ๊ณ , ๊ฒฐ๊ตญ 2020๋…„ ๋‰ด์š•ํƒ€์ž„์Šค๊ฐ€ ์œ ์ถœ๋œ ์ž๋ฃŒ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ํŠธ๋Ÿผํ”„๊ฐ€ ์ˆ˜๋…„๊ฐ„ ์—ฐ๋ฐฉ ์†Œ๋“์„ธ๋ฅผ ๊ฑฐ์˜ ๋‚ด์ง€ ์•Š์•˜๋‹ค๋Š” ์‚ฌ์‹ค์„ ํญ๋กœํ•˜๋ฉฐ ํฐ ํŒŒ์žฅ์„ ์ผ์œผ์ผฐ์Šต๋‹ˆ๋‹ค.

๋‰ด์š•ํƒ€์ž„์Šค ๋ณด๋„ ์ดํ›„ ์•ฝ 6๊ฐœ์›”์—์„œ 1๋…„ ๋’ค, ๋˜ ๋‹ค๋ฅธ ํƒ์‚ฌ๋ณด๋„ ์ „๋ฌธ ๋งค์ฒด์ธ ํ”„๋กœํผ๋ธ”๋ฆฌ์นด(ProPublica)๋Š” ์ œํ”„ ๋ฒ ์ด์กฐ์Šค(Jeff Bezos), ์ผ๋ก  ๋จธ์Šคํฌ(Elon Musk) ๋“ฑ ์„ธ๊ณ„ ์ตœ๊ณ  ๋ถ€์ž๋“ค์˜ ์„ธ๊ธˆ ์ „๋žต๊ณผ ์ƒํ™ฉ์„ ์ƒ์„ธํžˆ ๋ณด๋„ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ˆ˜๋งŽ์€ ๊ณ ์•ก ์ž์‚ฐ๊ฐ€๋“ค์˜ ์„ธ๊ธˆ ์ •๋ณด๊ฐ€ ๊ณต๊ฐœ๋˜๋ฉด์„œ, ๊ตญ์„ธ์ฒญ(IRS) ๋‚ด๋ถ€ ๊ด€๊ณ„์ž๊ฐ€ ์ด ์ •๋ณด ์œ ์ถœ์˜ ๊ทผ์›์ž„์ด ๋ช…ํ™•ํ•ด์กŒ์Šต๋‹ˆ๋‹ค. ์ฆ‰, IRS ๋‚ด๋ถ€์— ์ •๋ณด ์œ ์ถœ์ž(leaker)๊ฐ€ ์žˆ์—ˆ๋˜ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

2023๋…„ ๊ฐ€์„, ๋งˆ์นจ๋‚ด ์ฐฐ์Šค ๋ฆฌํ‹€์กด(Charles Littlejohn)์ด๋ผ๋Š” ์ „ IRS ๊ณ„์•ฝ์ง ์ง์›์ด ์ฒดํฌ๋˜์–ด ๊ธฐ์†Œ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ํŠธ๋Ÿผํ”„์™€ ๋‹ค๋ฅธ ๋ถ€์œ ์ธต ์ธ์‚ฌ๋“ค์˜ ์„ธ๊ธˆ ์ •๋ณด๋ฅผ ๋‰ด์š•ํƒ€์ž„์Šค์™€ ํ”„๋กœํผ๋ธ”๋ฆฌ์นด์— ์ œ๊ณตํ•œ ํ˜์˜๋ฅผ ์ธ์ •ํ–ˆ์œผ๋ฉฐ, 2024๋…„ ์ดˆ ์ง•์—ญํ˜•์„ ์„ ๊ณ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค. ํƒ€์ธ์˜ ์„ธ๊ธˆ ์ •๋ณด๋ฅผ ์œ ์ถœํ•˜๋Š” ๊ฒƒ์€ ๋ช…๋ฐฑํ•œ ๋ฒ”์ฃ„์ด๋ฉฐ, ํ”ผํ•ด์ž๋Š” IRS๋ฅผ ์ƒ๋Œ€๋กœ ๋ฏผ์‚ฌ ์†Œ์†ก์„ ์ œ๊ธฐํ•˜์—ฌ ์†ํ•ด๋ฐฐ์ƒ์„ ์ฒญ๊ตฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด์— ํŠธ๋Ÿผํ”„ ์ „ ๋Œ€ํ†ต๋ น๊ณผ ๊ทธ์˜ ๋‘ ์•„๋“ค, ๊ทธ๋ฆฌ๊ณ  ๊ฐ€์กฑ ๊ธฐ์—…์€ ์ง€๋‚œ 1์›”, IRS๊ฐ€ ์ฐฐ์Šค ๋ฆฌํ‹€์กด์˜ ์ •๋ณด ์œ ์ถœ์„ ๋ง‰์ง€ ๋ชปํ–ˆ๋‹ค๋Š” ์ด์œ ๋กœ ์ตœ์†Œ 1,000๋งŒ ๋‹ฌ๋Ÿฌ์˜ ์†ํ•ด๋ฐฐ์ƒ์„ ์š”๊ตฌํ•˜๋Š” ์†Œ์†ก์„ ์ œ๊ธฐํ–ˆ์Šต๋‹ˆ๋‹ค.

์ „๋ก€ ์—†๋Š” ๊ฐˆ๋“ฑ: ๋Œ€ํ†ต๋ น์ด ์ •๋ถ€๋ฅผ ๊ณ ์†Œํ•˜๋‹ค

๊ฐœ์ธ์ด ์ž์‹ ์˜ ์„ธ๊ธˆ ์ •๋ณด ์œ ์ถœ์— ๋Œ€ํ•ด IRS๋ฅผ ๊ณ ์†Œํ•˜๋Š” ๊ฒƒ์€ ํ•ฉ๋ฒ•์ ์ธ ๊ถŒ๋ฆฌ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํŠธ๋Ÿผํ”„์˜ ์‚ฌ๋ก€๋Š” ๋งค์šฐ ์ด๋ก€์ ์ด๋ฉฐ, ์ „๋ก€๊ฐ€ ์—†๋‹ค๊ณ  ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ๋‹จ์ˆœํ•œ ์ผ๋ฐ˜ ์‹œ๋ฏผ์ด ์•„๋‹ˆ๋ผ, ์ž์‹ ์ด ์ง์ ‘ ๊ฐ๋…ํ•˜๋Š” ์ •๋ถ€ ๊ธฐ๊ด€์„ ๊ณ ์†Œํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋Œ€ํ†ต๋ น์€ IRS ๊ตญ์žฅ๊ณผ ์žฌ๋ฌด๋ถ€(Treasury Department)์˜ ์ˆ˜์žฅ์„ ์ž„๋ช…ํ•˜๋ฉฐ, ์ด๋“ค ๊ธฐ๊ด€์˜ ์˜์‚ฌ ๊ฒฐ์ •์— ๊นŠ์ด ๊ด€์—ฌํ•ฉ๋‹ˆ๋‹ค. ๋”์šฑ์ด, ๊ทธ๋Š” ์ž์‹ ์˜ ๋‘ ๋ฒˆ์งธ ์ž„๊ธฐ ๋™์•ˆ ๋ฒ•๋ฌด๋ถ€(Justice Department)๋ฅผ ์ž์‹ ์˜ ๋œป๋Œ€๋กœ ์žฌํŽธํ•˜๋Š” ๋ฐ ๋ง‰๋Œ€ํ•œ ์˜ํ–ฅ๋ ฅ์„ ํ–‰์‚ฌํ•ด์™”์Šต๋‹ˆ๋‹ค.

๋ฌธ์ œ๋Š” ๋ฒ•๋ฌด๋ถ€๊ฐ€ ์ด ์†Œ์†ก์—์„œ IRS์˜ ๋ณ€ํ˜ธ์ธ ์—ญํ• ์„ ํ•ด์•ผ ํ•œ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ํŠธ๋Ÿผํ”„์˜ ๋‘ ๋ฒˆ์งธ ์ž„๊ธฐ์— ๋“ค์–ด์„œ๋ฉด์„œ ๋ฒ•๋ฌด๋ถ€๋Š” ์—ญ์‚ฌ์ ์œผ๋กœ ๋…๋ฆฝ์ ์ด์—ˆ๋˜ ์—ญํ• ์„ ๋„˜์–ด, ๋Œ€ํ†ต๋ น์˜ ์ •์น˜์  ์ ๋Œ€์ž๋“ค์„ ๊ธฐ์†Œํ•˜๊ณ  ๋ฏผ์ฃผ๋‹น ์˜์›๋“ค์„ ์—ฐ๋ฐฉ ์ˆ˜์‚ฌ ๋Œ€์ƒ์— ์˜ฌ๋ฆฌ๋Š” ๋“ฑ ์ด๋ก€์ ์ธ ํ–‰๋ณด๋ฅผ ๋ณด์—ฌ์™”์Šต๋‹ˆ๋‹ค. ํ˜„์žฌ ๋ฒ•๋ฌด๋ถ€์˜ ๋ฒ•๋ฌด์žฅ๊ด€ ๋Œ€ํ–‰์ธ ํ† ๋“œ ๋ธ”๋ž‘์‰ฌ(Todd Blanche)๋Š” ๋ฐฑ์•…๊ด€์— ๋ณต๊ท€ํ•˜๊ธฐ ์ „ ํŠธ๋Ÿผํ”„์˜ ๊ฐœ์ธ ๋ณ€ํ˜ธ์‚ฌ์˜€์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ƒํ™ฉ์—์„œ ๋ฒ•๋ฌด๋ถ€๊ฐ€ ๋Œ€ํ†ต๋ น์ด ์ œ๊ธฐํ•œ ์†Œ์†ก์— ๋Œ€ํ•ด ์–ด๋–ป๊ฒŒ ๋ฐฉ์–ดํ• ์ง€ ๊ฒฐ์ •ํ•˜๋Š” ๊ฒƒ์€ ์‹ฌ๊ฐํ•œ ์ดํ•ด ์ƒ์ถฉ(conflict of interest) ๋ฌธ์ œ๋ฅผ ์•ผ๊ธฐํ–ˆ์Šต๋‹ˆ๋‹ค. ํŠธ๋Ÿผํ”„๊ฐ€ ์‚ฌ์ ์œผ๋กœ ๊ณ ์šฉํ•œ ๋ณ€ํ˜ธ์ธ๋‹จ์„ ํ†ตํ•ด ์ •๋ถ€๋ฅผ ๊ณ ์†Œํ•˜๋Š” ๋™์‹œ์—, ์ •๋ถ€ ์ธก ๋ณ€ํ˜ธ์ธ๋‹จ๊นŒ์ง€ ์ด๋ก€์ ์œผ๋กœ ๊ฐ•๋ ฅํ•˜๊ฒŒ ํ†ต์ œํ•˜๋Š” ์ƒํ™ฉ์ด ๋ฒŒ์–ด์ง„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์žฌํŒ๋ถ€๋Š” ์ด๋Ÿฌํ•œ ๋ณต์žกํ•œ ์ดํ•ด ์ƒ์ถฉ ๊ด€๊ณ„๋ฅผ ์šฐ๋ คํ•˜๋ฉฐ ์†Œ์†ก ์ง„ํ–‰์— ์˜๋ฌธ์„ ์ œ๊ธฐํ–ˆ์Šต๋‹ˆ๋‹ค. ์†Œ์†ก ๋‹น์‚ฌ์ž๋“ค์€ ์‹ค์ œ๋กœ ์„œ๋กœ ๋Œ€๋ฆฝํ•ด์•ผ ํ•œ๋‹ค๋Š” ๋ฒ•์  ์›์น™์ด ์žˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ํŒ์‚ฌ๋Š” ์ด ์†Œ์†ก์ด โ€œ์ง„์ •ํ•œ ๋Œ€๋ฆฝ์ด ์—†๋Š” ๊ฐ€์žฅ๊ทน(charade)โ€œ์ผ ์ˆ˜ ์žˆ๋‹ค๊ณ  ํŒ๋‹จํ•˜๊ณ , ๋ฒ•๋ฌด๋ถ€์™€ ํŠธ๋Ÿผํ”„ ์ธก ๋ณ€ํ˜ธ์ธ๋‹จ์—๊ฒŒ ์†Œ์†ก์ด ์‹ค์ œ๋กœ ๋Œ€๋ฆฝํ•˜๋Š” ๊ด€๊ณ„์ธ์ง€, ์•„๋‹ˆ๋ฉด โ€œ๊ฑฐ์ง“๋œ ๊ตฌ์‹ค๋กœ ์ œ๊ธฐ๋œ ์†Œ์†กโ€์ธ์ง€ ์„ค๋ช…ํ•˜๋ผ๋Š” ์„œ๋ฉด ๋‹ต๋ณ€์„ ์š”๊ตฌํ–ˆ์Šต๋‹ˆ๋‹ค. ๋งŒ์•ฝ ์žฌํŒ๋ถ€๊ฐ€ ์ง„์ •ํ•œ ๋Œ€๋ฆฝ์ด ์—†๋‹ค๊ณ  ํŒ๋‹จํ–ˆ๋‹ค๋ฉด, ์†Œ์†ก์„ ๊ธฐ๊ฐํ•  ์ˆ˜๋„ ์žˆ๋Š” ์ƒํ™ฉ์ด์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ƒํ™ฉ์€ ํŠธ๋Ÿผํ”„์—๊ฒŒ ๋ถˆ๋ฆฌํ•˜๊ฒŒ ๋Œ์•„๊ฐ€๊ณ  ์žˆ์—ˆ์œผ๋ฉฐ, ๋ฒ•๋ฌด๋ถ€ ์ž…์žฅ์—์„œ๋„ ๋Œ€ํ†ต๋ น์˜ ์ฃผ์žฅ์— ๋ฐ˜๋Œ€ํ•˜๋Š” ์ž…์žฅ์„ ํ‘œ๋ช…ํ•ด์•ผ ํ•˜๋Š” ๋งค์šฐ ๊ณค๋ž€ํ•œ ์ƒํ™ฉ์ด์—ˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ๋Œ€ํ†ต๋ น์— ๋ฐ˜๋Œ€ํ•˜๋‹ค๊ฐ€ ํ•ด๊ณ ๋œ ์ „๋ก€๊ฐ€ ์žˆ๋Š” ํ˜„ ๋ฒ•๋ฌด๋ถ€์˜ ๋ถ„์œ„๊ธฐ์—์„œ๋Š” ๋”์šฑ ๊ทธ๋žฌ์Šต๋‹ˆ๋‹ค.

โ€˜๋ฌด๊ธฐํ™” ํ”ผํ•ด ๋ณด์ƒ ๊ธฐ๊ธˆโ€™์˜ ํƒ„์ƒ

์ด๋Ÿฌํ•œ ๊ณค๋ž€ํ•œ ์ƒํ™ฉ์„ ํ”ผํ•˜๊ธฐ ์œ„ํ•ด ํ–‰์ •๋ถ€ ๋‚ด๋ถ€์—์„œ๋Š” ์ง€๋‚œ์ฃผ ๋‚ด๋‚ด ํ™œ๋ฐœํ•œ ๋…ผ์˜๊ฐ€ ์ด๋ฃจ์–ด์กŒ์Šต๋‹ˆ๋‹ค. ๊ทธ ๊ฒฐ๊ณผ, ํŠธ๋Ÿผํ”„ ์ „ ๋Œ€ํ†ต๋ น์ด IRS ์†Œ์†ก์„ ์™„์ „ํžˆ ์ฒ ํšŒํ•˜๊ณ , ๊ทธ ๋Œ€๊ฐ€๋กœ ๋ฒ•๋ฌด๋ถ€๊ฐ€ โ€œ๋ฌด๊ธฐํ™”โ€์˜ ํ”ผํ•ด์ž๋“ค์„ ๋ณด์ƒํ•˜๊ธฐ ์œ„ํ•œ ์ „๋ก€ ์—†๋Š” ๊ฑฐ๋Œ€ํ•œ ๋‚ฉ์„ธ์ž ๊ธฐ๊ธˆ์„ ์กฐ์„ฑํ•œ๋‹ค๋Š” ํ•ฉ์˜์— ์ด๋ฅด๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.

์ด ๊ธฐ๊ธˆ์˜ ๊ทœ๋ชจ๋Š” 17์–ต 7,600๋งŒ ๋‹ฌ๋Ÿฌ์ž…๋‹ˆ๋‹ค. โ€˜1776โ€™์ด๋ผ๋Š” ์ˆซ์ž๋Š” ๋ฏธ๊ตญ ๊ฑด๊ตญ ์—ฐ๋„๋ฅผ ์ƒ์ง•ํ•˜๋ฉฐ, ์ด๋Š” ์šฐ์—ฐ์ด ์•„๋‹ ๊ฒƒ์ด๋ผ๋Š” ๋ถ„์„์ด ์ง€๋ฐฐ์ ์ž…๋‹ˆ๋‹ค. ์ด ๋ง‰๋Œ€ํ•œ ์ž๊ธˆ์€ โ€˜ํŒ๊ฒฐ ๊ธฐ๊ธˆ(Judgment Fund)โ€˜์ด๋ผ๋Š” ๊ณ„์ •์—์„œ ์ธ์ถœ๋  ์˜ˆ์ •์ž…๋‹ˆ๋‹ค. ํŒ๊ฒฐ ๊ธฐ๊ธˆ์€ ๋ฒ•๋ฌด๋ถ€๊ฐ€ ๋ฏธ๊ตญ ์ •๋ถ€๋ฅผ ์ƒ๋Œ€๋กœ ์ œ๊ธฐ๋œ ๋ฒ•์  ์†Œ์†ก์„ ํ•ด๊ฒฐํ•˜๋Š” ๋ฐ ์‚ฌ์šฉํ•˜๋Š” ์ž๊ธˆ์œผ๋กœ, ์ •๋ถ€๊ฐ€ ์†Œ์†ก์—์„œ ํ•ฉ์˜ํ•  ๋•Œ ์ด ๊ธฐ๊ธˆ์—์„œ ๋ˆ์„ ์ธ์ถœํ•ฉ๋‹ˆ๋‹ค. ํ‰์†Œ๋ณด๋‹ค ํ›จ์”ฌ ๋งŽ์€ ๊ธˆ์•ก์ด ์ธ์ถœ๋  ์˜ˆ์ •์ด๋ฉฐ, ์ด ๋ˆ์€ ๋ณ„๋„์˜ ๊ณ„์ •์œผ๋กœ ์˜ˆ์น˜๋˜์–ด ๋ฒ•๋ฌด์žฅ๊ด€์ด ์ž„๋ช…ํ•œ 5๋ช…์˜ ์œ„์›์ด ๊ด€๋ฆฌํ•˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค. ์ด๋“ค์€ ์—ฐ๋ฐฉ ์ •๋ถ€์— ์˜ํ•ด ํ”ผํ•ด๋ฅผ ์ž…์—ˆ๋‹ค๊ณ  ์ฃผ์žฅํ•˜๋Š” ์‚ฌ๋žŒ๋“ค์˜ ์ฒญ๊ตฌ๋ฅผ ์‹ฌ์‚ฌํ•˜๊ณ  ๋ณด์ƒ๊ธˆ์„ ์ง€๊ธ‰ํ•˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.

๊ทธ๋Ÿฌ๋‚˜ ์ด ๊ธฐ๊ธˆ์˜ ์ž‘๋™ ๋ฐฉ์‹์— ๋Œ€ํ•ด์„œ๋Š” ์—ฌ์ „ํžˆ ๋งŽ์€ ๊ฒƒ์ด ๋ถˆ๋ถ„๋ช…ํ•ฉ๋‹ˆ๋‹ค. ๋ณด์ƒ๊ธˆ ์ฒญ๊ตฌ๊ฐ€ ์–ด๋–ป๊ฒŒ ํ‰๊ฐ€๋ ์ง€, ๋ฒ•๋ฌด๋ถ€๊ฐ€ ๋งํ•˜๋Š” โ€œ๋ฌด๊ธฐํ™”(weaponization)โ€œ๊ฐ€ ์ •ํ™•ํžˆ ๋ฌด์—‡์„ ์˜๋ฏธํ•˜๋Š”์ง€, ๊ทธ๋ฆฌ๊ณ  ์–ผ๋งˆ๋‚˜ ๋งŽ์€ ๋ˆ์ด ์ง€๊ธ‰๋ ์ง€์— ๋Œ€ํ•œ ๊ตฌ์ฒด์ ์ธ ๊ธฐ์ค€์€ ์•„์ง ๊ณต๊ฐœ๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์‹ฌ์ง€์–ด ์ด ๊ธฐ๊ธˆ์„ ํ†ต์ œํ•˜๊ณ  ๊ฒฐ์ •์„ ๋‚ด๋ฆด 5๋ช…์˜ ์œ„์›์ด ๋ˆ„๊ตฌ์ธ์ง€์กฐ์ฐจ ์•Œ๋ ค์ง€์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์ด๋“ค์ด ๋ง‰๋Œ€ํ•œ ์—ญํ• ์„ ์ˆ˜ํ–‰ํ•  ๊ฒƒ์ด ๋ถ„๋ช…ํ•œ ๋งŒํผ, ์›Œ์‹ฑํ„ด์—์„œ๋Š” ์ด ๊ธฐ๊ธˆ์ด ๊ถ๊ทน์ ์œผ๋กœ ์–ด๋–ค ๋ชจ์Šต์ด ๋ ์ง€์— ๋Œ€ํ•œ ์ˆ˜๋งŽ์€ ์˜๋ฌธ์ด ์ œ๊ธฐ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฒ•๋ฌด๋ถ€๊ฐ€ ๋‹ค๋ฅธ ์†Œ์†ก๋“ค์„ ํ•ฉ์˜ํ•˜๋Š” ๊ฒฝ์šฐ๋Š” ๋งŽ์ง€๋งŒ, ์ด์ฒ˜๋Ÿผ ๋Œ€๊ทœ๋ชจ์˜ ํฌ๊ด„์ ์ธ โ€œํ”ผํ•ด ๋ณด์ƒ ๊ธฐ๊ธˆโ€์„ ์กฐ์„ฑํ•˜๋Š” ๊ฒƒ์€ ์ „๋ก€ ์—†๋Š” ์ผ์ž…๋‹ˆ๋‹ค.

๋ˆ„๊ฐ€ ์ด โ€˜๋ณต์ˆ˜ ๊ธฐ๊ธˆโ€™์˜ ์ˆ˜ํ˜œ์ž๊ฐ€ ๋  ๊ฒƒ์ธ๊ฐ€?

๋ฒ•๋ฌด๋ถ€๋Š” ์ด ๊ธฐ๊ธˆ์˜ ์ˆ˜ํ˜œ ๋Œ€์ƒ์„ โ€œ๋ฌด๊ธฐํ™” ๋ฐ ๋ฒ•๋ฅ ์ „์˜ ํ”ผํ•ด์ž๋“คโ€์ด๋ผ๊ณ ๋งŒ ๋ฐํ˜”์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ด ๋ฐœํ‘œ์— ๋Œ€ํ•ด ๊ฐ€์žฅ ๋จผ์ € ๊ธฐ๋Œ€๋ฅผ ํ‘œ๋ช…ํ•œ ์ง‘๋‹จ ์ค‘ ํ•˜๋‚˜๋Š” ๋ฐ”๋กœ 2021๋…„ 1์›” 6์ผ ๊ตญํšŒ์˜์‚ฌ๋‹น ๋‚œ์ž… ์‚ฌ๊ฑด์œผ๋กœ ์œ ์ฃ„ ํŒ๊ฒฐ์„ ๋ฐ›์•˜๋˜ ์‚ฌ๋žŒ๋“ค, ์ฆ‰ ํŠธ๋Ÿผํ”„ ์ „ ๋Œ€ํ†ต๋ น์ด ์ด์ „์— ์‚ฌ๋ฉดํ–ˆ๋˜ ์‚ฌ๋žŒ๋“ค์ž…๋‹ˆ๋‹ค. ์ด๋“ค์€ ๋ฌผ๋ก , ์˜ค๋ฐ”๋งˆ ํ–‰์ •๋ถ€ ์‹œ์ ˆ ๋ฒ•๋ฌด๋ถ€์™€ ์—ฐ๋ฐฉ์ˆ˜์‚ฌ๊ตญ(FBI)์ด ์ž์‹ ๋“ค์„ ๋ถ€๋‹นํ•˜๊ฒŒ ์ˆ˜์‚ฌํ–ˆ๋‹ค๊ณ  ๋ฏฟ๋Š” ํŠธ๋Ÿผํ”„ ์ง€์ง€์ž๋“ค๊ณผ ์ •์น˜์  ๋™๋งน์ž๋“ค๋„ ์ด ๊ธฐ๊ธˆ์„ ํ†ตํ•ด ๊ณผ๊ฑฐ์˜ ๊ฒฝํ—˜์œผ๋กœ๋ถ€ํ„ฐ โ€œ์™„์ „ํ•œ ๋ณด์ƒโ€์„ ๋ฐ›์„ ๊ธฐํšŒ๋กœ ๋ณด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๋Ÿฌ๋‚˜ ์ด ๊ธฐ๊ธˆ์˜ ์ง€๊ธ‰ ๋ฐฉ์‹์— ๋Œ€ํ•ด์„œ๋„ ์˜๋ฌธ์ด ๋งŽ์Šต๋‹ˆ๋‹ค. ์ง€๊ธ‰์•ก์ด ๋ฐœ์ƒํ•œ ๋ฒ•๋ฅ  ๋น„์šฉ์„ ์ƒํ™˜ํ•˜๋Š” ์šฉ๋„์ธ์ง€, ์•„๋‹ˆ๋ฉด ๊ทธ ์ด์ƒ์˜ ๊ธˆ์•ก์ด ์ง€๊ธ‰๋˜์–ด ์‚ฌ์‹ค์ƒ โ€œ๊ฒฝํ—˜์œผ๋กœ๋ถ€ํ„ฐ ์ด๋“์„ ์ทจํ• โ€ ์ˆ˜ ์žˆ๊ฒŒ ๋˜๋Š” ๊ฒƒ์ธ์ง€ ๋ช…ํ™•ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ˆ„๊ฐ€ ์–ผ๋งˆ์˜ ๋ˆ์„ ๋ฐ›์•˜๋Š”์ง€์— ๋Œ€ํ•œ ๋ช…ํ™•ํ•œ ์˜์ˆ˜์ฆ ๊ณต๊ฐœ๊ฐ€ ์—†์„ ๊ฐ€๋Šฅ์„ฑ๋„ ์ œ๊ธฐ๋˜์–ด, ์ด ๋ชจ๋“  ์šด์˜์ด ๋Œ€์ค‘์˜ ์‹œ์•ผ์—์„œ ์™„์ „ํžˆ ๊ฐ€๋ ค์งˆ ์ˆ˜ ์žˆ๋‹ค๋Š” ์šฐ๋ ค๋„ ๋‚˜์˜ต๋‹ˆ๋‹ค.

๋งŽ์€ ์‚ฌ๋žŒ๋“ค์—๊ฒŒ 1์›” 6์ผ ๊ตญํšŒ์˜์‚ฌ๋‹น์„ ์Šต๊ฒฉํ–ˆ๋˜ ์‚ฌ๋žŒ๋“ค์ด ์ •๋ถ€๋กœ๋ถ€ํ„ฐ ๋ˆ์„ ๋ฐ›๋Š”๋‹ค๋Š” ์ƒ๊ฐ์€ ์ถฉ๊ฒฉ์ ์ผ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์‹ฌํ•˜๊ฒŒ ๋งํ•˜๋ฉด, ์ž์‹ ๋“ค์ด ๊ณต๊ฒฉํ–ˆ๋˜ ์ •๋ถ€๋กœ๋ถ€ํ„ฐ ์œ ์ฃ„ ํŒ๊ฒฐ์„ ๋ฐ›์€ ๋ฒ”์ฃ„์ž๋“ค์ด ๋ˆ์„ ๋ฐ›์„ ์ˆ˜ ์žˆ๋‹ค๋Š” ์‹œ๋‚˜๋ฆฌ์˜ค๊ฐ€ ์ด ๊ธฐ๊ธˆ์˜ ๊ตฌ์กฐ ํ•˜์—์„œ๋Š” ์ถฉ๋ถ„ํžˆ ๊ฐ€๋Šฅํ•ด ๋ณด์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๊ธฐ๊ธˆ ๋ฐœํ‘œ์— ๋Œ€ํ•œ ์—ฌ์•ผ ์–‘์ธก์˜ ๊ฒฉ๋ ฌํ•œ ๋ฐ˜์‘์„ ์„ค๋ช…ํ•˜๋Š” ์ฃผ์š” ์š”์ธ์ด๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค.

์—ฌ์•ผ์˜ ๊ฒฉ๋ ฌํ•œ ๋น„ํŒ๊ณผ ๋‚ด๋ถ€ ๊ท ์—ด

๋ฒ•๋ฌด์žฅ๊ด€ ๋Œ€ํ–‰ ํ† ๋“œ ๋ธ”๋ž‘์‰ฌ๋Š” ํ™”์š”์ผ ๊ตญํšŒ์˜์‚ฌ๋‹น์—์„œ ์—ด๋ฆฐ ๋ฒ•๋ฌด๋ถ€ ์˜ˆ์‚ฐ ์ฒญ๋ฌธํšŒ์— ์ถœ์„ํ•˜์—ฌ ์ด ๊ธฐ๊ธˆ ์กฐ์„ฑ์„ ์˜นํ˜ธํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” โ€œ์ด๊ฒƒ์ด ์ด๋ก€์ ์ธ ๊ฒƒ์€ ์‚ฌ์‹ค์ด์ง€๋งŒ, ์ „๋ก€๊ฐ€ ์—†๋Š” ๊ฒƒ์€ ์•„๋‹ˆ๋‹คโ€๋ผ๊ณ  ์ฃผ์žฅํ•˜๋ฉฐ, ๊ธฐ๊ธˆ์ด ํŠน์ • ์ •์น˜ ์„ฑํ–ฅ์ด๋‚˜ 1์›” 6์ผ ์‚ฌ๊ฑด, ๋˜๋Š” ํŠน์ • ์ธ๋ฌผ์— ๊ตญํ•œ๋˜์ง€ ์•Š๊ณ  โ€œ๋ˆ„๊ตฌ๋‚˜ ์‹ ์ฒญํ•  ์ˆ˜ ์žˆ๋‹คโ€๊ณ  ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๋Ÿฌ๋‚˜ ์ฒญ๋ฌธํšŒ๋Š” ๊ณง ๋ฏผ์ฃผ๋‹น ์˜์›๋“ค์ด ์ด ๊ธฐ๊ธˆ์„ โ€œ์ •์น˜์  ์Œˆ์ง“๋ˆ(slush fund)โ€œ์ด๋ผ๊ณ  ๋งน๋น„๋‚œํ•˜๋ฉฐ ๋ธ”๋ž‘์‰ฌ ๋Œ€ํ–‰์„ ์ง‘์ค‘์ ์œผ๋กœ ์ถ”๊ถํ•˜๋Š” ์žฅ์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ํฌ๋ฆฌ์Šค ๋ฐด ํ™€๋ Œ(Chris Van Hollen) ์ƒ์›์˜์› ๋“ฑ ๋ฏผ์ฃผ๋‹น ์˜์›๋“ค์€ โ€œ์ด๊ฒƒ์€ ํ„ฐ๋ฌด๋‹ˆ์—†๊ณ  ์ „๋ก€ ์—†๋Š” ์Œˆ์ง“๋ˆโ€, โ€œ๊ณต๊ณต ์ž๊ธˆ์˜ ์ˆœ์ˆ˜ํ•œ ์ ˆ๋„โ€, โ€œ์™ธ์„ค์ ์ด๊ณ  ๋ถˆ๋ฒ•์ ์ด๋ฉฐ ๋ถ€ํŒจํ•œ ์…€ํ”„-๋”œ๋ง ๊ณ„ํšโ€์ด๋ผ๊ณ  ๊ฐ•ํ•˜๊ฒŒ ๋น„ํŒํ•˜๋ฉฐ, 1์›” 6์ผ ์˜์‚ฌ๋‹น ๊ฒฝ์ฐฐ๊ด€์„ ํญํ–‰ํ•œ ์‚ฌ๋žŒ๋“ค๋„ ์ด ๊ธฐ๊ธˆ์˜ ์ˆ˜ํ˜œ์ž๊ฐ€ ๋  ์ˆ˜ ์žˆ๋Š”์ง€ ๊ฑฐ๋“ญ ์งˆ๋ฌธํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ฐด ํ™€๋ Œ ์˜์›์€ ๋ธ”๋ž‘์‰ฌ ๋Œ€ํ–‰์ด โ€œ๋ฒ•๋ฌด์žฅ๊ด€ ๋Œ€ํ–‰์ด ์•„๋‹Œ ๋Œ€ํ†ต๋ น์˜ ๊ฐœ์ธ ๋ณ€ํ˜ธ์‚ฌ์ฒ˜๋Ÿผ ํ–‰๋™ํ•˜๊ณ  ์žˆ๋‹คโ€๊ณ  ๋น„๋‚œํ•˜๋ฉฐ, โ€œ์ด๋Ÿฌํ•œ ์ž˜๋ชป๋œ ํ–‰์œ„๊ฐ€ ๊ณ„์†๋˜๋„๋ก ํ•  ์–ด๋– ํ•œ ์ž๊ธˆ๋„ ์ง€๊ธ‰ํ•˜๋Š” ๊ฒƒ์„ ์ •๋‹นํ™”ํ•˜๊ธฐ ์–ด๋ ต๋‹คโ€๊ณ  ๋ง๋ถ™์˜€์Šต๋‹ˆ๋‹ค.

๋†€๋ž๊ฒŒ๋„, ๊ณตํ™”๋‹น ์˜์›๋“ค ์‚ฌ์ด์—์„œ๋„ ์ด ๊ธฐ๊ธˆ์— ๋Œ€ํ•œ ๋ฏธ๋ฌ˜ํ•œ ๋ถˆํŽธํ•จ๊ณผ ์˜๋ฌธ์ด ๊ฐ์ง€๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ƒ์› ๋‹ค์ˆ˜๋‹น ์›๋‚ด๋Œ€ํ‘œ์ธ ์กด ํŠ (John Thune)์กฐ์ฐจ ๊ธฐ์ž๋“ค์—๊ฒŒ โ€œํ–‰์ •๋ถ€๊ฐ€ ๋‹ตํ•ด์•ผ ํ•  ๋งŽ์€ ์งˆ๋ฌธ์ด ์žˆ์„ ๊ฒƒโ€์ด๋ผ๊ณ  ์–ธ๊ธ‰ํ•˜๋ฉฐ ์šฐ๋ ค๋ฅผ ํ‘œํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ํŠธ๋Ÿผํ”„์˜ ๋‹น๋‚ด ์ธ์‚ฌ๋“ค์กฐ์ฐจ ์ด ๊ธฐ๊ธˆ์— ๋Œ€ํ•ด ์—ด๊ด‘ํ•˜๊ฑฐ๋‚˜ ์ž๋ž‘์Šค๋Ÿฌ์›Œํ•˜์ง€ ์•Š๋Š”๋‹ค๋Š” ์‹ ํ˜ธ๋กœ ํ•ด์„๋ฉ๋‹ˆ๋‹ค.

์‹ฌ์ง€์–ด ํŠธ๋Ÿผํ”„ ํ–‰์ •๋ถ€ ๋‚ด๋ถ€์—์„œ๋„ ๊ท ์—ด์˜ ์กฐ์ง์ด ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ๊ธˆ ์กฐ์„ฑ์— ๊ด€์—ฌํ•  ์˜ˆ์ •์ด์—ˆ๋˜ ์žฌ๋ฌด๋ถ€์˜ ์ตœ๊ณ  ๋ณ€ํ˜ธ์‚ฌ๋Š” ๊ธฐ๊ธˆ ๋ฐœํ‘œ ์งํ›„์ธ ์›”์š”์ผ์— ๊ฐ‘์ž‘์Šค๋Ÿฝ๊ฒŒ ์‚ฌ์ž„ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ดˆ๊ธฐ ๋ณด๋„์— ๋”ฐ๋ฅด๋ฉด, ๊ทธ์˜ ์‚ฌ์ž„์€ ์ด ๊ธฐ๊ธˆ ์กฐ์„ฑ์— ๋Œ€ํ•œ ๋ฐ˜๋ฐœ ๋•Œ๋ฌธ์ธ ๊ฒƒ์œผ๋กœ ์•Œ๋ ค์กŒ์Šต๋‹ˆ๋‹ค.

ํŠธ๋Ÿผํ”„ ๊ฐœ์ธ์˜ ์ด๋“๊ณผ ํ–ฅํ›„ ์ „๋ง

๊ทธ๋ ‡๋‹ค๋ฉด ํŠธ๋Ÿผํ”„ ์ „ ๋Œ€ํ†ต๋ น ๋ณธ์ธ๋„ ์ด ๊ธฐ๊ธˆ์˜ ์ˆ˜ํ˜œ์ž๊ฐ€ ๋ ๊นŒ์š”? ๋ฒ•๋ฌด๋ถ€๋Š” ํŠธ๋Ÿผํ”„์™€ ๊ทธ์˜ ๋‘ ์•„๋“ค, ๊ทธ๋ฆฌ๊ณ  ํŠธ๋Ÿผํ”„ ์กฐ์ง์€ ์ด ๊ธฐ๊ธˆ์—์„œ ์ง์ ‘ ๋ˆ์„ ๋ฐ›์ง€ ์•Š์„ ๊ฒƒ์ด๋ผ๊ณ  ๋ฐํ˜”์Šต๋‹ˆ๋‹ค.

๊ทธ๋Ÿฌ๋‚˜ ํ™”์š”์ผ์— ๋ฐํ˜€์ง„ ๋˜ ๋‹ค๋ฅธ ํ•ฉ์˜ ์กฐํ•ญ์€ ๋”์šฑ ์ถฉ๊ฒฉ์ ์ž…๋‹ˆ๋‹ค. IRS๊ฐ€ ํŠธ๋Ÿผํ”„์™€ ๊ทธ์˜ ๊ฐ€์กฑ, ๊ทธ๋ฆฌ๊ณ  ๊ด€๋ จ ์‚ฌ์—…์ฒด์— ๋Œ€ํ•œ ๋ชจ๋“  ์„ธ๋ฌด ๊ฐ์‚ฌ(audit)๋ฅผ ์ค‘๋‹จํ•˜๊ธฐ๋กœ ํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๋Œ€ํ†ต๋ น์—๊ฒŒ ์ œ๊ณต๋˜๋Š” ๋งค์šฐ ์ด๋ก€์ ์ด๊ณ  ํŠน๋ณ„ํ•œ ์–‘๋ณด์ž…๋‹ˆ๋‹ค. ํ˜„์žฌ ํŠธ๋Ÿผํ”„๊ฐ€ ์–ด๋–ค ๊ฐ์‚ฌ๋ฅผ ๋ฐ›๊ณ  ์žˆ๋Š”์ง€ ๋ช…ํ™•ํ•˜์ง€ ์•Š์ง€๋งŒ, ์ด ์กฐํ•ญ์€ ๊ทธ์—๊ฒŒ ์ˆ˜์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ์— ๋‹ฌํ•˜๋Š” ์ง์ ‘์ ์ธ ์žฌ์ •์  ์ด์ต์„ ๊ฐ€์ ธ๋‹ค์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋งŒ์•ฝ IRS๊ฐ€ ๊ทธ์—๊ฒŒ ๋” ๋งŽ์€ ์„ธ๊ธˆ์„ ๋ถ€๊ณผํ•  ๊ณ„ํš์ด์—ˆ๋‹ค๋ฉด, ์ด ํ•ฉ์˜๋กœ ์ธํ•ด ๊ทธ๋Š” ๊ทธ ์„ธ๊ธˆ ์ฒญ๊ตฌ๋ฅผ ํ”ผํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋‰ด์š•ํƒ€์ž„์Šค ๋ณด๋„์— ๋”ฐ๋ฅด๋ฉด, ๊ณผ๊ฑฐ ํŠธ๋Ÿผํ”„๋Š” IRS๊ฐ€ ์Šน์†Œํ–ˆ์„ ๊ฒฝ์šฐ 1์–ต ๋‹ฌ๋Ÿฌ ์ด์ƒ์˜ ์ถ”๊ฐ€ ์„ธ๊ธˆ์„ ๋‚ฉ๋ถ€ํ•ด์•ผ ํ•  ์ˆ˜๋„ ์žˆ๋Š” ๊ฐ์‚ฌ๋ฅผ ๋ฐ›์€ ๋ฐ” ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฒˆ ํ•ฉ์˜๋Š” IRS๊ฐ€ ํŠธ๋Ÿผํ”„ ์ผ๊ฐ€์—๊ฒŒ ๋ฒ•๋ฅ ์— ๋”ฐ๋ผ ๋‚ฉ๋ถ€ํ•ด์•ผ ํ•  ์„ธ๊ธˆ์„ ์š”๊ตฌํ•  ๊ธฐํšŒ๋ฅผ ๋ฐ•ํƒˆํ•œ ๊ฒƒ์ด๋‚˜ ๋‹ค๋ฆ„์—†์Šต๋‹ˆ๋‹ค.

์˜ํšŒ๋‚˜ ๋‹ค๋ฅธ ๊ธฐ๊ด€์ด ์ด ๊ธฐ๊ธˆ ์กฐ์„ฑ์„ ๋ง‰์„ ์ˆ˜ ์žˆ์„๊นŒ์š”? ํ—Œ๋ฒ•์— ๋”ฐ๋ผ ์˜ํšŒ๋Š” โ€œ์˜ˆ์‚ฐ ์ง‘ํ–‰๊ถŒ(power of the purse)โ€œ์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์ด๋ก ์ ์œผ๋กœ ์˜ํšŒ๊ฐ€ ๋ฒ•๋ฅ ์  ํ•ฉ์˜์— ์‚ฌ์šฉ๋˜๋Š” ์ž๊ธˆ ์ง€์ถœ ๋ฐฉ์‹์„ ๋ณ€๊ฒฝํ•˜๊ฑฐ๋‚˜, ์ด ํŠน์ • ์ƒํ™ฉ์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ ๋ฒ•๋ฅ ์„ ํ†ต๊ณผ์‹œํ‚ฌ ์ˆ˜ ์žˆ์Œ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด ๊ถŒํ•œ์€ ํŠธ๋Ÿผํ”„ ํ–‰์ •๋ถ€ ํ•˜์—์„œ ์ƒ๋‹น ๋ถ€๋ถ„ ์นจ์‹๋˜์—ˆ์œผ๋ฉฐ, ์‹ค์ œ๋กœ ์˜ํšŒ๊ฐ€ ์ด๋Ÿฌํ•œ ์กฐ์น˜๋ฅผ ์ทจํ• ์ง€๋Š” ๋ถˆํ™•์‹คํ•ฉ๋‹ˆ๋‹ค.

๋Œ€๋‹ค์ˆ˜์˜ ์ผ๋ฐ˜ ๋Œ€์ค‘์€ ๋ฒ•๋ฌด๋ถ€๊ฐ€ ์ด๋Ÿฌํ•œ โ€˜ํŒ๊ฒฐ ๊ธฐ๊ธˆโ€™์„ ๊ฐ€์ง€๊ณ  ์žˆ๋‹ค๋Š” ์‚ฌ์‹ค์กฐ์ฐจ ์•Œ์ง€ ๋ชปํ–ˆ์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด ์ž๊ธˆ์ด ํ˜„์žฌ์™€ ๊ฐ™์€ ๋ฐฉ์‹์œผ๋กœ ์‚ฌ์šฉ๋˜๋Š” ๊ฒƒ์— ๋Œ€ํ•ด ์šฐ๋ฆฌ๋Š” ์–ด๋–ป๊ฒŒ ๋ฐ›์•„๋“ค์—ฌ์•ผ ํ• ๊นŒ์š”? โ€˜๋” ๋ฐ์ผ๋ฆฌโ€™์˜ ์•ค๋”” ๋“€๋ž€ ๊ธฐ์ž๋Š” ์ด๋ฅผ ํŠธ๋Ÿผํ”„๊ฐ€ ์žฌ์ง‘๊ถŒ ์‹œ ์ด๋ฃจ๊ณ ์ž ํ–ˆ๋˜ ๊ด‘๋ฒ”์œ„ํ•œ ๋ชฉํ‘œ์˜ ๋งฅ๋ฝ์—์„œ ๋ด์•ผ ํ•œ๋‹ค๊ณ  ๋งํ•ฉ๋‹ˆ๋‹ค. ๊ทธ์˜ ์ฃผ์š” ๋ชฉํ‘œ ์ค‘ ํ•˜๋‚˜๋Š” ์ž์‹ ์—๊ฒŒ ๋ฐ˜๋Œ€ํ–ˆ๋˜ ์‚ฌ๋žŒ๋“ค๊ณผ ๊ธฐ๊ด€์— ๋Œ€ํ•œ ๋ณด๋ณต์ด์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ๊ธˆ์˜ ์‚ฌ์šฉ์€ ํŠธ๋Ÿผํ”„ ํ–‰์ •๋ถ€๊ฐ€ ๋‹จ์ˆœํžˆ ์ ์„ ์ฒ˜๋ฒŒํ•˜๋Š” ๊ฒƒ์„ ๋„˜์–ด, ์ด์ œ ๊ทธ์˜ ๋™๋งน์ž๋“ค์—๊ฒŒ ๋ณด์ƒ์„ ์ œ๊ณตํ•˜๋Š” ๋‹ค์Œ ๋‹จ๊ณ„๋กœ ๋‚˜์•„๊ฐ”์Œ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.

์ผ๋ถ€ ์œค๋ฆฌ ์ „๋ฌธ๊ฐ€์™€ ๊ฐ์‹œ ๋‹จ์ฒด๋“ค์€ ์ด๋ฅผ โ€œ๋ฏธ๊ตญ ์—ญ์‚ฌ์ƒ ๊ฐ€์žฅ ๋ถ€ํŒจํ•œ ํ–‰์œ„โ€๋ผ๊ณ  ๋น„๋‚œํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฒ•๋ฌด๋ถ€๋ฅผ ์ ์„ ์ฒ˜๋ฒŒํ•˜๋Š” ๋ฐ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์„ ๋„˜์–ด, ๋™๋งน์ž๋“ค์—๊ฒŒ ๋ณด์ƒํ•˜๋Š” ๋ฐ ์ง์ ‘์ ์œผ๋กœ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ํŠนํžˆ, ๋Œ€ํ†ต๋ น์˜ ์ „ ๊ฐœ์ธ ๋ณ€ํ˜ธ์‚ฌ์ธ ํ† ๋“œ ๋ธ”๋ž‘์‰ฌ๊ฐ€ ์ž„๋ช…ํ•  5๋ช…์˜ ์œ„์›๋“ค์ด ์ด ๋ง‰๋Œ€ํ•œ ์ž๊ธˆ์„ ๋ˆ„๊ตฌ์—๊ฒŒ ์–ผ๋งˆ๋‚˜ ์ง€๊ธ‰ํ• ์ง€์— ๋Œ€ํ•ด ๊ฑฐ์˜ ๋ฌด์ œํ•œ์ ์ธ ์ž์œ ๋ฅผ ๊ฐ€์งˆ ๊ฒƒ์ด๋ผ๋Š” ์ ์€ ์‹ฌ๊ฐํ•œ ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๋ฏธ๊ตญ ๊ตญ๋ฏผ์ด ์„ธ๊ธˆ์œผ๋กœ ๋‚ฉ๋ถ€ํ•˜๊ณ  ์ •๋ถ€๊ฐ€ ์šฐ๋ฆฌ๋ฅผ ๋Œ€์‹ ํ•˜์—ฌ ์ฐจ์ž…ํ•œ ๋ˆ์ด ๋‹จ์ˆœํžˆ ๋Œ€ํ†ต๋ น์—๊ฒŒ ์ถฉ์„ฑํ–ˆ๋˜ ์‚ฌ๋žŒ๋“ค์—๊ฒŒ ์ง€๊ธ‰๋  ์ˆ˜ ์žˆ์Œ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜์ฒœ, ์ˆ˜๋งŒ, ์ˆ˜์‹ญ๋งŒ, ์‹ฌ์ง€์–ด ์ˆ˜๋ฐฑ๋งŒ ๋‹ฌ๋Ÿฌ๊ฐ€ ํ˜„๊ธˆ์œผ๋กœ ์ง€๊ธ‰๋  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์ •๋ง ๊ธ‰์ง„์ ์ด๊ณ  ์ถฉ๊ฒฉ์ ์ธ ์ƒํ™ฉ์ž…๋‹ˆ๋‹ค.

๊ฒฐ๋ก 

ํŠธ๋Ÿผํ”„ ์ „ ๋Œ€ํ†ต๋ น์˜ IRS ์†Œ์†ก ์ฒ ํšŒ์™€ โ€˜๋ฌด๊ธฐํ™” ํ”ผํ•ด ๋ณด์ƒ ๊ธฐ๊ธˆโ€™ ์กฐ์„ฑ์€ ๋‹จ์ˆœํžˆ ๋ฒ•์  ํ•ฉ์˜๋ฅผ ๋„˜์–ด, ๋ฏธ๊ตญ์˜ ๋ฒ•์น˜์ฃผ์˜์™€ ์ •๋ถ€ ๊ธฐ๊ด€์˜ ๋…๋ฆฝ์„ฑ์— ๋Œ€ํ•œ ์‹ฌ๊ฐํ•œ ๋„์ „์œผ๋กœ ํ‰๊ฐ€๋ฉ๋‹ˆ๋‹ค. ๋‚ฉ์„ธ์ž์˜ ๋ˆ์ด ์ •์น˜์  ๋ณด๋ณต๊ณผ ๋™๋งน์ž์— ๋Œ€ํ•œ ๋ณด์ƒ์ด๋ผ๋Š” ์‚ฌ์  ๋ชฉ์ ์œผ๋กœ ์‚ฌ์šฉ๋  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฐ€๋Šฅ์„ฑ์€ ๋ฏผ์ฃผ์ฃผ์˜์˜ ๊ทผ๊ฐ„์„ ํ”๋“œ๋Š” ์œ„ํ—˜ํ•œ ์„ ๋ก€๊ฐ€ ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ๊ธˆ์˜ ์ž‘๋™ ๋ฐฉ์‹๊ณผ ์ˆ˜ํ˜œ ๋Œ€์ƒ์— ๋Œ€ํ•œ ํˆฌ๋ช…์„ฑ ํ™•๋ณด, ๊ทธ๋ฆฌ๊ณ  ์˜ํšŒ์˜ ๊ฐ์‹œ์™€ ๊ฐœ์ž…์ด ์ ˆ์‹คํžˆ ์š”๊ตฌ๋˜๋Š” ์ƒํ™ฉ์ž…๋‹ˆ๋‹ค. ๊ณผ์—ฐ ๋ฏธ๊ตญ์€ ์ด ์ „๋ก€ ์—†๋Š” ๋…ผ๋ž€ ์†์—์„œ ๋ฒ•์น˜์™€ ๋ฏผ์ฃผ์  ๊ฐ€์น˜๋ฅผ ์ˆ˜ํ˜ธํ•  ์ˆ˜ ์žˆ์„์ง€, ์ „ ์„ธ๊ณ„์˜ ์ด๋ชฉ์ด ์ง‘์ค‘๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.