English
Based on โAnthropic Just Bought a Dev Tools Startup for $300M. Hereโs What Its Founder Told Me.โ from Every Watch the original video
The Internetโs 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, theyโre at the forefront of enabling AI to do the same, grappling with the complexities of whatโs known as Model Context Protocol (MCP).
โAPIs are the dendrites of the internet,โ explains Ratray, drawing a vivid analogy to the neural connections in our brains. Just as thoughts donโt 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 โnew computer has entered the chatโ: LLMs.
The Promise and Pitfalls of Model Context Protocol (MCP)
Model Context Protocol (MCP) represents the internetโs attempt to give LLMs their own native interface. Imagine every website or web service transforming into a set of โtoolsโ an AI can use directly to get work done. For instance, a Gmail MCP might offer tools like โsend mailโ or โread 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. โLetโs say Dan walks into my store and buys a pair of stripey socks,โ he begins. โThe next day, somethingโs wrong, and I tell my team, โHey, can we refund Dan for those stripey socks and send him a discount code with a thank you note?โโ
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. โThatโs the big vision of MCP,โ Ratray emphasizes. โTo 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, โitโs 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, youโd 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 โcontext budgetโโhundreds 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 โnew research problem.โ Humans can learn Python SDKs; LLMs struggle to โthinkโ 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. โYou donโt 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 โdynamic 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. โI 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, โHey, 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 โmaster 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 โcacheโ information, avoiding repetitive searches and quickly retrieving relevant data for future queries or board preparations.
โThatโs [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 โCyborgโ 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 โcyborgs.โ This doesnโt mean literal human-robot hybrids, but rather AI agents that are โpart GPT neural net LLM, part AI, and part code.โ The โcodeโ refers to traditional CPU-based software, not just GPU-driven neural networks.
Ratray envisions a future where LLMs donโt just pick from a vast menu of pre-defined tools, but instead write the code themselves. He proposes a model where, instead of a โbajillion tools,โ the LLM has just two:
- Execute Code: The LLM is given a text box to write TypeScript (or another language) using an APIโs SDK. For example, it could write
Stripe.transactions.list()orStripe.refunds.create(). - 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 minimalโperhaps a thousand tokens or less.
- Efficient Execution: The AI writes the code, which is then executed by the API providerโs 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 doesnโt exist), the code execution tool can immediately identify the error, suggest correct alternatives (likepaymentIntentsororders), and provide relevant documentation. This โsub-agentโ is well-trained and unburdened by the broader conversationโs context.
โLLMs 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 โcyborgโ 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 whatโs exposed through MCP. โRight now you see people trying to implement security by sort of limiting whatโs exposed through MCP, and that kind of makes sense, but at the end of the day, you could do anything thatโs 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 โWas 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 giantโs 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 โHow I AI,โ the 2026 iteration was no different. This year, Google went โfull 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 didnโt quite work โ at least not yet.
The Foundation: Gemini 3.5 Flash โ Speed Meets Intelligence
At the core of Googleโs 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: โtheir fastest, smartest coding model.โ What truly sets Flash apart, according to Googleโs benchmarks, is its dual prowess. It not only โrivals the intelligence of some of our favorite coding modelsโ like Opus and GPT-4 but is also โfour times as fast.โ This combination promises a โsuper smart modelโฆ 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 โcatch-upโ to competitors like Anthropic and OpenAI, they come with two distinct advantages: the sheer speed of the Flash models and a pronounced โcreative consumer bent.โ
Beyond coding, Geminiโs multimodal strength continues to be a standout. Claraveville has consistently championed Gemini for its ability to handle diverse data types: โI have always told people if you are working with files, videos, any sort of transformative work where you have to go from one modalityโฆ 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 โprojectsโ (folder-constrained workspaces) and โscheduled 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. โHooksโ allow developers to tap into various lifecycle events of their agent, offering on-demand customization. Claraveville was particularly keen on the new โslash 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 โaggressiveโ clarification process, where the agent will โask clarification questions back and really get to the heart of what your requirements are.โ Claraveville mused, โNow 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? Weโll 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 didnโt quite match the โFlashโ hype in this specific instance. โIโm not noticing it is particularly fasterโฆ it doesnโt feel that much faster to me,โ she noted, concluding that while it was โvery nice, fast, kind of what I would expect from a coding agent,โ it still felt like โplaying 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 โconnected to your Google Workspace appsโ โ reading Sheets, drafting Gmails, organizing Drive, and managing calendars. This is Googleโs strategic move to โown that experience themselves,โ carving off internal enterprise and personal assistant use cases. The studio also now supports creating Android apps.
However, Claravevilleโs 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. โIt didnโt do anything magical,โ she reported, highlighting a recurring frustration: โtheyโve announced a bunch of stuff and then at the very, very bottom theyโre like itโs available to some subset of people or itโs 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: Googleโs product naming strategy. โI 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 โGemini, Google AI, the consumerish AI product.โ Her plea to the Google team was clear: โa comprehensive brand analysis and singular brand that I can work with.โ
The consumer-facing Gemini AI product received a redesign, dubbed โsomething ridiculousโ by Claraveville but intended to โmake you feel goodโ with its โglowโ 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 โupscale 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: โThis 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 โpretty fast.โ
The real excitement in the creative space revolved around Omni, Googleโs new video generation model, positioned as โNano Banana for video.โ Omni promises โlonger, more consistent, more photorealistic videos using AI,โ capable of leveraging โreference materials.โ Claraveville emphasized Geminiโs strong multimodal capabilities, noting that Omni allows users to โcreate video from anythingโ and โground 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 Omniโs power by animating a drawing her child made: โAnimate 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 โkid 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 โcinematic quality production.โ Flow allows users to blend multimodal references, edit conversationally, and even define characters (e.g., โCaptain 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: โPlease donโt steal my identity!โ she quipped. The outcome? Another I/O disappointment. โNo. No sir, I couldnโtโฆ it didnโt even create the avatar.โ This led to a stark conclusion: โThe promise is really good for some of these things, but the reality is if youโre not able to use them or theyโre broken on the dayโฆ people are going to lose patience for some of this.โ
Design & Marketing Tools: Pamelli and Stitch
The product naming confusion continued with Pamelli, Googleโs brand and marketing content generator, and Stitch, their design tool. Pamelli introduced an agent that can create a โcore brand identity from a website,โ including brand books, campaign assets, and photoshoots โ a feature reminiscent of Claude Design. Claraveville highlighted Googleโs 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 Pamelliโs ability to generate a brand book, which later yielded โdefinitely my brand colors. This is my tagline. And now I can createโฆ a new website.โ Her final assessment of the generated website: โitโs fine.โ
Stitch, described as an โin-browser Figma,โ showcased features like streaming into a design canvas, starting with existing designs, and inline AI edits. Claraveville noted, โThis 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: โEvery AI tool ships their own version of slop. Googleโs 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, Googleโs portfolio is undeniably expansive.
Claravevilleโs 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, Claravevilleโs key takeaways were clear: โThe video piece is the part that Iโm 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: โthere 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 โTrumpโs Taxpayer-Funded Revenge Planโ from New York Times Podcasts Watch the original video
Taxpayer-Funded Retribution: Unpacking Trumpโs $1.7 Billion Slush Fund
In a move that has sent shockwaves through Washington and ignited bipartisan fury, President Trumpโs administration has unveiled an unprecedented $1.776 billion fund of taxpayer money. Ostensibly designed to compensate โvictims 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 Presidentโs 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 Trumpโs 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 Trumpโs 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 worldโs 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 someoneโs 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 Trumpโs 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, โYou 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 Judgeโs 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 โcharadeโ filed under false pretenses. She ordered both the Justice Department and Trumpโs 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 โfeverishlyโ 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 โunprecedented gigantic pot of taxpayer moneyโ to compensate โvictims of weaponization.โ
The $1.776 Billion โVictimโ Fund Takes Shape
The fund, totaling $1,776,000,000 โ a figure many observers note is no coincidence, referencing the year of Americaโs 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 โVictimโ?
While the Justice Departmentโs official language refers to โvictims 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 โsomebody 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 โprofit 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 โweaponization,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 โmight 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 โunusual,โ it was โnot unprecedentedโ and claimed the fund was โnot limited to Republicans,โ โnot limited to January 6th,โ and open to โanybody in this countryโฆ if they believe they are a victim of weaponization.โ
Democrats Cry Foul, Republicans Get โSqueamishโ
However, Blancheโs defense did little to quell the outrage. Democratic lawmakers lambasted the fund as an โoutrageous unprecedented slush fundโ designed to pay the presidentโs political allies. Senator Chris Van Hollen accused Blanche of โpure theft of public fundsโ and called the idea of rewarding convicted criminals โobscene.โ He further asserted that it was part of President Trumpโs โbroader revenge tourโ and bluntly told Blanche, โYou are still acting as the presidentโs 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 โthere are and will be continue to be a lot of questions around that that the administration is going to have to answer.โ This โmild discontentโ from top Republican lawmakers signaled that even within Trumpโs 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 fundโs 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.
Trumpโs 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 โextremely 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 โrobbed 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 โtens 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 Trumpโs original lawsuit against the IRS were ultimately unsuccessful. Now, the focus shifts to Congressโs โpower 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 โerodedโ under the current presidency, and it remains โunclear 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 Trumpโs stated goal of โretribution 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 โcompensation for President Trumpโs alliesโ with taxpayer dollars.
Ethicists are calling this โthe 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 Presidentโs former personal lawyer, will have โseemingly 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 โradical and striking situationโ means that those deemed โloyal to the presidentโ could be โcashing 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 publicโs money, fundamentally challenging the principles of impartial governance and public trust.
ํ๊ตญ์ด
โAnthropic Just Bought a Dev Tools Startup for $300M. Hereโs What Its Founder Told Me.โ โ Every ๊ธฐ๋ฐ ๊ธฐ์ฌ ์๋ณธ ์์ ๋ณด๊ธฐ
AI ์๋, ์ปดํจํฐ ์ํต์ ๋ฏธ๋๋ฅผ ๋ฌป๋ค: Stainless CEO ์๋ ์ค ๋ซ๋ ์ด์์ ๋ํ
์ธํฐ๋ท์ ์ปดํจํฐ๋ค์ด ์๋ก ๋ํํ๋ฉฐ ์๋ํ์ง๋ง, ๊ทธ ์ ์ฒด์ ์ธ ์ํคํ ์ฒ(architecture)๋ ์ธ๊ณต์ง๋ฅ(AI)์ด ๋ฑ์ฅํ๊ธฐ ์ ์ ์ธ์์ ๋ง์ถฐ ์ค๊ณ๋์์ต๋๋ค. ์ด์ ์ฐ๋ฆฌ๋ ๋ชจ๋ธ ์ปจํ ์คํธ ํ๋กํ ์ฝ(Model Context Protocol, MCP)์ด๋ผ๋ ์๋ก์ด ๋ฐฉ์์ผ๋ก AI๋ฅผ ์ธํฐ๋ท์ ์ฐ๊ฒฐํ๋ ค ํฉ๋๋ค. MCP๋ ๋ชจ๋ ์น์ฌ์ดํธ๋ ์น ์๋น์ค๋ฅผ AI๊ฐ ์์ ์ ์ํํ๋ ๋ฐ ๊ธฐ๋ณธ์ ์ผ๋ก ์ฌ์ฉํ ์ ์๋ ๋๊ตฌ ์ธํธ๋ก ์ ํํฉ๋๋ค. ์์ผ๋ก 10๋ ๋์ MCP๋ฅผ ์ ํ์ฉํ๋ ์ํํธ์จ์ด ๊ธฐ์ ๋ค์ด ์์ฅ์ ์ ๋ํ ๊ฒ์ ๋๋ค.
์ด๋ฌํ ๋ณํ์ ์ค์ฌ์์ ์ค์ํ ์ญํ ์ ํ๋ ๊ธฐ์ ์ด ๋ฐ๋ก Stainless์ ๋๋ค. Stainless๋ ์ปดํจํฐ๋ค์ด ์๋ก ์ํํ๊ฒ ์ํตํ๋๋ก ๋๋ ํ์ฌ๋ก, OpenAI, Anthropic๊ณผ ๊ฐ์ ๋๊ธฐ์ ๋ค์ ์ํ API(Application Programming Interface) ๋ฐ SDK(Software Development Kit)๋ฅผ ๊ฐ๋ฐํฉ๋๋ค. ์ต๊ทผ์๋ MCP ์๋ฒ ๊ตฌ์ถ์๋ ์ ๊ทน์ ์ผ๋ก ๋์๊ณ ์์ต๋๋ค.
Every ์ฑ๋์ Stainless์ ์ฐฝ๋ฆฝ์์ด์ CEO์ธ ์๋ ์ค ๋ซ๋ ์ด(Alex Ratray)๋ฅผ ์ด๋ํ์ฌ MCP์ ๋ฏธ๋, ํจ๊ณผ์ ์ธ MCP ์ค๊ณ ๋ฐฉ๋ฒ, ํ์ฅ์ฑ ๋ฐ ๋ณด์ ๋ฌธ์ , ๊ทธ๋ฆฌ๊ณ AI๊ฐ ์ธํฐ๋ท์ ๋ ์ ํ์ฉํ ์ ์๋ ๋ชจ๋ธ์ ๋ํ ์ฌ์ธต์ ์ธ ๋ํ๋ฅผ ๋๋ด์ต๋๋ค.
์๋ ์ค ๋ซ๋ ์ด์ Stainless: ์ปดํจํฐ ์ํต์ ์ ๋ฌธ๊ฐ
์๋ ์ค ๋ซ๋ ์ด๋ Stainless์ ์ฐฝ๋ฆฝ์์ด์ CEO์ ๋๋ค. ๊ทธ๋ ์ด์ ์ Stripe์์ API ๊ฐ๋ฐ ์ ๋ฌด๋ฅผ ๋ด๋นํ์ผ๋ฉฐ, ํ์ฌ Stainless๋ฅผ ํตํด ์ปดํจํฐ ๊ฐ์ ์ํต์ ์ํ API๋ฅผ ์ ๊ณตํ๋ฉฐ OpenAI, Anthropic ๋ฑ ์ฃผ์ ๊ธฐ์ ๋ค์ ๊ธฐ์ ๊ธฐ๋ฐ์ ๋ค์ง๊ณ ์์ต๋๋ค. ์๋ ์ค๋ ์งํ์ ๋(Dan)๊ณผ ๋ํ ์์ ๋ถํฐ ์น๊ตฌ์์ผ๋ฉฐ, ์๋ก ์คํํธ์ ์ ์์ํ๋ฉฐ ํจ๊ป ์ฑ์ฅํด์จ ํน๋ณํ ์ธ์ฐ์ ๊ฐ์ง๊ณ ์์ต๋๋ค.
์๋ ์ค๋ Stainless์ ๋ฏธ์ ์ด โ์ปดํจํฐ๋ค์ด ์๋ก ์ฝ๊ฒ ๋ํํ ์ ์๋๋ก ๋ง๋๋ ๊ฒโ์ด๋ผ๊ณ ์ค๋ช ํฉ๋๋ค. ์ด๋ฅผ ์ดํดํ๊ธฐ ์ํด ๋จผ์ API์ ๊ฐ๋ ๋ถํฐ ์ง๊ณ ๋์ด๊ฐ๋๋ค.
API: ์ธํฐ๋ท์ ์ ๊ฒฝ๋ง
API๋ ์ ํ๋ฆฌ์ผ์ด์ ํ๋ก๊ทธ๋๋ฐ ์ธํฐํ์ด์ค(Application Programming Interface)์ ์ฝ์๋ก, ํ ์ปดํจํฐ ํ๋ก๊ทธ๋จ์ด ๋ค๋ฅธ ์ปดํจํฐ ํ๋ก๊ทธ๋จ๊ณผ ๋ํํ๋ ๋ฐฉ์, ์ฆ ์ฑ๊ณผ ์ฑ์ด ์ํตํ๋ ํต๋ก๋ฅผ ์๋ฏธํฉ๋๋ค. ์๋ ์ค๋ API๋ฅผ โ์ธํฐ๋ท์ ์์๋๊ธฐ(dendrite)โ์ ๋น์ ํฉ๋๋ค. ๋์ ๋ด๋ฐ๋ค์ด ์์๋๊ธฐ๋ฅผ ํตํด ์ ๋ณด๋ฅผ ๊ตํํด์ผ ์ฌ๊ณ ๊ฐ ๊ฐ๋ฅํ๋ฏ์ด, ํด๋ผ์ฐ๋์ ๋ชจ๋ ์๋ฒ๋ค์ด ์๋ก ๋ํํ์ง ์๋๋ค๋ฉด ์ธํฐ๋ท์ ์กด์ฌํ ์ ์์ต๋๋ค. ํ๋ ์ํํธ์จ์ด์ ๊ฑฐ์ ๋ชจ๋ ๋ถ๋ถ์ด API๋ฅผ ํตํด ์ฐ๊ฒฐ๋์ด ์์ผ๋ฉฐ, ์ด๋ ๊ธฐ์ ์๋ํ์ ํต์ฌ ๋๋ ฅ์ ๋๋ค.
ํ์ง๋ง AI์ ๋ฑ์ฅ์ผ๋ก ์๋ก์ด ์ ํ์ โ์ปดํจํฐโ๊ฐ ์ด ๋ํ์ ์ฐธ์ฌํ๊ฒ ๋์์ต๋๋ค. ๊ณผ๊ฑฐ์๋ ์ธ๊ฐ์ด ์ฌ์ฉ์ ์ธํฐํ์ด์ค(UI)๋ฅผ ํตํด ์ปดํจํฐ์ ์ํธ์์ฉํ๊ฑฐ๋, ์ปดํจํฐ๊ฐ API๋ฅผ ํตํด ๋ค๋ฅธ ์ปดํจํฐ์ ๋ํํ์ต๋๋ค. ์ด์ ๋ ๋๊ท๋ชจ ์ธ์ด ๋ชจ๋ธ(Large Language Model, LLM)์ด ์ปดํจํฐ์ ์ํธ์์ฉํด์ผ ํ๋ ์๋๊ฐ ์จ ๊ฒ์ ๋๋ค. ๋ฐ๋ก ์ด ์ง์ ์์ MCP, ์ฆ ๋ชจ๋ธ ์ปจํ ์คํธ ํ๋กํ ์ฝ์ด ์ค์ํด์ง๋๋ค. MCP๋ LLM์ ์ปดํจํฐ์ ์ฐ๊ฒฐํ๊ธฐ ์ํ ์์คํ ์ผ๋ก, Stainless๊ฐ ์ง์ค์ ์ผ๋ก ํฌ์ํ๊ณ ์๋ ๋ถ์ผ์ ๋๋ค.
SDK์ MCP: ์ธ๊ฐ๊ณผ AI๋ฅผ ์ํ ์ธํฐํ์ด์ค
Stainless์ ํต์ฌ ์ ํ์ ์ํํธ์จ์ด ๊ฐ๋ฐ ํคํธ(Software Development Kit, SDK)์
๋๋ค. ์๋ฅผ ๋ค์ด, Stripe์ REST API๋ฅผ ํ์ด์ฌ(Python) ๊ฐ๋ฐ์๊ฐ ํธ๋ฆฌํ๊ฒ ์ฌ์ฉํ๋ ค๋ฉด pip install stripe ๋ช
๋ น์ผ๋ก Stripe ํ์ด์ฌ SDK๋ฅผ ์ค์นํ ํ Stripe.customers.create์ ๊ฐ์ ์ฝ๋๋ฅผ ํตํด ์์ฝ๊ฒ ๊ณ ๊ฐ ๊ฐ์ฒด๋ฅผ ์์ฑํ ์ ์์ต๋๋ค. SDK๋ ๊ฐ๋ฐ์๋ค์ด API์ ์ฝ๊ฒ ์ํธ์์ฉํ ์ ์๋๋ก ๋๋ ๋๊ตฌ์
๋๋ค.
๊ทธ๋ ๋ค๋ฉด LLM์ด API์ ์ฝ๊ฒ ์ํธ์์ฉํ ์ ์๋ ๋ฐฉ๋ฒ์ ๋ฌด์์ผ๊น์? ๋ฐ๋ก MCP์ ๋๋ค. ์ด๋ก ์ ์ผ๋ก MCP๋ ์น์ฌ์ดํธ๊ฐ ์ธ๊ฐ์ ์ํด ์ค๊ณ๋ ๊ฒ์ฒ๋ผ, LLM์ ์ํด ๋๊ตฌ ์ธํธ๋ฅผ ๋ ธ์ถํ๋ ์ญํ ์ ํฉ๋๋ค. ์๋ฅผ ๋ค์ด, Gmail MCP๋ โ๋ฉ์ผ ๋ณด๋ด๊ธฐโ, โ๋ฉ์ผ ์์ฑํ๊ธฐโ, โ๋ฐ์ ํธ์งํจ ์ฝ๊ธฐโ์ ๊ฐ์ ๋๊ตฌ๋ค์ LLM์ ์ ๊ณตํ์ฌ, LLM์ด ์ง์ ๋ก๊ทธ์ธํ์ฌ ์์ ์ ์ํํ ์ ์๋๋ก ํฉ๋๋ค.
ํ์ง๋ง ์๋ ์ค๋ ํ์ฌ MCP์ ๊ตฌํ์ด ์์ง ์ด์์ ์ด์ง ์๋ค๊ณ ์ง์ ํฉ๋๋ค. โ์ฌ๋๋ค์ด MCP๋ฅผ ์คํํ๊ณ ์์ง๋ง, ๊ธฐ๋๋งํผ ์ ์๋ํ์ง ์๋๋คโ๋ ๊ฒ์ ๋๋ค.
MCP์ ์ฝ์๊ณผ ํ์ค: โ์์ด์ ํธ AIโ์ ๋น์
์๋ ์ค๋ MCP์ ์์ด์ ํธ AI(Agentic AI)์ ๊ถ๊ทน์ ์ธ ๋น์ ์ ์ค๋ช ํ๊ธฐ ์ํด ๋งค์ฐ ์ผ์์ ์ธ ์์๋ฅผ ์ ์ํฉ๋๋ค.
์ผ์ ์ ๋ฌด์ ์๋ํ: ์ค๋ฌด๋ฌ ์๋ง ํ๋ถ ์ฌ๋ก
์ด๋ค ๊ณ ๊ฐ(๋)์ด ๊ฐ๊ฒ์์ ์ค๋ฌด๋ฌ ์๋ง์ ํฌํจํ ๋ช ๊ฐ์ง ๋ฌผ๊ฑด์ ๊ตฌ๋งคํ๋๋ฐ, ๋ค์ ๋ ๋ฌธ์ ๊ฐ ๋ฐ์ํ์ฌ ํ๋ถ์ ์์ฒญํ๋ค๊ณ ๊ฐ์ ํด ๋ด ์๋ค. ๋งค์ฅ ์ง์์ด โ๋์ด ์ด์ ์ฐ ์ค๋ฌด๋ฌ ์๋ง์ ํ๋ถํด์ฃผ๊ณ , ๋ค์ ๋ฐฉ๋ฌธ ์ ์ฌ์ฉํ ํ ์ธ ์ฝ๋์ ๊ฐ์ฌ ํธ์ง๋ฅผ ๋ณด๋ด์คโ๋ผ๊ณ ์ง์ํฉ๋๋ค.
ํ์ฌ์ ๋ฐฉ์์ผ๋ก๋ ์ง์์ด ์ฌ๋ฌ ๋จ๊ณ๋ฅผ ๊ฑฐ์ณ์ผ ํฉ๋๋ค. ๋ด๋ถ ๊ด๋ฆฌ ์์คํ ์ ์ด๊ณ , Stripe ๋์๋ณด๋์์ ๊ฒฐ์ ๋ด์ญ์ ํ์ธํ๋ฉฐ โ๋โ์ด๋ผ๋ ์ด๋ฆ๊ณผ โ์ค๋ฌด๋ฌ ์๋งโ์ด ํฌํจ๋ ์ฃผ๋ฌธ์ ์ฐพ์ต๋๋ค. ์ฌ๋ฌ ๋ฒ์ ํด๋ฆญ๊ณผ ์คํฌ๋กค์ ๊ฑฐ์ณ ํด๋น ์ฃผ๋ฌธ์ ์ฐพ์ ํ ํ๋ถ ํ๋ฉด์ผ๋ก ์ด๋ํ์ฌ ์ ํํ ๊ธ์ก์ ํ๋ถ ์ฒ๋ฆฌํฉ๋๋ค. ์ด์ด์ ํ ์ธ ์ฝ๋๋ฅผ ์์ฑํ๊ณ , ๋ค๋ฅธ SaaS(์๋น์คํ ์ํํธ์จ์ด) ์ฑ์ ๋ก๊ทธ์ธํ์ฌ ๊ฐ์ฌ ํธ์ง์ ํจ๊ป ํ ์ธ ์ฝ๋๋ฅผ ๋ณด๋ ๋๋ค. ์ด๋ฌํ ํ๋์ ๊ฐ๋จํ ์์ ์ ์ํด 5๊ฐ์ ๋ค๋ฅธ ์ฑ์ ์ค๊ฐ๋ฉฐ 15๋ฒ์ ํด๋ฆญ๊ณผ ๋ก๋ฉ์ ๊ธฐ๋ค๋ ค์ผ ํ๋ ๊ฒ์ด ํ์ค์ ๋๋ค.
์์ด์ ํธ AI์ ์ฝ์์ ๋ฐ๋ก ์ด ์ง์ ์์ ๋น์ ๋ฐํฉ๋๋ค. ์ง์์ด ChatGPT๋ Claude ๊ฐ์ AI์๊ฒ โ์ฑ๋ด ์น๊ตฌ, ๋ด ์น๊ตฌ ๋ D.์ ํ๋ถ์ ๋์์ค ์ ์์๊น?โ๋ผ๊ณ ์ ๋ ฅํ๋ฉด, AI๊ฐ ์ค์ค๋ก 5๊ฐ์ ์ฑ๊ณผ 15๊ฐ์ ํ๋ฉด์ ์ค๊ฐ๋ฉฐ ํ์ํ ์์ ์ ์๋ฃํ๊ณ โ์๋ฃํ์ต๋๋คโ๋ผ๊ณ ๋ณด๊ณ ํ๋ ๊ฒ์ ๋๋ค.
๊ด๋ฒ์ํ ๋๊ตฌ ๋ ธ์ถ์ ๋์ : ์ปจํ ์คํธ ์๋์ฐ ํ๊ณ
์์ด์ ํธ AI๊ฐ ์ด๋ฌํ ๋น์ ์ ์คํํ๋ ค๋ฉด, ์ธ๊ฐ ์ด์์๊ฐ ํ ์ ์๋ ๋ชจ๋ ์์ ์ ์ํํ ์ ์์ด์ผ ํฉ๋๋ค. ์ด๋ ๋จ์ํ โํ๋ถ ์์ฑโ์ด๋ โ๊ฑฐ๋ ๋ด์ญ ์กฐํโ ๊ฐ์ ๋ช ๊ฐ์ง ๋๊ตฌ๋ฟ๋ง ์๋๋ผ, Stripe ๋์๋ณด๋์์ ๊ฐ๋ฅํ ๋ชจ๋ ๊ธฐ๋ฅ, ์ฆ Stripe API์์ ์ ๊ณตํ๋ ์๋ฐฑ ๊ฐ์ ์๋ํฌ์ธํธ(endpoint)๋ฅผ ๋ชจ๋ ํ์ฉํ ์ ์์ด์ผ ํจ์ ์๋ฏธํฉ๋๋ค.
๋ฌธ์ ๋ ์ฌ๊ธฐ์ ์์ต๋๋ค. ํ์ฌ LLM์ ์ด๋ฌํ ๋ฐฉ๋ํ ๋๊ตฌ ๋ชฉ๋ก๊ณผ ๊ฐ ๋๊ตฌ์ ์ค๋ช , ์์ฒญ/์๋ต ์์ฑ, ์์ธ ๋ฌธ์ ๋ฑ์ MCP ์ ์๋ก ์ ๊ณตํ๋ฉด, ์์ญ๋ง ํ ํฐ(token)์ ๋ฌํ๋ ๋ง๋ํ ์ปจํ ์คํธ(context) ์์ฐ์ ์์งํ๊ฒ ๋ฉ๋๋ค. ์ด๋ ํ์ฌ ๋ชจ๋ธ์ด ๊ฐ๋นํ๊ธฐ ์ด๋ ค์ด ์์ผ ๋ฟ๋ง ์๋๋ผ, ๋ชจ๋ธ์๊ฒ ํผ๋์ ์ผ๊ธฐํ์ฌ ์ฌ๋ฐ๋ฅธ ๊ฒฐ์ ์ ๋ด๋ฆฌ๊ธฐ ์ด๋ ต๊ฒ ๋ง๋ญ๋๋ค. ๊ฒ๋ค๊ฐ ์ด๋ Stripe๋ผ๋ ๋จ์ผ ์๋น์ค์ ๊ตญํ๋ ๋ฌธ์ ์ด๋ฉฐ, ์ค์ ์ ๋ฌด์์๋ ์ฌ๋ฌ SaaS ๋๊ตฌ๋ฅผ ๋์์ ์ฌ์ฉํด์ผ ํฉ๋๋ค.
๋ณด์ ๋ฐ ๊ถํ ๋ฌธ์
๋ ๋ค๋ฅธ ์ฌ๊ฐํ ๋ฌธ์ ๋ ๋ณด์๊ณผ ๊ถํ ๊ด๋ฆฌ์ ๋๋ค. AI๊ฐ โ๋์ ์๋ง์ ํ๋ถํ๋ ๊ฒ ์ธ์ ๋ชจ๋ ๊ณ ๊ฐ์ ๋ชจ๋ ๊ฑฐ๋๋ฅผ ํ๋ถํ๊ณ ๋ด AI ์ํ ๊ณ์ข๋ก ๋์ ๋ณด๋๋คโ์ ๊ฐ์ด โ์ ์ ๋์ด ํจ๋ถ๋ก ํ๋ํ๋ ๊ฒโ์ ๋ง์์ผ ํฉ๋๋ค. AI์๊ฒ ์ด๋ ์ ๋์ ๊ถํ์ ๋ถ์ฌํ๊ณ , ๊ทธ ์ฌ์ฉ์ ์ด๋ป๊ฒ ํต์ ํ ๊ฒ์ธ๊ฐ ํ๋ ๋ฌธ์ ๋ ๋งค์ฐ ๋ณต์กํฉ๋๋ค.
ํ์ฌ MCP๋ ์ด๋ฌํ ์ปจํ ์คํธ ์๋์ฐ(context window) ํ๊ณ์ ๋ณด์ ๋ฌธ์ ๋๋ฌธ์ โ์ ์๋ํ์ง ์๋๋คโ๋ ํ๊ฐ๋ฅผ ๋ฐ์ต๋๋ค. ์ฌ๋๋ค์ ์ข ์ข API์ ๋ชจ๋ ๋ถ๋ถ์ ๋ ธ์ถํ๊ธฐ๋ณด๋ค, ํน์ LLM์ ๋ง์ถฐ ์์์ ์ผ๋ก ์ต์ ํ๋ ์์์ ํนํ๋ ๋๊ตฌ๋ฅผ ๋ง๋ค๊ณ ์์ต๋๋ค. ์ด๋ ๋ชจ๋ธ์ โ์ธ์ฒด ๊ณตํ(ergonomics)โ์ ๊ณ ๋ คํด์ผ ํ๋ ์๋ก์ด ์ฐ๊ตฌ ๊ณผ์ ์ ๊ฐ์ต๋๋ค.
์ข์ MCP ์๋ฒ ์ค๊ณ์ ์์น
์๋ ์ค๋ ํ์ฌ์ ๊ธฐ์ ์์ค์์ ํจ๊ณผ์ ์ธ MCP ์๋ฒ๋ฅผ ๊ตฌ์ถํ๊ธฐ ์ํ ์์น๋ค์ ์ ์ํฉ๋๋ค.
- ์ ํ ๊ด๋ฆฌ ์ค์ฌ ์ ๊ทผ: ์์ฅ์ ๋๊ฐ ๊ณ ๊ฐ์ ์ค์ ๋์ฆ๋ฅผ ํ์ ํ๊ณ , ์ํํธ์จ์ด๋ฅผ ์ฌ์ฉํ๋ ๋ชจ์ต์ ๊ด์ฐฐํ๋ฉฐ AI๋ฅผ ํตํด ์ด๋ค ์๋ก์ด ๊ฐ์น๋ฅผ ์ฐฝ์ถํ ์ ์์์ง ๊ณ ๋ฏผํด์ผ ํฉ๋๋ค.
- ์ ๊ตํ ์์ง๋์ด๋ง ์์ : ๋ชจ๋ธ์ ๋ง์ถฐ ๋๊ตฌ๋ฅผ ์ ๊ตํ๊ฒ ๋ํ(wrapping)ํ๊ณ , ๋ค์ํ ํด๋ผ์ด์ธํธ(Cursor, Cloud Code ๋ฑ)์ ๊ธฐ๋ณธ ๋ชจ๋ธ์ ๊ณ ๋ คํ ํ๊ฐ ์์คํ (eval system)์ ๊ตฌ์ถํด์ผ ํฉ๋๋ค.
- ํผ๋๋ฐฑ ๋ฉ์ปค๋์ฆ: ํ์ฌ๋ AI๊ฐ ๋๊ตฌ ํธ์ถ(tool call)์ ์ฑ๊ณต์ ์ผ๋ก ์ํํ๋์ง, ์ฌ์ฉ์์๊ฒ ์ ์ฉํ๋์ง ํผ๋๋ฐฑ์ ๋ฐ๊ธฐ ์ด๋ ต์ต๋๋ค. โํผ๋๋ฐฑ ๋ณด๋ด๊ธฐ(send feedback)โ์ ๊ฐ์ ๋๊ตฌ๋ฅผ MCP์ ํฌํจํ์ฌ, ์ฌ์ฉ์๊ฐ โ์ด๊ฑฐ ์ ๋ง ์ธ๋ชจ์์์ดโ๋ผ๊ณ ๋งํ๋ฉด MCP ์๋ฒ๊ฐ ์ด๋ฅผ ์ธ์งํ๊ณ ๊ฐ์ ํ ์ ์๋๋ก ํด์ผ ํฉ๋๋ค.
- ๊ตฌ์ฒด์ ์ธ ์ค๊ณ ์์น:
- ๋๊ตฌ ์ ์ต์ํ: ๋๊ตฌ์ ๊ฐ์๋ฅผ ์๋์ ์ผ๋ก ์ ๊ฒ ์ ์งํด์ผ ํฉ๋๋ค.
- ์ ํํ ๋๊ตฌ ์ด๋ฆ ๋ฐ ์ค๋ช : ๋๊ตฌ ์ด๋ฆ๊ณผ ์ค๋ช ์ ๋งค์ฐ ์ ํํ๊ณ ๊ตฌ์ฒด์ ์ผ๋ก ์์ฑํด์ผ ํฉ๋๋ค. (์ด๊ฒ์ด ๋งค์ฐ ์ด๋ ต์ต๋๋ค.)
- ๊ฐ๊ฒฐํ ์ ๋ ฅ ์คํค๋ง: ์ ๋ ฅ ํ๋ผ๋ฏธํฐ(parameter)์ ์๋ฅผ ์ ๊ฒ ํ๊ณ , ๊ฐ๊ฒฐํ๋ฉด์๋ ์ถฉ๋ถํ ์ค๋ช ๋์ด์ผ ํฉ๋๋ค.
- ์ ํํ ์๋ต ๋ฐ์ดํฐ: ๋ชจ๋ธ์ด ํ์๋ก ํ๋ ์ต์ํ์ ๋ฐ์ดํฐ๋ง ์๋ต์ผ๋ก ๋ฐํํด์ผ ํฉ๋๋ค. (์ด๋ค ๋ฐ์ดํฐ๊ฐ ํ์ํ ์ง ๋ฏธ๋ฆฌ ์๊ธฐ ์ด๋ ต๊ธฐ ๋๋ฌธ์ ์ด ์ญ์ ์ด๋ ต์ต๋๋ค.) Stainless๋ ๋ชจ๋ธ์ JSON ๋ฐ์ดํฐ๋ฅผ ํํฐ๋งํ๋ JQ ํํฐ(JQ filter)๋ฅผ ์ ๊ณตํ๋ ๊ธฐ์ ์ ์ฌ์ฉํ๊ธฐ๋ ํฉ๋๋ค.
๋๊ตฌ ํ์(Tool Browsing) ๋ฌธ์ ์ ๋์ ๋ชจ๋(Dynamic Mode)
๋๊ตฌ ํ์ ๋ฌธ์ ๋ ๋ฐฉ๋ํ ๋๊ตฌ ๋ชฉ๋ก ์ค์์ AI๊ฐ ์ ์ ํ ๋๊ตฌ๋ฅผ ์ฐพ์๋ด๋ ์ด๋ ค์์ ์๋ฏธํฉ๋๋ค. ์ด์ ๋ํ ํ ๊ฐ์ง ์ ๊ทผ ๋ฐฉ์์ Stainless๊ฐ ์ ๊ณตํ๋ โ๋์ ๋ชจ๋(dynamic mode)โ์ ๋๋ค. API์ ํฌ๊ธฐ์ ์๊ด์์ด ์ธ ๊ฐ์ง ๊ธฐ๋ณธ ๋๊ตฌ๋ง ๋ ธ์ถํฉ๋๋ค.
list endpoints: ์ฌ์ฉ ๊ฐ๋ฅํ ์๋ํฌ์ธํธ ๋ชฉ๋ก์ ์กฐํํฉ๋๋ค.get endpoint: ํน์ ์๋ํฌ์ธํธ์ ๋ํ ์ ๋ณด๋ฅผ ํ์ตํฉ๋๋ค.execute endpoint: ์๋ํฌ์ธํธ๋ฅผ ์คํํฉ๋๋ค.
์ด ๋ฐฉ์์ ์ปจํ ์คํธ ์๋์ฐ ๋ฌธ์ ๋ฅผ ํจ๊ณผ์ ์ผ๋ก ํด๊ฒฐํ ์ ์์ง๋ง, ํ๋์ ์์ ์ ์ํํ๊ธฐ ์ํด ๋ชจ๋ธ์ด ์ธ ๋ฒ์ โํด(turn)โ์ ๊ฑฐ์ณ์ผ ํ๋ฏ๋ก ์๋๊ฐ ๋๋ ค์ง๊ณ ๋น์ฉ์ด ๋ ๋ฐ์ํ ์ ์์ต๋๋ค. ๋ํ, ๋๊ตฌ๋ค์ด ์ง์ ๋ก๋๋๋ ๋ฐฉ์์ด ์๋๋ฏ๋ก ์ฑ๋ฅ ์ ํ๊ฐ ๋ฐ์ํ ์๋ ์์ต๋๋ค.
์๋ ์ค ๋ซ๋ ์ด์ AI ํ์ฉ๋ฒ: ๊ฐ์ธ์ ์ธ MCP ๊ฒฝํ
์๋ ์ค๋ Stainless์ MCP ์๋ฒ๋ฅผ ์ฝ๋ฉ ์์ ๋ณด๋ค๋ ๋น์ฆ๋์ค ์ธก๋ฉด์์ ํ์ฉํ๊ณ ์๋ค๊ณ ๋งํฉ๋๋ค. ๊ทธ๋ Notion, HubSpot, Gong ๋ฑ์ MCP ์๋ฒ์ ์ฝ๊ธฐ ์ ์ฉ ๋ฐ์ดํฐ๋ฒ ์ด์ค MCP ์๋ฒ๋ฅผ ์ฌ์ฉํ์ฌ ๋ค์๊ณผ ๊ฐ์ ์ง๋ฌธ์ ๋์ง๋๋ค.
โ์ง๋์ฃผ Stainless์ ๊ฐ์ ํ ํฅ๋ฏธ๋ก์ด ๊ณ ๊ฐ์ ๋๊ตฌ์ธ๊ฐ?โ
AI๋ PostgreSQL ๋ฐ์ดํฐ๋ฒ ์ด์ค๋ฅผ ์ฟผ๋ฆฌํ๊ณ , HubSpot์์ ๊ต์ฐจ ์ฐธ์กฐํ๋ฉฐ, Notion์์ ๋ ธํธ๋ฅผ ์ฐพ๊ณ , Gong์์ ๋ น์ทจ๋ก์ ํ์ธํ์ฌ ์ด์ ๋ํ ์์ธํ ์ ๋ณด๋ฅผ ์ ๊ณตํฉ๋๋ค. ์์ง์ โ์คํ ๋ชจ๋โ์ ๊ฐ๊น๊ณ ๊ฐ๋ ์ฐ๊ฒฐ์ด ๋๊ธฐ๋ ๋ฑ์ โ์ฌ์ํ ๋ถํธํจ(paper cuts)โ์ด ์์ง๋ง, ๊ทธ ์ ์ฌ๋ ฅ์ ์์ฒญ๋๋ค๊ณ ํฉ๋๋ค.
์๋ ์ค๋ ํนํ AI๋ฅผ ํ์ฉํ์ฌ โAI๋ฅผ ์ํ ๋ ธํธ ์์งโ ๊ดํ์ ๋งค์ฐ ์ ์ฉํ๋ค๊ณ ๊ฐ์กฐํฉ๋๋ค. ๊ทธ๋ ํน๋ณํ Git ๋ ํฌ์งํ ๋ฆฌ(repository)์ โ๋ ธํธ ํด๋โ๋ โ์ฐ๊ตฌ ํด๋โ๋ฅผ ๋ง๋ค์ด, AI์๊ฒ ํฅ๋ฏธ๋ก์ด ๊ณ ๊ฐ ์ธ์ฉ๋ฌธ์ด ๋ฐ๊ฒฌ๋๋ฉด ์ ์ฒด ์ถ์ฒ์ ํจ๊ป ์ด ํด๋์ ์ ์ฅํ๋๋ก ์ง์ํฉ๋๋ค. ์ด๋ ๊ฒ ํ๋ฉด ๋ค์๋ฒ์ ์ง๋ฌธํ ๋๋ง๋ค MCP ์๋ฒ๋ฅผ ๋ค์ ๊ฒ์ํ ํ์ ์์ด AI๊ฐ ์บ์๋ ์ ๋ณด๋ฅผ ํ์ฉํ ์ ์์ต๋๋ค. Claude Code์ ๊ฐ์ ๋๊ตฌ๋ฅผ ์ฌ์ฉํ์ฌ ์ด๋ฌํ ์์ ์ ์ํํ๋ฉฐ, ํ์ฌ๋ ๊ตฌ์กฐ๊ฐ ๋ค์ ํผ๋์ค๋ฝ๋๋ผ๋ AI๊ฐ ๋น์ ํ ๋ฐ์ดํฐ๋ฅผ ์ ์ฒ๋ฆฌํ๊ธฐ ๋๋ฌธ์ ํฐ ๋ฌธ์ ๋ ์๋ค๊ณ ํฉ๋๋ค.
๊ทธ๋ ์ด ๋ ํฌ์งํ ๋ฆฌ์ ๊ณ ๊ฐ ์ธ์ฉ๋ฌธ ์ธ์๋ SQL ์ฟผ๋ฆฌ(SQL queries)๋ฅผ ์ ์ฅํฉ๋๋ค. ์๋ฅผ ๋ค์ด, ์ด์ฌํ(board) ์ค๋น๋ฅผ ์ํด โ์ง๋ 3๊ฐ์๊ฐ XYZ ์งํ์ ์๋ณ ์ฑ์ฅ๋ฅ ์ ์ด๋ ๋์?โ๋ผ๊ณ ์ง๋ฌธํ๊ณ , AI์ ๋ฐ๋ณต์ ์ธ ๋ํ๋ฅผ ํตํด ํน์ ๋น์ฆ๋์ค ๋งฅ๋ฝ์ ๋ฐ์ํ SQL ์ฟผ๋ฆฌ๋ฅผ ์์ฑํ ํ, ์ด๋ฅผ โ๋ถ์ ํด๋โ์ ์ ์ฅํ์ฌ ๋ค์์ ์ฌ์ฌ์ฉํฉ๋๋ค.
๋ํ, Stainless ํ์ ๊ณ ๊ฐ ์ง์์๋ AI๋ฅผ ํ์ฉํ๊ณ ์์ต๋๋ค. ๊ณ ๊ฐ ๋ฌธ์๊ฐ ์ ์๋๋ฉด, AI(Claude Code)์๊ฒ ๋จผ์ ํด๊ฒฐ์ ์๋ํ๋๋ก ํฉ๋๋ค. 100% ๋๋ 50%์ ์ฑ๊ณต๋ฅ ์ ์๋์ง๋ง, ์ ๋ฐ์ ์ธ ํจ์จ์ฑ์ ๊ฐ์ ํ ๊ฐ๋ฅ์ฑ์ด ์๋ค๊ณ ๋ณด๊ณ ์คํ์ ์งํ ์ค์ ๋๋ค.
AI์ ๋ฏธ๋: ์ฌ์ด๋ณด๊ทธ์ ์ฝ๋ ์คํ
์๋ ์ค๋ ์์ด์ ํธ AI๋ฅผ ๋์ด AI์ ๋ฏธ๋์ ๋ํ ํฐ ๋น์ ์ ์ ์ํฉ๋๋ค. ๊ทธ๋ โAI์ ๋ฏธ๋๋ ์ฌ์ด๋ณด๊ทธ(cyborg)๊ฐ ๋ ๊ฒโ์ด๋ผ๊ณ ๋งํฉ๋๋ค. ์ฌ๊ธฐ์ โ์ฌ์ด๋ณด๊ทธโ๋ ๋ถ๋ถ์ ์ผ๋ก๋ GPT์ ๊ฐ์ ์ ๊ฒฝ๋ง LLM AI์ด๊ณ , ๋ถ๋ถ์ ์ผ๋ก๋ ์ ํต์ ์ธ CPU ๊ธฐ๋ฐ ์ํํธ์จ์ด์ธ ๊ธฐ๊ณ๋ฅผ ์๋ฏธํฉ๋๋ค. ์ด๋ฌํ ์ฌ์ด๋ณด๊ทธ AI๋ ํฌ๊ฒ ๋ ๊ฐ์ง ๋ฐฉ์์ผ๋ก ์ ๊ฐ๋ ๊ฒ์ด๋ผ๊ณ ํฉ๋๋ค. ํ๋๋ ๋ฐฉ๊ธ ์ธ๊ธํ ์ผํ์ฑ ์ด์ ์ฌ์ฉ ์ฌ๋ก์ด๊ณ , ๋ค๋ฅธ ํ๋๋ ํ๋ก๋์ ์ํํธ์จ์ด(production software)์ ๋๋ค.
์ฝ๋ ์คํ ๋๊ตฌ(Code Execution Tool)์ ๋ถ์
์๋ ์ค๋ ๋ชจ๋ธ์ด ์๋ง์ ๋๊ตฌ๋ฅผ ๊ฐ์ง๋ ๋์ , ๋จ ๋ ๊ฐ์ง ๋๊ตฌ๋ง ๊ฐ์ง๋ ๋ฐฉํฅ์ผ๋ก ๋์๊ฐ ๊ฒ์ด๋ผ๊ณ ์์ธกํฉ๋๋ค.
- ์ฝ๋ ์คํ ๋๊ตฌ (Execute Code Tool): ๋ชจ๋ธ์ ํ
์คํธ ์์์ ํ์
์คํฌ๋ฆฝํธ(TypeScript)์ ๊ฐ์ ์ฝ๋๋ฅผ ์์ฑํ๊ณ , ์ด ์ฝ๋๋ API์ ํ์
์คํฌ๋ฆฝํธ SDK๋ฅผ ์ฌ์ฉํ์ฌ
Stripe.transactions.list๋๋Stripe.refunds.create์ ๊ฐ์ API ํธ์ถ์ ์ํํฉ๋๋ค. ๋ชจ๋ธ์ ์ฝ๋ ์์ฑ์ ๋งค์ฐ ๋ฅ์ํ๋ฉฐ, SDK์ API๊ฐ ์ ๊ตฌ์ฑ๋์ด ์๋ค๋ฉด ํจํด์ ํตํด ์ฝ๊ฒ ํ์ฅํ ์ ์์ต๋๋ค. - ๋ฌธ์ ๊ฒ์ ๋๊ตฌ (Search Docs Tool): ๋ชจ๋ธ์ด ํ์คํ์ง ์๊ฑฐ๋ ์ฒซ ์๋์์ ์ค๋ฅ๊ฐ ๋ฐ์ํ๋ฉด, ์ด ๋๊ตฌ๋ฅผ ์ฌ์ฉํ์ฌ ๋ฌธ์๋ฅผ ๊ฒ์ํ๊ณ ์ง๋ฌธํฉ๋๋ค.
์ด๋ฌํ ๋ฐฉ์์ ์์ ์ธ๊ธ๋ ์ปจํ ์คํธ ์๋์ฐ ๋ฌธ์ ๋ฅผ ๊ทน์ ์ผ๋ก ํด๊ฒฐํฉ๋๋ค. ์ด๊ธฐ ์ปจํ ์คํธ ์ํฅ์ 1,000 ํ ํฐ ๋ฏธ๋ง์ผ๋ก ๋งค์ฐ ์ ํ์ ์ด๋ฉฐ, ํ์ด์ง๊ฐ ๋งค๊ฒจ์ง(paginated) ๋ชฉ๋ก ์์ฒญ๊ณผ ๊ฐ์ ์์ ์ ์ปจํ ์คํธ ์ํฅ์ ๊ฑฐ์ 0์ ๊ฐ๊น์ต๋๋ค. ๋ชจ๋ธ์ ์ฌ๋ฌ ๋ฐ๋ณต ์์ ์ ์ํํ ํ ์ต์ข ์ ์ผ๋ก ํ์ํ ์ ๋ณด(์: ๊ณ ๊ฐ ID, ๊ฑฐ๋ ID, ํ๋ถ ID)๋ง ์ฝ์์ ๊ธฐ๋ก(console.log)ํ ๊ฒ์ด๊ณ , ์ด๋ก ์ธํด ๋์์ค๋ ์ปจํ ์คํธ๋ ๋จ 10์ค ์ ๋์ ํ ์คํธ๋ก ๋งค์ฐ ์ต์ํ๋ฉ๋๋ค. ๋ํ, ์ด๋ฌํ ์ฝ๋๋ ๋ชจ๋ธ๊ณผ์ ์๋ณต(round trip) ์์ด CPU ๊ธฐ๋ฐ ์๋ฒ์์ Stripe API ์์์ ์ด๊ณ ์์ผ๋ก ์คํ๋ ๊ฒ์ ๋๋ค.
์ ์ง์ ์ฝ๋๋ฅผ ์คํํ์ง ์๋๊ฐ?
์ผ๊ฐ์์๋ ๋ชจ๋ธ์ด ์ฝ๋๋ฅผ ์์ฑํ๋ฉด ์ฌ์ฉ์๊ฐ ์ง์ ์คํํ๋ฉด ๋๋ ๊ฒ ์๋๋๋ ์๋ฌธ์ ์ ๊ธฐํ ์ ์์ต๋๋ค. ์๋ ์ค๋ ์ด์ ๋ํด ๋ค์๊ณผ ๊ฐ์ด ์ค๋ช ํฉ๋๋ค.
- ๋ผ์ด๋ธ๋ฌ๋ฆฌ ๋ฌธ์ : LLM์ ๋ผ์ด๋ธ๋ฌ๋ฆฌ์ ์์ ํ๊ณ , ํน์ ๋ฒ์ ์ ๋ผ์ด๋ธ๋ฌ๋ฆฌ๋ฅผ ์ ํํ ์ฌ์ฉํ๋ฉฐ, API์ ํ๊ฐ(hallucination)์ ํผํ๊ณ , ์ค๋ฅ ๋ฐ์ ์ ๋ฐ๋ณต์ ์ผ๋ก ์์ ํ๋ ๋ฐ ์ด๋ ค์์ ๊ฒช์ต๋๋ค. npm์ด๋ ํ์ด์ฌ ํจํค์ง ์ธ๋ฑ์ค์ ๋ผ์ด๋ธ๋ฌ๋ฆฌ๋ฅผ ์๋ฒฝํ๊ฒ ์ฌ์ฉํ ์ ์๋ค๋ฉด, ๊ฒฐ๊ตญ ์์ HTTP API๋ฅผ ์ง์ ํธ์ถํด์ผ ํฉ๋๋ค.
- OpenAPI ์คํ์ ๋ฐฉ๋ํจ: ์์ HTTP API๋ฅผ ์ฌ์ฉํ๋ ค๋ฉด ๋ฐฉ๋ํ OpenAPI ์คํ ๋ฌธ์๋ฅผ ์ดํดํด์ผ ํ๋๋ฐ, ์ด๋ ์ปจํ ์คํธ ์๋์ฐ ๋ฌธ์ ๋ฅผ ๋ค์ ์ผ๊ธฐํฉ๋๋ค.
- ํ์ ๊ฒ์ฌ(Type Checking) ๋ฐ ์ค๋ฅ ์ฒ๋ฆฌ: ์ฝ๋ ์คํ ๋๊ตฌ๋ ํ์ ์ฒด์ปค(type checker)๋ฅผ ์คํํ์ฌ โstripe.transactions.list๋ ์กด์ฌํ์ง ์๋๋ค. payment intents, orders, balance transactions ์ค ์ด๋ ๊ฒ์ ์ํ๋๊ฐ?โ์ ๊ฐ์ด ๋ชจ๋ธ์ ์ค๋ฅ๋ฅผ ์ฆ์ ๊ฐ์งํ๊ณ , ๊ด๋ จ ๋ฌธ์์ AI ๊ธฐ๋ฐ ์ ์์ ์ ๊ณตํ ์ ์์ต๋๋ค. ์ด๋ฌํ ๊ธฐ๋ฅ์ API ์ ๊ณต์๊ฐ ์ ๊ตฌ์ถํด์ผ ํ๋ฉฐ, ์ด๋ ์ ์ฒด ๋ํ์ ์ปจํ ์คํธ์ ์ฝ๋งค์ด์ง ์๋ ์๋ธ ์์ด์ ํธ(sub agent)์ ์ํด ์ํ๋ ์ ์์ต๋๋ค.
๋ณด์ ๋ชจ๋ธ: API ๋ ์ด์ด์์ ํด๊ฒฐ
๋ณด์ ๋ฌธ์ ์ ๋ํด์๋ ๊ถ๊ทน์ ์ผ๋ก API ๋ ์ด์ด ์์ฒด์์ ํด๊ฒฐ๋์ด์ผ ํ๋ค๊ณ ์๋ ์ค๋ ์ฃผ์ฅํฉ๋๋ค. ํ์ฌ๋ MCP๋ฅผ ํตํด ๋ ธ์ถ๋๋ ๊ธฐ๋ฅ์ ์ ํํจ์ผ๋ก์จ ๋ณด์์ ๊ตฌํํ๋ ค ํ์ง๋ง, ๊ฒฐ๊ตญ ๋ด๋ถ์ ์ผ๋ก๋ API์ ๋ชจ๋ ๊ธฐ๋ฅ์ ํธ์ถํ ์ ์์ต๋๋ค. ์ฌ๋ฐ๋ฅธ ์ ๊ทผ ๋ฐฉ์์ granular permission(์ธ๋ถํ๋ ๊ถํ)๊ณผ ์ ์ ํ ์ค์ฝํ(scope)๋ฅผ ๊ฐ์ง OAuth(์ค์ด์ฐ)๋ฅผ ์ฌ์ฉํ๋ ๊ฒ์ ๋๋ค. ์ด๋ ๊ตฌ์ถํ๊ธฐ ์ด๋ ต์ง๋ง, ๋ณด์์ด ์ ๋๋ก ์๋ํด์ผ ํ๋ ์ฌ๋ฐ๋ฅธ ๊ณ์ธต์ ๋๋ค.
๋ธ๋ ์ธ ํธ๋ฌ์คํธ(Brain Trust): AI ์ต์ ๋ฒ๋น๋ฆฌํฐ ํ๋ซํผ
๋ง์ ๊ธฐ์ ๋ค์ด AI๋ฅผ ํ๋ก๋์ ํ๊ฒฝ์ ๋ฐฐํฌํ๋ฉฐ ์์ฐ์ฑ์ ๋์ด๊ณ ์์ง๋ง, ๋์์ ๋ถ์๊ฐ๋ ์ปค์ง๊ณ ์์ต๋๋ค. ํ๋กฌํํธ(prompt)๋ฅผ ์์ ํ๊ณ , ๋ชจ๋ธ์ ๊ต์ฒดํ๊ณ , ํ๋ผ๋ฏธํฐ(parameter)๋ฅผ ์กฐ์ ํ๋ฉด ํ ์คํธ์์๋ ๋ฌธ์ ๊ฐ ์์ด ๋ณด์ด์ง๋ง, ๋ฉฐ์น ํ ๊ณ ๊ฐ ์ง์ ํฐ์ผ์ด ์์์ง๊ธฐ ์์ํฉ๋๋ค. AI๊ฐ ์์์น ๋ชปํ ๋ต๋ณ์ ์ ๊ณตํ๊ณ , ์ธ์ ์ ์ด๋ฐ ๋ฌธ์ ๊ฐ ๋ฐ์ํ๋์ง ์ ์ ์์ต๋๋ค.
๋ธ๋ ์ธ ํธ๋ฌ์คํธ(Brain Trust)๋ ์ด๋ฌํ ๋ฌธ์ ๋ฅผ ํด๊ฒฐํ๋ AI ์ต์ ๋ฒ๋น๋ฆฌํฐ(observability) ํ๋ซํผ์ ๋๋ค. ์ด๋ ํ๊ฐ(eval)์ ๊ด์ฐฐ ๊ฐ๋ฅ์ฑ์ ํ๋์ ์ํฌํ๋ก์ฐ(workflow)๋ก ํตํฉํ์ฌ, ํ๋ก๋์ ํ๊ฒฝ์์ ์ค์ ๋ก ๋ฐ์ํ ์ผ์ ํ์ธํ๊ณ ๋ณ๊ฒฝ ์ฌํญ์ด ๊ฐ์ ๋์๋์ง ์ ํ๋์๋์ง ์ธก์ ํ ์ ์๋๋ก ๋์ต๋๋ค. ํธ๋ ์ด์ค(trace)๋ ์ ์ฒด ์คํ ๊ฒฝ๋ก๋ฅผ ๋ณด์ฌ์ฃผ๊ณ , ํ๊ฐ๋ โ์ข์ ๊ฒฐ๊ณผโ๊ฐ ๋ฌด์์ธ์ง ์ ์ํ๋ฉฐ, ์คํ์ ๋ฐฐํฌ ์ ์ ํ๋กฌํํธ์ ๋ชจ๋ธ์ ๋น๊ตํ ์ ์๊ฒ ํฉ๋๋ค.
ํ๋ก๋์ ํธ๋ ์ด์ค๋ ํ๊ฐ ๋ฐ์ดํฐ ์ธํธ์ ์ง์ ๊ณต๊ธ๋์ด ๋ชจ๋ ์คํจ๋ฅผ ํ ์คํธ ์ฌ๋ก๋ก ๋ง๋ญ๋๋ค. ์ด๋ฅผ ํตํด ์ฌ์ฉ์์๊ฒ ๋๋ฌํ๊ธฐ ์ ์ CI(Continuous Integration) ๋จ๊ณ์์ ํ๊ท(regression)๋ฅผ ์ก์๋ผ ์ ์์ต๋๋ค. Notion, Stripe, Zapier, Vercel, RAMP์ ๊ฐ์ ๊ธฐ์ ๋ค์ด ๋ธ๋ ์ธ ํธ๋ฌ์คํธ๋ฅผ ์ฌ์ฉํ์ฌ ๊ณ ํ์ง AI๋ฅผ ๋๊ท๋ชจ๋ก ๋ฐฐํฌํ๊ณ ์์ต๋๋ค. ๋ธ๋ ์ธ ํธ๋ฌ์คํธ๋ ํ์ด์ฌ, ํ์ ์คํฌ๋ฆฝํธ, ๊ณ , ๋ฃจ๋น, C์ฉ SDK๋ฅผ ์ ๊ณตํ๋ฉฐ, ํน์ ํ๋ ์์ํฌ๋ ๋ฒค๋์ ์ข ์๋์ง ์์ต๋๋ค. SOCK 2 Type 2 ์ธ์ฆ์ ๋ฐ์๊ณ GDPR ๋ฐ HIPAA๋ฅผ ์ค์ํฉ๋๋ค. ๋ ์์ธํ ์ ๋ณด๋ brain.dev์์ ํ์ธํ ์ ์์ต๋๋ค.
๊ฒฐ๋ก : AI์ ์ธ๊ฐ, ์ฝ๋์ ๊ณต์กด
์๋ ์ค ๋ซ๋ ์ด์์ ๋ํ๋ API๋ฅผ ํตํ ์ปดํจํฐ ๊ฐ์ ์ํต์ด ์ด๋ป๊ฒ AI ์๋์ ์งํํ๊ณ ์๋์ง์ ๋ํ ๊น์ ํต์ฐฐ์ ์ ๊ณตํ์ต๋๋ค. ์ธํฐ๋ท์ ๊ธฐ์กด ์ํคํ ์ฒ๊ฐ AI ์๋์ ๋ง์ถฐ ์ฌํธ๋์ด์ผ ํ๋ฉฐ, MCP๊ฐ ๊ทธ ํต์ฌ์ ์ธ ์ญํ ์ ํ ๊ฒ์ด๋ผ๋ ์ ์ ๋ถ๋ช ํฉ๋๋ค.
ํ์ฌ MCP๋ ์ปจํ ์คํธ ์๋์ฐ์ ํ๊ณ์ ๋ณต์กํ ๋๊ตฌ ์ค๊ณ, ๋ณด์ ๋ฌธ์ ๋ฑ ์ฌ๋ฌ ๋๊ด์ ๋ถ๋ชํ ์์ง๋ง, ์๋ ์ค๊ฐ ์ ์ํ๋ โ์ฌ์ด๋ณด๊ทธโ ๋น์ ๊ณผ โ์ฝ๋ ์คํ ๋๊ตฌโ๋ ์ด๋ฌํ ๋ฌธ์ ๋ฅผ ํด๊ฒฐํ ํ์ ์ ์ธ ๋์์ด ๋ ์ ์์ต๋๋ค. LLM์ด ์ง์ ์ฝ๋๋ฅผ ์์ฑํ๊ณ ์ด๋ฅผ API ์ ๊ณต์๊ฐ ์์ ํ๊ณ ํจ์จ์ ์ผ๋ก ์คํํ๋ ๋ฐฉ์์ AI๊ฐ ์ธํฐ๋ท์ ๋ฐฉ๋ํ ๊ธฐ๋ฅ์ ํ์ฉํ๋ ๋ฐ ์์ด ๊ฒ์ ์ฒด์ธ์ ๊ฐ ๋ ๊ฒ์ ๋๋ค.
Stainless๋ ์ด๋ฌํ ๋ฏธ๋๋ฅผ ๊ตฌ์ถํ๋ ์ต์ ์ ์ ์ ์์ต๋๋ค. AI์ ์ธ๊ฐ, ๊ทธ๋ฆฌ๊ณ ์ฝ๋๊ฐ ์ ๊ธฐ์ ์ผ๋ก ํ๋ ฅํ๋ฉฐ ์์ํ ์ ์์๋ ์๋ํ์ ํจ์จ์ฑ์ ์คํํ โ์ฌ์ด๋ณด๊ทธโ ์๋๋ ์๊ฐ๋ณด๋ค ๋ ๋น ๋ฅด๊ฒ ๋ค๊ฐ์ฌ์ง ๋ชจ๋ฆ ๋๋ค. ์ด ์ฌ์ ์์ Stainless์ ๊ฐ์ ๊ธฐ์ ๋ค์ ์ญํ ์ ๋์ฑ ์ค์ํด์ง ๊ฒ์ ๋๋ค.
โWas anything good released at Google I/O 2026?โ โ How I AI ๊ธฐ๋ฐ ๊ธฐ์ฌ ์๋ณธ ์์ ๋ณด๊ธฐ
Google I/O 2026, ๊ณผ์ฐ ๊ธฐ๋๋งํผ ํ์ ์ ์ด์๋? AI ์๋์ ๋ช ๊ณผ ์
์๋ก : AI ํ์ ์ ํฅํ ๊ตฌ๊ธ์ ์ผ์ฌ
โHow I AIโ ์ฑ๋์ ํ๋ก๋ํธ ๋ฆฌ๋์ด์ AI ์ ๋ฌธ๊ฐ์ธ ํด๋ผ๋ผ๋น(Claraveville)์ ์ด๋ฒ Google I/O ํ์ฌ๋ฅผ ํตํด ๊ณต๊ฐ๋ ๊ตฌ๊ธ์ ์๋ง์ AI ์ ์ ํ๊ณผ ๊ธฐ๋ฅ๋ค์ ์ฌ์ธต์ ์ผ๋ก ๋ถ์ํ์ต๋๋ค. ๋งค๋ ๊ฐ๋ฐ์๋ค์ ๊ธฐ๋๋ฅผ ํ๋ชธ์ ๋ฐ๋ ๊ตฌ๊ธ์ ์ฃผ๋ ฅ ์ด๋ฒคํธ์ธ Google I/O๋ ์ฌํด๋ ์ด๊น์์ด ๋ค์ํ ๊ธฐ์ ์ ๋ฐํ์ ํจ๊ป ์๋น์์ฉ ํฌ๋ฆฌ์์ดํฐ๋ธ ๊ธฐ๋ฅ์ ์ ์ฌ๋ ฅ์ ์ ๋ณด์์ต๋๋ค. ๊ณผ์ฐ ๊ตฌ๊ธ์ด ์ ์ํ ์๋ก์ด ๊ธฐ๋ฅ๋ค์ด ๋จ์ํ ๊ณผ๋๊ด๊ณ ๋ฅผ ๋์ด ์ค์ง์ ์ธ ํ์ ์ ๊ฐ์ ธ์ฌ ์ ์์์ง, ํด๋ผ๋ผ๋น์ ์์ ์ผ๋ก ๊ธฐ์ ์ ์ธก๋ฉด๋ถํฐ ํฅ๋ฏธ๋ก์ด ์๋น์ ๊ธฐ๋ฅ๊น์ง ํจ๊ป ์ดํด๋ณด๊ฒ ์ต๋๋ค.
์ด๋ฒ Google I/O์ ํต์ฌ์ ํฌ๊ฒ ์ธ ๊ฐ์ง์์ต๋๋ค. ์ฒซ์งธ, ๊ฐ๋ฐ์๋ฅผ ์ํ ๋ ๋น ๋ฅด๊ณ ๊ฐ๋ ฅํ AI ๋ชจ๋ธ๊ณผ ๊ฐ๋ฐ ๋๊ตฌ์ ๋ฐ์ . ๋์งธ, ๊ตฌ๊ธ ์ํฌ์คํ์ด์ค(Google Workspace)์์ ์ฐ๋์ ํตํ ์ผ๋ฐ ์ฌ์ฉ์์ ๊ธฐ์ ์ ์์ฐ์ฑ ํฅ์. ์ ์งธ, ์ด๋ฏธ์ง์ ๋น๋์ค ์์ฑ, ๋์์ธ ๋๊ตฌ ๋ฑ ์ฐฝ์์ ์์ญ์์์ AI ํ์ฉ ํ๋์ ๋๋ค. ํ์ง๋ง ํด๋ผ๋ผ๋น์ ์ด๋ฌํ ์ผ์ฌ ์ฐฌ ๋ฐํ๋ค ์์ ์์ง์ โ๋ฏธ์์ ์กฐ๊ฐโ๋ค์ด ์กด์ฌํ๋ฉฐ, ๋ช๋ช ๊ธฐ๋ฅ๋ค์ ์ค์ ๊ตฌํ์ ์์ด ์์ฌ์์ ๋จ๊ฒผ๋ค๊ณ ์ง์ ํฉ๋๋ค.
1๋ถ: ๊ฐ๋ฐ์๋ฅผ ์ํ ์ฌ์ธต ๊ธฐ์ ํ์ โ ๋ชจ๋ธ ๋ฐ ๊ฐ๋ฐ ๋๊ตฌ
Gemini 3.5: ์๋์ ์ง๋ฅ์ ๊ฒฐํฉ
๊ตฌ๊ธ์ ์ด๋ฒ I/O์์ ์๋ก์ด ์ ๋ฏธ๋ 3.5(Gemini 3.5) ๋ชจ๋ธ๊ตฐ์ ๊ณต๊ฐํ์ผ๋ฉฐ, ํนํ ๊ฐ์ฅ ๋น ๋ฅด๊ณ ์ค๋งํธํ ์ฝ๋ฉ ๋ชจ๋ธ์ธ ์ ๋ฏธ๋ 3.5 ํ๋์(Gemini 3.5 Flash)๋ฅผ ๊ฐ์กฐํ์ต๋๋ค. ํด๋ผ๋ผ๋น์ ์ด ๋ชจ๋ธ์ด ๊ฒฝ์ ๋ชจ๋ธ์ธ ํด๋ก๋ 5.5 ์คํธ์ค(Claude 5.5 Opus)๋ GPT-4์ ๊ฐ์ ์ต์์ ์ฝ๋ฉ ๋ชจ๋ธ์ ์ง๋ฅ์ ํ์ ํ๋ฉด์๋, ์ฒ๋ฆฌ ์๋๋ ๋ฌด๋ ค 4๋ฐฐ๋ ๋น ๋ฅด๋ค๋ ์ ์ ์ฃผ๋ชฉํ์ต๋๋ค. ๊ตฌ๊ธ์ ๋ฒค์น๋งํฌ์ ๋ฐ๋ฅด๋ฉด, ์ ๋ฏธ๋ 3.5 ํ๋์๋ ์ฝ๋ฑ์ค(CodeX) 47๊ณผ ์ ์ฌํ ์ค๋งํธํจ์ 3.1 ํ๋์ ๋ชจ๋ธ ์์ค์ ์๋๋ก ์ ๊ณตํ์ฌ, โ์ด๊ณ ์-์ด๊ณ ์ง๋ฅโ์ด๋ผ๋ ๋ ๋ง๋ฆฌ ํ ๋ผ๋ฅผ ์ก์๋ค๋ ํ๊ฐ์ ๋๋ค.
ํนํ ๊ตฌ๊ธ์ ์ ๋ฏธ๋ 3.5 ๋ชจ๋ธ์ โ์์ด์ ํธ(Agent) ๊ธฐ๋ฅโ์ ์ง์คํ๊ณ ์์ต๋๋ค. ์ด๋ AI๊ฐ ๋ณต์กํ ์์ ์ ์ค์ค๋ก ๊ณํํ๊ณ ์คํํ๋ฉฐ, ํ์์ ๋ฐ๋ผ ๋ค๋ฅธ AI๋ฅผ ํธ์ถํ๋ ๋ฅ๋ ฅ์ ์๋ฏธํฉ๋๋ค. ํด๋ผ๋ผ๋น์ ๊ตฌ๊ธ์ด AI ์์ด์ ํธ ๋ถ์ผ์ ์ ๋ ฅ์ ๋คํ๊ณ ์์ผ๋ฉฐ, ์ด๋ ๊ฒฝ์์ฌ(์คํธ๋กํฝ(Anthropic), ์คํAI(OpenAI))์ ๋ฐ์์ทจ๋ฅผ ๋ฐ๋ผ์ก๋(Catch-up) ํ๋ณด๋ก ๋ณด์ผ ์ ์์ง๋ง, ํ๋์ ๋ชจ๋ธ์ ์๋์ ์๋น์ ์นํ์ ์ธ ์ฐฝ์์ฑ์ด๋ผ๋ ๋ ๊ฐ์ง ์ฐจ๋ณ์ ์ ๊ฐ์ง๊ณ ์๋ค๊ณ ๋ถ์ํ์ต๋๋ค. ๋ํ ์ ๋ฏธ๋ ๋ชจ๋ธ์ ๊ฐ์ ์ธ โ๋ฉํฐ๋ชจ๋ฌ(Multimodal)โ ๋ฅ๋ ฅ, ์ฆ ํ์ผ, ๋น๋์ค ๋ฑ ๋ค์ํ ํํ์ ๋ฐ์ดํฐ๋ฅผ ์ฒ๋ฆฌํ๊ณ ๋ณํํ๋ ๋ฅ๋ ฅ์ ์ฌ์ ํ ๋ ๋ณด์ ์ด๋ผ๊ณ ๊ฐ์กฐํ์ต๋๋ค.
Anti-gravity 2.0: ๊ฐ๋ฐ ์์ฐ์ฑ์ ์ํ ์์ด์ ํธ IDE
๊ฐ๋ฐ์๋ฅผ ์ํ ๋๊ตฌ๋ก ๊ตฌ๊ธ์ ์์ด์ ํธ ๊ธฐ๋ฐ IDE(ํตํฉ ๊ฐ๋ฐ ํ๊ฒฝ)์ธ โ์ํฐ๊ทธ๋๋นํฐ(Anti-gravity)โ์ 2.0 ๋ฒ์ ์ ๊ณต๊ฐํ์ต๋๋ค. ์ด๋ ํด๋ก๋ ์ฝ๋(Claude Code)๋ ์ฝ๋ฑ์ค(CodeX)์ ๊ฐ์ ๊ฒฝ์ ๋๊ตฌ์์ ๋ณผ ์ ์์๋ ๊ฐ๋ ๋ค์ ๊ตฌ๊ธ๋ง์ ๋ฐฉ์์ผ๋ก ๊ตฌํํ ๊ฒ์ผ๋ก ํ๊ฐ๋ฉ๋๋ค.
์ฃผ์ ๊ธฐ๋ฅ:
- ํ๋ก์ ํธ(Projects): ํด๋ ๋จ์๋ก ์ ํ๋ ์์ ํ๊ฒฝ์ ์ ๊ณตํ์ฌ ํจ์จ์ ์ธ ์ฝ๋ ๊ด๋ฆฌ๊ฐ ๊ฐ๋ฅํฉ๋๋ค.
- ์์ฝ ์์ (Scheduled Tasks): ํน์ ํ๋กฌํํธ(Prompt)๋ฅผ ์ ํด์ง ์ค์ผ์ค์ ๋ฐ๋ผ ์ ๊ธฐ์ ์ผ๋ก ์คํํ ์ ์์ต๋๋ค.
- ์๋ธ ์์ด์ ํธ(Sub-agents): ๋ฉ์ธ ์์ด์ ํธ๊ฐ ํน์ ์์ ์ ์ํด ๋ณด์กฐ ์์ด์ ํธ๋ฅผ ์์ฑํ๊ณ ํ์ฉํ ์ ์์ผ๋ฉฐ, ํนํ ๋ธ๋ผ์ฐ์ ์๋ธ ์์ด์ ํธ๋ ์น ๊ธฐ๋ฐ ์์ ์ ์ ์ฉํฉ๋๋ค.
- ํ (Hooks): ์์ด์ ํธ์ ์์ ์ฃผ๊ธฐ(์: ๋๊ตฌ ์คํ, ์์ ์๋ฃ, ์ ์ธ์ ์์)์ ํน์ ์ด๋ฒคํธ์ ๊ฐ์ ํ์ฌ ๋ง์ถคํ ๋์์ ์ํํ ์ ์์ต๋๋ค.
- ๋ค์ดํฐ๋ธ ๊น(Native Git) ๋ฐ ๋ก์ปฌ ๊ฐ๋ฐ ํ๊ฒฝ: ์ฝ๋ ๋ฒ์ ๊ด๋ฆฌ ๋ฐ ๋ก์ปฌ ๊ฐ๋ฐ ํ๊ฒฝ๊ณผ์ ํตํฉ์ ๊ฐํํ์ต๋๋ค.
- ์ฌ๋์ ๋ช
๋ น์ด(Slash Commands):
/goal: ์์ด์ ํธ์๊ฒ ํน์ ๋ชฉํ๋ฅผ ๋ถ์ฌํ๊ณ , ๋ชฉํ ๋ฌ์ฑ๊น์ง ๋ฐ๋ณต์ ์ผ๋ก ์์ ์ ์ํํ๊ฒ ํฉ๋๋ค./grill me: ํด๋ก๋ ์ฝ๋์ ์ ์คํ ์ง๋ฌธ๊ณผ ๋ฌ๋ฆฌ, ์ด ๋ช ๋ น์ด๋ ์์ด์ ํธ๊ฐ ์ฌ์ฉ์์ ์๊ตฌ์ฌํญ์ ๋์ฑ ์ฒ ์ ํ๊ฒ ๊ฒ์ฆํ๊ณ ํต์ฌ์ ํ๊ณ ๋๋ โ๊ฐ๋ ๋์ ์ง๋ฌธโ์ ๋์ง๊ฒ ํฉ๋๋ค./schedule,/browser: ์์ฝ ์์ ๋ฐ ๋ธ๋ผ์ฐ์ ๊ธฐ๋ฐ ์์ ์ ์ํ ๋ช ๋ น์ด์ ๋๋ค.
ํด๋ผ๋ผ๋น์ ์ํฐ๊ทธ๋๋นํฐ CLI(Command Line Interface)๋ฅผ ํตํด ๋ธ๋ก๊ทธ ์์ฑ๊ธฐ API ๊ตฌ์ถ์ ์์ฐํ์ต๋๋ค. ์์ฒญ์ ๋ฐ๋ผ API ์๋ํฌ์ธํธ(Endpoint)์ ์ธ์ฆ, ์๊ฐ์ ์ํฌํ๋ก์ฐ(Workflow) ํธ๋ฆฌ๊ฑฐ ๊ธฐ๋ฅ ๋ฑ์ ๋น ๋ฅด๊ฒ ์์ฑํ๋ ๋ชจ์ต์ ๋ณด์ฌ์ฃผ์์ง๋ง, ํด๋ผ๋ผ๋น์ โ๊ธฐ๋ํ๋ ๋งํผ ํน๋ณํ ๋น ๋ฅด๋ค๋ ๋๋์ ๋ค์ง ์์๋คโ๊ณ ์ธ๊ธํ๋ฉฐ, โ๊ฒฝ์์ฌ ๊ธฐ๋ฅ์ ๋ฐ๋ผ์ก๋ ์์คโ์ด๋ผ๊ณ ํ๊ฐํ์ต๋๋ค. ๊ทธ๋ผ์๋ ์ ๋ฏธ๋ 3.5 ํ๋์ ๋ชจ๋ธ์ ๋น ๋ฅธ ์๋๊ฐ ๊ฒฐํฉ๋๋ค๋ฉด, ๊ฐ๋ฐ์๋ค์ด ๊ฐ๋จํ ์์ ์ ๋น ๋ฅด๊ฒ ์ฒ๋ฆฌํ๋ ๋ฐ ์ ์ฉํ๊ฒ ํ์ฉ๋ ์ ์์ ๊ฒ์ด๋ผ๊ณ ์ ๋งํ์ต๋๋ค.
2๋ถ: ์ผ๋ฐ ์ฌ์ฉ์ ๋ฐ ๊ธฐ์ ์ ์ํ AI ๊ฒฝํ ํ์ฅ
Google AI Studio: ์ํฌ์คํ์ด์ค ์ฐ๋ ๋ฐ Android ์ฑ ๊ฐ๋ฐ
๊ตฌ๊ธ์ ๋ก์ฐ์ฝ๋/๋ ธ์ฝ๋(Low-code/No-code) ๊ฐ๋ฐ ํ๋ซํผ์ธ โGoogle AI Studioโ์ ์ ๋ฏธ๋ 3.5 ํ๋์๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ํ ์๋ก์ด ๊ธฐ๋ฅ๋ค์ ์ถ๊ฐํ์ต๋๋ค. ๊ฐ์ฅ ์ฃผ๋ชฉํ ๋งํ ์ ์ ๊ตฌ๊ธ ์ํฌ์คํ์ด์ค ์ฑ(Google Workspace Apps)๊ณผ์ ํตํฉ์ ๋๋ค. ์ด์ ์ฌ์ฉ์๋ ์คํ๋ ๋์ํธ(Sheets)๋ฅผ ์ฝ๊ณ , Gmail ์ด์์ ์์ฑํ๋ฉฐ, ๋๋ผ์ด๋ธ(Drive)๋ฅผ ์ ๋ฆฌํ๊ณ , ์บ๋ฆฐ๋๋ฅผ ํ์ธํ๋ ๋ฑ ์ํฌ์คํ์ด์ค ๋ฐ์ดํฐ์ ์ฐ๊ฒฐ๋ AI ์ฑ์ ์ง์ ๊ตฌ์ถํ ์ ์๊ฒ ๋์์ต๋๋ค. ์ด๋ ๊ธฐ์ ๋ด๋ถ ์์ฐ์ฑ ๋๊ตฌ๋ ๊ฐ์ธ ๋น์ ์ฑ ๋ฑ ๋ค์ํ ํ์ฉ ์ฌ๋ก๋ฅผ ๊ฐ๋ฅํ๊ฒ ํ ์ ์ฌ๋ ฅ์ ๊ฐ์ง๋๋ค. ๋ํ, Google AI Studio ๋ด์์ ์๋๋ก์ด๋ ์ฑ์ ์์ฑํ ์ ์๋ ๊ธฐ๋ฅ๋ ์ถ๊ฐ๋์ด, ๋ชจ๋ฐ์ผ ์ฑ ๊ฐ๋ฐ์ ์ง์ ์ฅ๋ฒฝ์ ๋ฎ์ถ์์ต๋๋ค.
ํ์ง๋ง ํด๋ผ๋ผ๋น์ ์บ๋ฆฐ๋ ์ฐ๋ ์ฑ์ ์๋ํ๋ ๊ณผ์ ์์ ๊ธฐ๋ฅ์ ๋ํ ์ ๊ทผ์ฑ ๋ฌธ์ ๋ฅผ ๊ฒช์์ต๋๋ค. ์ ๋ฃ ๊ณ ๊ฐ์์๋ ๋ถ๊ตฌํ๊ณ ์ํฌ์คํ์ด์ค ํตํฉ ๊ธฐ๋ฅ์ ์ฐพ๊ฑฐ๋ ํ์ฑํํ ์ ์์์ผ๋ฉฐ, ์ด๋ ๊ตฌ๊ธ์ด ๋ฐํํ ๊ธฐ๋ฅ๋ค์ด ์์ง ๋ชจ๋ ์ฌ์ฉ์์๊ฒ ์ฆ์ ์ ๊ณต๋์ง ์๊ฑฐ๋, ์ฌ์ฉ ๋ฐฉ๋ฒ์ด ๋ช ํํ์ง ์๋ค๋ ์ ์ ์์ฌํฉ๋๋ค. ํด๋ผ๋ผ๋น์ โ์์ง์ ์ฝ์๋ ๋น์ ์ผ ๋ฟ, ์ค์ ๊ตฌํ๊น์ง๋ ์๊ฐ์ด ๊ฑธ๋ฆด ๊ฒโ์ด๋ผ๊ณ ์์ฌ์์ ํํ์ต๋๋ค.
๋ณต์กํ ๋ธ๋๋๋ช : ๊ตฌ๊ธ์ ์์
ํด๋ผ๋ผ๋น์ ์ด๋ฒ Google I/O ๋ฐํ๋ฅผ ํตํ์ด ๊ฐ์ฅ ํฐ ๋ถ๋ง ์ค ํ๋๋ก โ๋ณต์กํ๊ณ ์ผ๊ด์ฑ ์๋ ์ ํ ๋ฐ ๋ธ๋๋๋ช โ์ ๊ผฝ์์ต๋๋ค. ์ ๋ฏธ๋ 3.5 ํ๋์, ์ํฐ๊ทธ๋๋นํฐ IDE, ์ํฐ๊ทธ๋๋นํฐ CLI, Google AI Studio, ์๋น์์ฉ ์ ๋ฏธ๋(Gemini/Google AI), ์ด๋(Omni), ํ๋ก์ฐ(Flow), ํ๋ฉ๋ฆฌ(Pomelli), ์คํฐ์น(Stitch) ๋ฑ ๋๋ฌด๋ ๋ง์ ์ด๋ฆ๋ค์ด ํผ์ฌ๋์ด ์์ด ์ด๋ค ์ ํ์ด ์ด๋ค ๊ธฐ๋ฅ์ ํ๋์ง ํผ๋์ค๋ฝ๋ค๋ ์ง์ ์ ๋๋ค. ํด๋ผ๋ผ๋น์ ๊ตฌ๊ธ ํ์๊ฒ โํฌ๊ด์ ์ธ ๋ธ๋๋ ๋ถ์๊ณผ ๋จ์ผํ๋ ๋ธ๋๋ ์ ๋ต์ด ์ ์คํ๋คโ๊ณ ๊ฐ๋ ฅํ ์์ฒญํ์ต๋๋ค. ์ด๋ ์ฌ์ฉ์ ๊ฒฝํ๊ณผ ์ ํ ์ธ์ง์ ์์ด ์ค์ํ ๋ถ๋ถ์ด๋ฉฐ, ๊ตฌ๊ธ์ด ํด๊ฒฐํด์ผ ํ ์์ ๋ก ๋จ์์ต๋๋ค.
3๋ถ: ์ฐฝ์์ฑ์ ์ํ AI โ ์ด๋ฏธ์ง, ๋น๋์ค, ๋์์ธ
Gemini (์๋น์์ฉ): ์ฌ์ค๊ณ์ ์ด๋ฏธ์ง ์์ฑ์ ์งํ
์๋น์์ฉ AI ์ ํ์ธ ์ ๋ฏธ๋(Gemini)๋ ์๋ก์ด UI ๋์์ธ๊ณผ ํจ๊ป ์ฌ์ฉ์ ๊ฒฝํ์ ๋ํญ ๊ฐ์ ํ์ต๋๋ค. ํนํ ์ด๋ฏธ์ง ์์ฑ ๊ธฐ๋ฅ์ ๋๋ ธ ๋ฐ๋๋(Nano Banana) ๋ชจ๋ธ์ ๊ธฐ๋ฐ์ผ๋ก ๋ค์ํ ํ ํ๋ฆฟ๊ณผ ๊ธฐ๋ฅ์ ์ ๊ณตํฉ๋๋ค. ํด๋ผ๋ผ๋น์ ์์ ์ ์ฌ์ง์ ํ์ฉํด โํ์บ์คํธ ์งํ์์ ์ด๋ฏธ์ง๋ฅผ ์ ๋ฌธ ์คํ๋์ค ๋ฐฐ๊ฒฝ์ผ๋ก ์ ์ค์ผ์ผ ๋ฐ ๋ฏธํโํ๋ ํ ์คํธ๋ฅผ ์งํํ์ต๋๋ค. ๊ฒฐ๊ณผ๋ฌผ์ โ๋ด ์ผ๊ตด์ด ์๋๋ฉฐ ๋ชจ๋ ๋ฉด์์ ๋์ฐํ๋คโ๊ณ ํ๊ฐํ ์ ๋๋ก ์ผ๊ตด ์ธ์์๋ ์คํจํ์ง๋ง, ์ฌ์ง๊ณผ ๊ฐ์ ์ฌ์ค์ ์ธ(Photorealistic) ์ง๊ฐ๊ณผ ๋น ๋ฅธ ์์ฑ ์๋๋ ์ธ์์ ์ด์๋ค๊ณ ์ธ๊ธํ์ต๋๋ค. ์ด๋ AI ์ด๋ฏธ์ง ์์ฑ ๋ชจ๋ธ๋ค์ด ํ ์คํธ ์์ฑ ๋ฅ๋ ฅ๊ณผ ํจ๊ป ์ด๋ฏธ์ง์ ์ฌ์ค์ฑ์ ๋์ด๋ ๋ฐ ์ฃผ๋ ฅํ๊ณ ์์์ ๋ณด์ฌ์ค๋๋ค.
Omni & Flow: ๋น๋์ค ์์ฑ ๋ฐ ํธ์ง์ ํ์
์ด๋ฒ I/O์์ ๊ฐ์ฅ ํฅ๋ฏธ๋ก์ด ๋ฐํ ์ค ํ๋๋ ์๋ก์ด ๋น๋์ค ์์ฑ ๋ชจ๋ธ์ธ โ์ด๋(Omni)โ์์ต๋๋ค. ์ด๋๋ ๊ธฐ์กด ๋ชจ๋ธ๋ณด๋ค ๋ ๊ธธ๊ณ , ์ผ๊ด์ ์ด๋ฉฐ, ํฌํ ๋ฆฌ์ผ๋ฆฌ์คํฑํ ๋น๋์ค๋ฅผ ์์ฑํ ์ ์์ผ๋ฉฐ, ์ฐธ๊ณ ์๋ฃ๋ฅผ ํ์ฉํด ๋น๋์ค๋ฅผ ๋ง๋๋ ๋ฅ๋ ฅ๋ ๊ฐ์ท์ต๋๋ค. ๊ตฌ๊ธ์ ์ด๋๋ฅผ โ๋น๋์ค๋ฅผ ์ํ ๋๋ ธ ๋ฐ๋๋โ๋ผ๊ณ ์นญํ๋ฉฐ, ์ถ๋ก ๋ชจ๋ธ๊ณผ ๋น๋์ค ์์ฑ ๋ชจ๋ธ์ ๊ฒฐํฉํ์ฌ โ๋ฌด์์ด๋ ๋น๋์ค๋ก ๋ง๋ค ์ ์๋คโ๊ณ ๊ฐ์กฐํ์ต๋๋ค.
ํด๋ผ๋ผ๋น์ ์์ ์ ์์ด๊ฐ ๊ทธ๋ฆฐ ๊ทธ๋ฆผ์ ํ์ฉํด ์ํผํ์ด๋ก๊ฐ ์์ด๋ฅผ ์์ ์์ ํ์ถ์์ผ ์ฆ๊ฒ๊ฒ ๋ ธ๋ ๋น๋์ค๋ฅผ ์์ฑํ๋ ํ ์คํธ๋ฅผ ์งํํ์ต๋๋ค. ์ด๋๋ ๊ณํ์ ์ธ์ฐ๊ณ ์ฝ 10์ด ๊ธธ์ด์ ๋น๋์ค๋ฅผ ์ฑ๊ณต์ ์ผ๋ก ์์ฑํ์ผ๋ฉฐ, ํด๋ผ๋ผ๋น์ โ์์ด๊ฐ ์ ๋ง ์ข์ํ ๊ฒโ์ด๋ผ๋ฉฐ ๊ฒฐ๊ณผ์ ๋ง์กฑํ์ต๋๋ค. ์ด๋ ์๋ผ(Sora)์ ๊ฐ์ ๊ธฐ์กด ๋น๋์ค ๋ชจ๋ธ๋ณด๋ค ๋ ๊ธด ๊ธธ์ด์ ์ผ๊ด์ฑ์ ๋ณด์ฌ์ฃผ๋ฉฐ, ๋ฉํฐ๋ชจ๋ฌ ๊ธฐ๋ฅ์ ๊ฐ๋ ฅํจ์ ์ ์ฆํ์ต๋๋ค.
์ด๋์ ์ฃผ์ ๋น๋์ค ํธ์ง ๊ธฐ๋ฅ:
- ๋ํํ ๋น๋์ค ํธ์ง: ํ๋กฌํํธ๋ง์ผ๋ก ๋น๋์ค์ ํน์ ์์๋ฅผ ๋ณ๊ฒฝ(์: ๊ฑด๋ฌผ์ ๊ฑฐํ์ผ๋ก ๋ฐ๊พธ๊ธฐ)ํ ์ ์์ต๋๋ค.
- ๋น๋์ค ์ค๋ช ๋ฐ ์ฌํธ์ง: ์ด๋๊ฐ ๋น๋์ค๋ฅผ ์ค๋ช ํ๊ณ , ์ด ์ค๋ช ์ ํธ์งํ์ฌ ์๋ก์ด ๋ฒ์ ์ ๋น๋์ค๋ฅผ ์์ฑํ ์ ์์ต๋๋ค.
- ์ผ๊ด์ฑ ์ ์ง: ํ๊ฒฝ, ๊ฐ๋ ๋ฑ์ ๋ณ๊ฒฝํ๋ฉด์๋ ์บ๋ฆญํฐ์ ์ผ๊ด์ฑ์ ์ ์งํ์ฌ ํ๋ก๋์ ์์ค์ ๋น๋์ค ์ ์์ ์ ์ฉํฉ๋๋ค.
๋์ฑ ์ฌ์ธต์ ์ธ ๋น๋์ค ํธ์ง์ ์ํด ๊ตฌ๊ธ์ โํ๋ก์ฐ(Flow)โ๋ผ๋ ์๋ก์ด ๋๊ตฌ๋ฅผ ๊ณต๊ฐํ์ต๋๋ค. ์ด๋ ๋ชจ๋ธ์ด ๋ด์ฅ๋ ํ๋ก์ฐ๋ ์๋ค๋งํฑ(Cinematic) ํ์ง์ ํธ์ง์ ์ค์ ์ ๋๋ฉฐ, ๋ฉํฐ๋ชจ๋ฌ ์ฐธ์กฐ ์๋ฃ๋ฅผ ํผํฉํ๊ณ ๋ํํ์ผ๋ก ๋น๋์ค๋ฅผ ํธ์งํ ์ ์์ต๋๋ค. ํนํ, ์ฌ์ฉ์๊ฐ ์บ๋ฆญํฐ๋ฅผ ์ ์ํ๊ณ ํฅํ ๋น๋์ค์์ ์ด ์บ๋ฆญํฐ๋ฅผ ๊ณ์ ์ฌ์ฉํ ์ ์๋ ๊ธฐ๋ฅ(์: โํ์ถ๋์ฅ ์ค์ฟจโ์ด๋ผ๋ ์บ๋ฆญํฐ ์์ฑ)๊ณผ ์์ ์ ์๋ฐํ๋ฅผ ์์ฑํ๋ ๊ธฐ๋ฅ์ด ์ฃผ๋ชฉ๋ฐ์์ต๋๋ค.
ํ์ง๋ง ํด๋ผ๋ผ๋น์ ์์ ์ ์๋ฐํ๋ฅผ ์์ฑํ๋ ๊ณผ์ ์์ ๋ค์ ํ๋ฒ ์ข์ ์ ๋ง๋ด์ผ ํ์ต๋๋ค. QR ์ฝ๋๋ฅผ ์ค์บํ๊ณ ์ผ๊ตด์ ์ดฌ์ํ๋ ์ ์ฐจ๋ฅผ ๊ฑฐ์ณค์์๋ ๋ถ๊ตฌํ๊ณ ์๋ฐํ๋ ์์ฑ๋์ง ์์์ต๋๋ค. ํด๋ผ๋ผ๋น์ โ๋ด ์ ๋ถ์ ๋์ฉํ์ง ๋ง์ ๋ฌ๋ผโ๊ณ ๋๋ดํ๋ฉฐ, โ๋ง์ ๊ฒ์ ๋ฐํํ์ง๋ง, ์์ง ์๋ํ์ง ์๊ฑฐ๋ ๊ธฐ๋ฅ์ด ๋ถ์์ ํ ์ ์ด ์ค๋ง์ค๋ฝ๋คโ๊ณ ์์งํ ์ฌ์ ์ ํ ๋กํ์ต๋๋ค.
Pamelli & Stitch: ๋ง์ผํ ๋ฐ ๋์์ธ์ AI ํ๋ช
๋ง์ผํฐ์ ๋์์ด๋๋ฅผ ์ํ ๋ ๊ฐ์ง ํฅ๋ฏธ๋ก์ด ๋๊ตฌ๋ ์๊ฐ๋์์ต๋๋ค.
-
ํ๋ฉ๋ฆฌ(Pamelli): ๋ธ๋๋ ๋ฐ ๋ง์ผํ ์ฝํ ์ธ ์์ฑ ๋๊ตฌ์ ๋๋ค. ์น์ฌ์ดํธ URL์ ์ ๋ ฅํ๋ฉด ํด๋น ์น์ฌ์ดํธ์ ๋ธ๋๋ ์์ด๋ดํฐํฐ๋ฅผ ๋ถ์ํ์ฌ ๋ธ๋๋ ๋ถ, ์บ ํ์ธ ์์ฐ, ํฌํ ์ํธ ๋ฑ์ ์์ฑํด ์ค๋๋ค. ๊ตฌ๊ธ์ด ๋ฐํํ
design.mdํ์ค์ ํ์ฉํ์ฌ AI๊ฐ ๋์์ธ ์์คํ ์ ์ดํดํ๊ณ ์ ์ฉํ๋ ๋ฐฉ์์ผ๋ก ์๋ํฉ๋๋ค. ํด๋ผ๋ผ๋น์ ์์ ์ ์น์ฌ์ดํธ๋ฅผ ๊ธฐ๋ฐ์ผ๋ก ๋ธ๋๋ ๋ถ๊ณผ ์น์ฌ์ดํธ๋ฅผ ์์ฑํ๋ ๊ณผ์ ์ ์์ฐํ์ต๋๋ค. ๊ฒฐ๊ณผ๋ โ๊ทธ๋ฅ ๊ด์ฐฎ์(fine)โ ์์ค์ด์์ง๋ง, ๋ธ๋๋ ์ ์ฒด์ฑ์ ์๋์ผ๋ก ์์ฑํ๊ณ ์ด๋ฅผ ๋ฐํ์ผ๋ก ๋ค์ํ ๋ง์ผํ ์์ฐ์ ๋ง๋๋ ์ ์ฌ๋ ฅ์ ๋์ด ํ๊ฐํ์ต๋๋ค. -
์คํฐ์น(Stitch): ํผ๊ทธ๋ง(Figma) ์คํ์ผ์ ์ธ ๋ธ๋ผ์ฐ์ (in-browser) ๋์์ธ ๋๊ตฌ์ ๋๋ค. ๊ธฐ์กด ๋์์ธ์ ํ์ฉํ๊ฑฐ๋ ์ธ๋ผ์ธ(inline) AI ํธ์ง์ ํตํด ๊ตฌ์ฑ ์์๋ฅผ ํ ์คํธ๋ก ์์ ํ ์ ์์ผ๋ฉฐ, ๋์์ธ ์บ๋ฒ์ค์ ์ค์๊ฐ์ผ๋ก ๋์์ธ์ ์คํธ๋ฆฌ๋ฐํ๋ ๊ธฐ๋ฅ์ด ํน์ง์ ๋๋ค. ํด๋ผ๋ผ๋น์ Google I/O ์ฌ์ดํธ ์คํ์ผ๋ก ์คํฐ์น ์ฑ ๋์์ธ์ ์์ฒญํ๊ณ , ์์ด์ ํธ๊ฐ ๋์์ธ ์์คํ ์ ๊ตฌ์ถํ๊ณ ์ฝํ ์ธ ๋ฅผ ๊ฐ์ ธ์ ๋ชจ๋ฐ์ผ ์ฑ ํ๋ฉด์ ์ค์๊ฐ์ผ๋ก ์คํธ๋ฆฌ๋ฐํ๋ ๋ชจ์ต์ ๋ณด์ฌ์ฃผ์์ต๋๋ค. ํด๋ผ๋ผ๋น์ โ๋์์ธ ๋ถ์ผ์์๋ ํ๋์ ๋ชจ๋ธ์ ์๋๋ฅผ ์ค์ ๋ก ์ฒด๊ฐํ ์ ์์๋คโ๋ฉฐ, โ๋ ๋ง์ ๋์์ด๋๋ค์ด ์ด ๋๊ตฌ๋ฅผ ์ฌ์ฉํด๋ณด๊ณ ์ถ์ด ํ ๊ฒโ์ด๋ผ๊ณ ํธํํ์ต๋๋ค.
๊ฒฐ๋ก : ๊ธฐ๋์ ํ์ค ์ฌ์ด์ Google I/O
์ด๋ฒ Google I/O 2026์ ๊ตฌ๊ธ์ AI์ ๋ํ ๊น์ ์ผ์ฌ๊ณผ ๊ธฐ์ ์ ์ง๋ณด๋ฅผ ๋ช ํํ ๋ณด์ฌ์ฃผ์์ต๋๋ค. ์๋ก์ด ์ ๋ฏธ๋ 3.5 ๋ชจ๋ธ๊ตฐ์ ์๋์ ์ง๋ฅ, ๊ฐ๋ฐ์๋ฅผ ์ํ ์ํฐ๊ทธ๋๋นํฐ IDE์ ๋ค์ํ ๊ธฐ๋ฅ, ์ํฌ์คํ์ด์ค ์ฐ๋์ ํตํ ์์ฐ์ฑ ํฅ์, ๊ทธ๋ฆฌ๊ณ ์ด๋์ ํ๋ก์ฐ๋ฅผ ํตํ ํ์ ์ ์ธ ๋น๋์ค ์์ฑ ๋ฐ ํธ์ง ๊ธฐ๋ฅ, ํ๋ฉ๋ฆฌ์ ์คํฐ์น ๊ฐ์ ๋ง์ผํ /๋์์ธ ๋๊ตฌ๊น์ง, ํฅ๋ฏธ๋ก์ด ๋ฐํ๋ค์ด ์ค์ ์ด์์ต๋๋ค. ํนํ ๋น๋์ค ์์ฑ ๊ธฐ๋ฅ๊ณผ ํ๋์ ๋ชจ๋ธ ๊ธฐ๋ฐ์ ์ฝ๋ฉ ์๋ ํฅ์์ ํด๋ผ๋ผ๋น์ด ๊ฐ์ฅ ๊ธฐ๋ํ๋ ๋ถ๋ถ์ด์์ต๋๋ค.
ํ์ง๋ง ํด๋ผ๋ผ๋น์ ์ด๋ฌํ ๋ฐํ๋ค ์์์ ๋ช ๊ฐ์ง โ๋ ์นด๋ก์ด ๋ชจ์๋ฆฌ(sharp edges)โ๋ฅผ ์ง์ ํ์ต๋๋ค. ์ผ๋ถ ํต์ฌ ๊ธฐ๋ฅ๋ค(์: Google AI Studio์ ์ํฌ์คํ์ด์ค ์ฐ๋, Flow์ ์๋ฐํ ์์ฑ)์ ์์ง ์ ๋๋ก ์๋ํ์ง ์๊ฑฐ๋, ์ผ๋ฐ ์ฌ์ฉ์์๊ฒ๋ ์ ๊ทผ์ฑ์ด ๋จ์ด์ง๋ ๋ฌธ์ ๊ฐ ์์์ต๋๋ค. ๋ํ ๋๋ฌด๋ ๋ง์ ์ ํ๋ช ๊ณผ ๋ธ๋๋๊ฐ ํผ์ฌ๋์ด ์ฌ์ฉ์๋ค์๊ฒ ํผ๋์ ์ผ๊ธฐํ๋ค๋ ์ ๋ ํฐ ๊ณผ์ ๋ก ๋จ์์ต๋๋ค.
๊ตฌ๊ธ์ AI ํ์ ์ ๋ถ๋ช ํ์ฌ ์งํํ์ด๋ฉฐ, ์์ฒญ๋ ์ ์ฌ๋ ฅ์ ๊ฐ์ง๊ณ ์์ต๋๋ค. ๊ทธ๋ฌ๋ ๋ฐํ๋ ๊ธฐ๋ฅ๋ค์ด ์ค์ ์ฌ์ฉ์๋ค์๊ฒ ์จ์ ํ ์ ๋ฌ๋๊ณ ๋งค๋๋ฌ์ด ๊ฒฝํ์ ์ ๊ณตํ๊ธฐ ์ํด์๋ ์ ํ์ ์์ฑ๋์ ์ ๊ทผ์ฑ, ๊ทธ๋ฆฌ๊ณ ์ผ๊ด๋ ๋ธ๋๋ ์ ๋ต์ ๋ํ ๊ตฌ๊ธ์ ๋ ธ๋ ฅ์ด ๋์ฑ ํ์ํ ๊ฒ์ผ๋ก ๋ณด์ ๋๋ค. ์ด๋ฒ Google I/O๋ AI ์๋์ ๋ฐ์ ๋น์ ์ ์ ์ํจ๊ณผ ๋์์, ๊ทธ ๋น์ ์ ํ์ค๋ก ๋ง๋ค๊ธฐ ์ํ ์ง๋ํ ๊ณผ์ ๋ํ ํจ๊ป ๋ณด์ฌ์ค ํ์ฌ์์ต๋๋ค.
โTrumpโs Taxpayer-Funded Revenge Planโ โ New York Times Podcasts ๊ธฐ๋ฐ ๊ธฐ์ฌ ์๋ณธ ์์ ๋ณด๊ธฐ
ํธ๋ผํ์ โ๋ฉ์ธ์ ๋ณด๋ณต ๊ธฐ๊ธโ ๋ ผ๋: 17์ต ๋ฌ๋ฌ์ ์ง์ค๊ณผ ๋ฏธ๊ตญ์ ๋ถ์ด
์ง๋์ฃผ, ์์ฑํด ์ ๊ฐ๋ฅผ ๋คํ๋ ์ถฉ๊ฒฉ์ ์ธ ์์์ด ์ ํด์ก์ต๋๋ค. ๋๋๋ ํธ๋ผํ ์ ๋ํต๋ น์ด ๊ตญ์ธ์ฒญ(IRS)์ ์๋๋ก ์ ๊ธฐํ๋ 100์ต ๋ฌ๋ฌ ๊ท๋ชจ์ ์์ก์ ์ ๊ฒฉ ์ฒ ํํ๋ ๋์ , ๋ฒ๋ฌด๋ถ(Justice Department)๊ฐ 17์ต 7,600๋ง ๋ฌ๋ฌ์ ๋ฌํ๋ ๊ฑฐ๋ํ ๊ธฐ๊ธ์ ์กฐ์ฑํ๊ฒ ๋ค๊ณ ๋ฐํํ ๊ฒ์ ๋๋ค. ์ด ๊ธฐ๊ธ์ ํ์ ๋ถ๊ฐ ์ฃผ์ฅํ๋ โ๋ฌด๊ธฐํ(weaponization) ๋ฐ ๋ฒ๋ฅ ์ (lawfare)โ์ ํผํด์๋ค์ ๋ณด์ํ๊ธฐ ์ํ ๊ฒ์ผ๋ก, ํนํ 1์ 6์ผ ๋ฏธ๊ตญ ๊ตญํ์์ฌ๋น ๋์ ์ฌ๊ฑด์ ์ฐ๋ฃจ๋ ์ฝ 1,600๋ช ์ ํผ๊ณ ์ธ๋ค๊น์ง ์ํ ๋์์ด ๋ ์ ์๋ค๋ ์ ์์ ์์ฒญ๋ ๋ ผ๋์ ๋ถ๋ฌ์ผ์ผํค๊ณ ์์ต๋๋ค.
๋ด์ํ์์ค ํ์บ์คํธ โ๋ ๋ฐ์ผ๋ฆฌ(The Daily)โ๋ ์ด ์ ๋ก ์๋ ๊ธฐ๊ธ์ ๋ฐฐ๊ฒฝ, ์๋ ๋ฐฉ์, ๊ทธ๋ฆฌ๊ณ ์ด๋ก ์ธํด ์ด๋ฐ๋ ์ด๋น์ ๋ถ๋ ธ์ ๋ํด ์ฌ์ธต์ ์ผ๋ก ๋ค๋ฃจ์์ต๋๋ค. ๋นํ๊ฐ๋ค์ ์ด ๊ธฐ๊ธ์ โํ๋ผ์ฐ๋ ๋ณด์ด์ค(Proud Boys)์ ์ค์ค ํคํผ์ค(Oath Keepers) ๋ฑ ํธ๋ผํ์ ์ ์น์ ์ถฉ์ ๋ค์ ์ํ 17์ต ๋ฌ๋ฌ ๊ท๋ชจ์ ์ ์น์ ์์ง๋(political slush fund)โ์ด๋ผ๊ณ ๋งน๋น๋ํ๋ฉฐ, ํ์ ๋ถ๊ฐ ๋ตํด์ผ ํ ์๋ง์ ์ง๋ฌธ์ด ์๋ค๊ณ ์ง์ ํฉ๋๋ค. ์ด ๊ธฐ์ฌ๋ โ๋ ๋ฐ์ผ๋ฆฌโ์ ๋ณด๋๋ฅผ ๋ฐํ์ผ๋ก, ํธ๋ผํ์ โ๋ฉ์ธ์ ๋ณด๋ณต ๊ธฐ๊ธโ์ ๋๋ฌ์ผ ๋ณต์กํ ์ง์ค์ ํํค์นฉ๋๋ค.
๋ ผ๋์ ์์: ํธ๋ผํ์ IRS ์์ก๊ณผ ์ธ๊ธ ์ ๋ณด ์ ์ถ
์ด ์ด์ผ๊ธฐ๋ 2016๋ ํธ๋ผํ๊ฐ ์ฒ์ ๋์ ์ ์ถ๋งํ์ ๋๋ก ๊ฑฐ์ฌ๋ฌ ์ฌ๋ผ๊ฐ๋๋ค. ๋น์ ๊ทธ๋ ์ญ๋ ๋์ ํ๋ณด๋ค๊ณผ ๋ฌ๋ฆฌ ์์ ์ ๋ฉ์ธ ์๋ฃ๋ฅผ ๊ณต๊ฐํ์ง ์์ ํฐ ์ค์บ๋ค์ด ๋์์ต๋๋ค. ๋ํต๋ น ํ๋ณด๊ฐ ์์ ์ ์ฌ์ ์ํ๋ฅผ ํฌ๋ช ํ๊ฒ ๊ณต๊ฐํ๋ ๊ฒ์ ์ค๋ซ๋์ ๊ด๋ก์ฒ๋ผ ์ฌ๊ฒจ์ก์ง๋ง, ํธ๋ผํ๋ ์ด๋ฅผ ๊ฑฐ๋ถํ์ต๋๋ค. ๊ทธ์ ์ฒซ ์๊ธฐ ๋ด๋ด ๋ฉ์ธ ์๋ฃ ๊ณต๊ฐ ์๊ตฌ๋ ๊ณ์๋์๊ณ , ๊ฒฐ๊ตญ 2020๋ ๋ด์ํ์์ค๊ฐ ์ ์ถ๋ ์๋ฃ๋ฅผ ๋ฐํ์ผ๋ก ํธ๋ผํ๊ฐ ์๋ ๊ฐ ์ฐ๋ฐฉ ์๋์ธ๋ฅผ ๊ฑฐ์ ๋ด์ง ์์๋ค๋ ์ฌ์ค์ ํญ๋กํ๋ฉฐ ํฐ ํ์ฅ์ ์ผ์ผ์ผฐ์ต๋๋ค.
๋ด์ํ์์ค ๋ณด๋ ์ดํ ์ฝ 6๊ฐ์์์ 1๋ ๋ค, ๋ ๋ค๋ฅธ ํ์ฌ๋ณด๋ ์ ๋ฌธ ๋งค์ฒด์ธ ํ๋กํผ๋ธ๋ฆฌ์นด(ProPublica)๋ ์ ํ ๋ฒ ์ด์กฐ์ค(Jeff Bezos), ์ผ๋ก ๋จธ์คํฌ(Elon Musk) ๋ฑ ์ธ๊ณ ์ต๊ณ ๋ถ์๋ค์ ์ธ๊ธ ์ ๋ต๊ณผ ์ํฉ์ ์์ธํ ๋ณด๋ํ์ต๋๋ค. ์๋ง์ ๊ณ ์ก ์์ฐ๊ฐ๋ค์ ์ธ๊ธ ์ ๋ณด๊ฐ ๊ณต๊ฐ๋๋ฉด์, ๊ตญ์ธ์ฒญ(IRS) ๋ด๋ถ ๊ด๊ณ์๊ฐ ์ด ์ ๋ณด ์ ์ถ์ ๊ทผ์์์ด ๋ช ํํด์ก์ต๋๋ค. ์ฆ, IRS ๋ด๋ถ์ ์ ๋ณด ์ ์ถ์(leaker)๊ฐ ์์๋ ๊ฒ์ ๋๋ค.
2023๋ ๊ฐ์, ๋ง์นจ๋ด ์ฐฐ์ค ๋ฆฌํ์กด(Charles Littlejohn)์ด๋ผ๋ ์ IRS ๊ณ์ฝ์ง ์ง์์ด ์ฒดํฌ๋์ด ๊ธฐ์๋์์ต๋๋ค. ๊ทธ๋ ํธ๋ผํ์ ๋ค๋ฅธ ๋ถ์ ์ธต ์ธ์ฌ๋ค์ ์ธ๊ธ ์ ๋ณด๋ฅผ ๋ด์ํ์์ค์ ํ๋กํผ๋ธ๋ฆฌ์นด์ ์ ๊ณตํ ํ์๋ฅผ ์ธ์ ํ์ผ๋ฉฐ, 2024๋ ์ด ์ง์ญํ์ ์ ๊ณ ๋ฐ์์ต๋๋ค. ํ์ธ์ ์ธ๊ธ ์ ๋ณด๋ฅผ ์ ์ถํ๋ ๊ฒ์ ๋ช ๋ฐฑํ ๋ฒ์ฃ์ด๋ฉฐ, ํผํด์๋ IRS๋ฅผ ์๋๋ก ๋ฏผ์ฌ ์์ก์ ์ ๊ธฐํ์ฌ ์ํด๋ฐฐ์์ ์ฒญ๊ตฌํ ์ ์์ต๋๋ค. ์ด์ ํธ๋ผํ ์ ๋ํต๋ น๊ณผ ๊ทธ์ ๋ ์๋ค, ๊ทธ๋ฆฌ๊ณ ๊ฐ์กฑ ๊ธฐ์ ์ ์ง๋ 1์, IRS๊ฐ ์ฐฐ์ค ๋ฆฌํ์กด์ ์ ๋ณด ์ ์ถ์ ๋ง์ง ๋ชปํ๋ค๋ ์ด์ ๋ก ์ต์ 1,000๋ง ๋ฌ๋ฌ์ ์ํด๋ฐฐ์์ ์๊ตฌํ๋ ์์ก์ ์ ๊ธฐํ์ต๋๋ค.
์ ๋ก ์๋ ๊ฐ๋ฑ: ๋ํต๋ น์ด ์ ๋ถ๋ฅผ ๊ณ ์ํ๋ค
๊ฐ์ธ์ด ์์ ์ ์ธ๊ธ ์ ๋ณด ์ ์ถ์ ๋ํด IRS๋ฅผ ๊ณ ์ํ๋ ๊ฒ์ ํฉ๋ฒ์ ์ธ ๊ถ๋ฆฌ์ ๋๋ค. ๊ทธ๋ฌ๋ ํธ๋ผํ์ ์ฌ๋ก๋ ๋งค์ฐ ์ด๋ก์ ์ด๋ฉฐ, ์ ๋ก๊ฐ ์๋ค๊ณ ํ ์ ์์ต๋๋ค. ๊ทธ๋ ๋จ์ํ ์ผ๋ฐ ์๋ฏผ์ด ์๋๋ผ, ์์ ์ด ์ง์ ๊ฐ๋ ํ๋ ์ ๋ถ ๊ธฐ๊ด์ ๊ณ ์ํ ๊ฒ์ ๋๋ค. ๋ํต๋ น์ IRS ๊ตญ์ฅ๊ณผ ์ฌ๋ฌด๋ถ(Treasury Department)์ ์์ฅ์ ์๋ช ํ๋ฉฐ, ์ด๋ค ๊ธฐ๊ด์ ์์ฌ ๊ฒฐ์ ์ ๊น์ด ๊ด์ฌํฉ๋๋ค. ๋์ฑ์ด, ๊ทธ๋ ์์ ์ ๋ ๋ฒ์งธ ์๊ธฐ ๋์ ๋ฒ๋ฌด๋ถ(Justice Department)๋ฅผ ์์ ์ ๋ป๋๋ก ์ฌํธํ๋ ๋ฐ ๋ง๋ํ ์ํฅ๋ ฅ์ ํ์ฌํด์์ต๋๋ค.
๋ฌธ์ ๋ ๋ฒ๋ฌด๋ถ๊ฐ ์ด ์์ก์์ IRS์ ๋ณํธ์ธ ์ญํ ์ ํด์ผ ํ๋ค๋ ์ ์ ๋๋ค. ํ์ง๋ง ํธ๋ผํ์ ๋ ๋ฒ์งธ ์๊ธฐ์ ๋ค์ด์๋ฉด์ ๋ฒ๋ฌด๋ถ๋ ์ญ์ฌ์ ์ผ๋ก ๋ ๋ฆฝ์ ์ด์๋ ์ญํ ์ ๋์ด, ๋ํต๋ น์ ์ ์น์ ์ ๋์๋ค์ ๊ธฐ์ํ๊ณ ๋ฏผ์ฃผ๋น ์์๋ค์ ์ฐ๋ฐฉ ์์ฌ ๋์์ ์ฌ๋ฆฌ๋ ๋ฑ ์ด๋ก์ ์ธ ํ๋ณด๋ฅผ ๋ณด์ฌ์์ต๋๋ค. ํ์ฌ ๋ฒ๋ฌด๋ถ์ ๋ฒ๋ฌด์ฅ๊ด ๋ํ์ธ ํ ๋ ๋ธ๋์ฌ(Todd Blanche)๋ ๋ฐฑ์ ๊ด์ ๋ณต๊ทํ๊ธฐ ์ ํธ๋ผํ์ ๊ฐ์ธ ๋ณํธ์ฌ์์ต๋๋ค. ์ด๋ฌํ ์ํฉ์์ ๋ฒ๋ฌด๋ถ๊ฐ ๋ํต๋ น์ด ์ ๊ธฐํ ์์ก์ ๋ํด ์ด๋ป๊ฒ ๋ฐฉ์ดํ ์ง ๊ฒฐ์ ํ๋ ๊ฒ์ ์ฌ๊ฐํ ์ดํด ์์ถฉ(conflict of interest) ๋ฌธ์ ๋ฅผ ์ผ๊ธฐํ์ต๋๋ค. ํธ๋ผํ๊ฐ ์ฌ์ ์ผ๋ก ๊ณ ์ฉํ ๋ณํธ์ธ๋จ์ ํตํด ์ ๋ถ๋ฅผ ๊ณ ์ํ๋ ๋์์, ์ ๋ถ ์ธก ๋ณํธ์ธ๋จ๊น์ง ์ด๋ก์ ์ผ๋ก ๊ฐ๋ ฅํ๊ฒ ํต์ ํ๋ ์ํฉ์ด ๋ฒ์ด์ง ๊ฒ์ ๋๋ค.
์ฌํ๋ถ๋ ์ด๋ฌํ ๋ณต์กํ ์ดํด ์์ถฉ ๊ด๊ณ๋ฅผ ์ฐ๋ คํ๋ฉฐ ์์ก ์งํ์ ์๋ฌธ์ ์ ๊ธฐํ์ต๋๋ค. ์์ก ๋น์ฌ์๋ค์ ์ค์ ๋ก ์๋ก ๋๋ฆฝํด์ผ ํ๋ค๋ ๋ฒ์ ์์น์ด ์๊ธฐ ๋๋ฌธ์ ๋๋ค. ํ์ฌ๋ ์ด ์์ก์ด โ์ง์ ํ ๋๋ฆฝ์ด ์๋ ๊ฐ์ฅ๊ทน(charade)โ์ผ ์ ์๋ค๊ณ ํ๋จํ๊ณ , ๋ฒ๋ฌด๋ถ์ ํธ๋ผํ ์ธก ๋ณํธ์ธ๋จ์๊ฒ ์์ก์ด ์ค์ ๋ก ๋๋ฆฝํ๋ ๊ด๊ณ์ธ์ง, ์๋๋ฉด โ๊ฑฐ์ง๋ ๊ตฌ์ค๋ก ์ ๊ธฐ๋ ์์กโ์ธ์ง ์ค๋ช ํ๋ผ๋ ์๋ฉด ๋ต๋ณ์ ์๊ตฌํ์ต๋๋ค. ๋ง์ฝ ์ฌํ๋ถ๊ฐ ์ง์ ํ ๋๋ฆฝ์ด ์๋ค๊ณ ํ๋จํ๋ค๋ฉด, ์์ก์ ๊ธฐ๊ฐํ ์๋ ์๋ ์ํฉ์ด์์ต๋๋ค. ์ด๋ฌํ ์ํฉ์ ํธ๋ผํ์๊ฒ ๋ถ๋ฆฌํ๊ฒ ๋์๊ฐ๊ณ ์์์ผ๋ฉฐ, ๋ฒ๋ฌด๋ถ ์ ์ฅ์์๋ ๋ํต๋ น์ ์ฃผ์ฅ์ ๋ฐ๋ํ๋ ์ ์ฅ์ ํ๋ช ํด์ผ ํ๋ ๋งค์ฐ ๊ณค๋ํ ์ํฉ์ด์์ต๋๋ค. ํนํ ๋ํต๋ น์ ๋ฐ๋ํ๋ค๊ฐ ํด๊ณ ๋ ์ ๋ก๊ฐ ์๋ ํ ๋ฒ๋ฌด๋ถ์ ๋ถ์๊ธฐ์์๋ ๋์ฑ ๊ทธ๋ฌ์ต๋๋ค.
โ๋ฌด๊ธฐํ ํผํด ๋ณด์ ๊ธฐ๊ธโ์ ํ์
์ด๋ฌํ ๊ณค๋ํ ์ํฉ์ ํผํ๊ธฐ ์ํด ํ์ ๋ถ ๋ด๋ถ์์๋ ์ง๋์ฃผ ๋ด๋ด ํ๋ฐํ ๋ ผ์๊ฐ ์ด๋ฃจ์ด์ก์ต๋๋ค. ๊ทธ ๊ฒฐ๊ณผ, ํธ๋ผํ ์ ๋ํต๋ น์ด IRS ์์ก์ ์์ ํ ์ฒ ํํ๊ณ , ๊ทธ ๋๊ฐ๋ก ๋ฒ๋ฌด๋ถ๊ฐ โ๋ฌด๊ธฐํโ์ ํผํด์๋ค์ ๋ณด์ํ๊ธฐ ์ํ ์ ๋ก ์๋ ๊ฑฐ๋ํ ๋ฉ์ธ์ ๊ธฐ๊ธ์ ์กฐ์ฑํ๋ค๋ ํฉ์์ ์ด๋ฅด๊ฒ ๋ฉ๋๋ค.
์ด ๊ธฐ๊ธ์ ๊ท๋ชจ๋ 17์ต 7,600๋ง ๋ฌ๋ฌ์ ๋๋ค. โ1776โ์ด๋ผ๋ ์ซ์๋ ๋ฏธ๊ตญ ๊ฑด๊ตญ ์ฐ๋๋ฅผ ์์งํ๋ฉฐ, ์ด๋ ์ฐ์ฐ์ด ์๋ ๊ฒ์ด๋ผ๋ ๋ถ์์ด ์ง๋ฐฐ์ ์ ๋๋ค. ์ด ๋ง๋ํ ์๊ธ์ โํ๊ฒฐ ๊ธฐ๊ธ(Judgment Fund)โ์ด๋ผ๋ ๊ณ์ ์์ ์ธ์ถ๋ ์์ ์ ๋๋ค. ํ๊ฒฐ ๊ธฐ๊ธ์ ๋ฒ๋ฌด๋ถ๊ฐ ๋ฏธ๊ตญ ์ ๋ถ๋ฅผ ์๋๋ก ์ ๊ธฐ๋ ๋ฒ์ ์์ก์ ํด๊ฒฐํ๋ ๋ฐ ์ฌ์ฉํ๋ ์๊ธ์ผ๋ก, ์ ๋ถ๊ฐ ์์ก์์ ํฉ์ํ ๋ ์ด ๊ธฐ๊ธ์์ ๋์ ์ธ์ถํฉ๋๋ค. ํ์๋ณด๋ค ํจ์ฌ ๋ง์ ๊ธ์ก์ด ์ธ์ถ๋ ์์ ์ด๋ฉฐ, ์ด ๋์ ๋ณ๋์ ๊ณ์ ์ผ๋ก ์์น๋์ด ๋ฒ๋ฌด์ฅ๊ด์ด ์๋ช ํ 5๋ช ์ ์์์ด ๊ด๋ฆฌํ๊ฒ ๋ฉ๋๋ค. ์ด๋ค์ ์ฐ๋ฐฉ ์ ๋ถ์ ์ํด ํผํด๋ฅผ ์ ์๋ค๊ณ ์ฃผ์ฅํ๋ ์ฌ๋๋ค์ ์ฒญ๊ตฌ๋ฅผ ์ฌ์ฌํ๊ณ ๋ณด์๊ธ์ ์ง๊ธํ๊ฒ ๋ฉ๋๋ค.
๊ทธ๋ฌ๋ ์ด ๊ธฐ๊ธ์ ์๋ ๋ฐฉ์์ ๋ํด์๋ ์ฌ์ ํ ๋ง์ ๊ฒ์ด ๋ถ๋ถ๋ช ํฉ๋๋ค. ๋ณด์๊ธ ์ฒญ๊ตฌ๊ฐ ์ด๋ป๊ฒ ํ๊ฐ๋ ์ง, ๋ฒ๋ฌด๋ถ๊ฐ ๋งํ๋ โ๋ฌด๊ธฐํ(weaponization)โ๊ฐ ์ ํํ ๋ฌด์์ ์๋ฏธํ๋์ง, ๊ทธ๋ฆฌ๊ณ ์ผ๋ง๋ ๋ง์ ๋์ด ์ง๊ธ๋ ์ง์ ๋ํ ๊ตฌ์ฒด์ ์ธ ๊ธฐ์ค์ ์์ง ๊ณต๊ฐ๋์ง ์์์ต๋๋ค. ์ฌ์ง์ด ์ด ๊ธฐ๊ธ์ ํต์ ํ๊ณ ๊ฒฐ์ ์ ๋ด๋ฆด 5๋ช ์ ์์์ด ๋๊ตฌ์ธ์ง์กฐ์ฐจ ์๋ ค์ง์ง ์์์ต๋๋ค. ์ด๋ค์ด ๋ง๋ํ ์ญํ ์ ์ํํ ๊ฒ์ด ๋ถ๋ช ํ ๋งํผ, ์์ฑํด์์๋ ์ด ๊ธฐ๊ธ์ด ๊ถ๊ทน์ ์ผ๋ก ์ด๋ค ๋ชจ์ต์ด ๋ ์ง์ ๋ํ ์๋ง์ ์๋ฌธ์ด ์ ๊ธฐ๋๊ณ ์์ต๋๋ค. ๋ฒ๋ฌด๋ถ๊ฐ ๋ค๋ฅธ ์์ก๋ค์ ํฉ์ํ๋ ๊ฒฝ์ฐ๋ ๋ง์ง๋ง, ์ด์ฒ๋ผ ๋๊ท๋ชจ์ ํฌ๊ด์ ์ธ โํผํด ๋ณด์ ๊ธฐ๊ธโ์ ์กฐ์ฑํ๋ ๊ฒ์ ์ ๋ก ์๋ ์ผ์ ๋๋ค.
๋๊ฐ ์ด โ๋ณต์ ๊ธฐ๊ธโ์ ์ํ์๊ฐ ๋ ๊ฒ์ธ๊ฐ?
๋ฒ๋ฌด๋ถ๋ ์ด ๊ธฐ๊ธ์ ์ํ ๋์์ โ๋ฌด๊ธฐํ ๋ฐ ๋ฒ๋ฅ ์ ์ ํผํด์๋คโ์ด๋ผ๊ณ ๋ง ๋ฐํ์ต๋๋ค. ํ์ง๋ง ์ด ๋ฐํ์ ๋ํด ๊ฐ์ฅ ๋จผ์ ๊ธฐ๋๋ฅผ ํ๋ช ํ ์ง๋จ ์ค ํ๋๋ ๋ฐ๋ก 2021๋ 1์ 6์ผ ๊ตญํ์์ฌ๋น ๋์ ์ฌ๊ฑด์ผ๋ก ์ ์ฃ ํ๊ฒฐ์ ๋ฐ์๋ ์ฌ๋๋ค, ์ฆ ํธ๋ผํ ์ ๋ํต๋ น์ด ์ด์ ์ ์ฌ๋ฉดํ๋ ์ฌ๋๋ค์ ๋๋ค. ์ด๋ค์ ๋ฌผ๋ก , ์ค๋ฐ๋ง ํ์ ๋ถ ์์ ๋ฒ๋ฌด๋ถ์ ์ฐ๋ฐฉ์์ฌ๊ตญ(FBI)์ด ์์ ๋ค์ ๋ถ๋นํ๊ฒ ์์ฌํ๋ค๊ณ ๋ฏฟ๋ ํธ๋ผํ ์ง์ง์๋ค๊ณผ ์ ์น์ ๋๋งน์๋ค๋ ์ด ๊ธฐ๊ธ์ ํตํด ๊ณผ๊ฑฐ์ ๊ฒฝํ์ผ๋ก๋ถํฐ โ์์ ํ ๋ณด์โ์ ๋ฐ์ ๊ธฐํ๋ก ๋ณด๊ณ ์์ต๋๋ค.
๊ทธ๋ฌ๋ ์ด ๊ธฐ๊ธ์ ์ง๊ธ ๋ฐฉ์์ ๋ํด์๋ ์๋ฌธ์ด ๋ง์ต๋๋ค. ์ง๊ธ์ก์ด ๋ฐ์ํ ๋ฒ๋ฅ ๋น์ฉ์ ์ํํ๋ ์ฉ๋์ธ์ง, ์๋๋ฉด ๊ทธ ์ด์์ ๊ธ์ก์ด ์ง๊ธ๋์ด ์ฌ์ค์ โ๊ฒฝํ์ผ๋ก๋ถํฐ ์ด๋์ ์ทจํ โ ์ ์๊ฒ ๋๋ ๊ฒ์ธ์ง ๋ช ํํ์ง ์์ต๋๋ค. ๋ํ ๋๊ฐ ์ผ๋ง์ ๋์ ๋ฐ์๋์ง์ ๋ํ ๋ช ํํ ์์์ฆ ๊ณต๊ฐ๊ฐ ์์ ๊ฐ๋ฅ์ฑ๋ ์ ๊ธฐ๋์ด, ์ด ๋ชจ๋ ์ด์์ด ๋์ค์ ์์ผ์์ ์์ ํ ๊ฐ๋ ค์ง ์ ์๋ค๋ ์ฐ๋ ค๋ ๋์ต๋๋ค.
๋ง์ ์ฌ๋๋ค์๊ฒ 1์ 6์ผ ๊ตญํ์์ฌ๋น์ ์ต๊ฒฉํ๋ ์ฌ๋๋ค์ด ์ ๋ถ๋ก๋ถํฐ ๋์ ๋ฐ๋๋ค๋ ์๊ฐ์ ์ถฉ๊ฒฉ์ ์ผ ๊ฒ์ ๋๋ค. ์ฌํ๊ฒ ๋งํ๋ฉด, ์์ ๋ค์ด ๊ณต๊ฒฉํ๋ ์ ๋ถ๋ก๋ถํฐ ์ ์ฃ ํ๊ฒฐ์ ๋ฐ์ ๋ฒ์ฃ์๋ค์ด ๋์ ๋ฐ์ ์ ์๋ค๋ ์๋๋ฆฌ์ค๊ฐ ์ด ๊ธฐ๊ธ์ ๊ตฌ์กฐ ํ์์๋ ์ถฉ๋ถํ ๊ฐ๋ฅํด ๋ณด์ ๋๋ค. ์ด๋ ๊ธฐ๊ธ ๋ฐํ์ ๋ํ ์ฌ์ผ ์์ธก์ ๊ฒฉ๋ ฌํ ๋ฐ์์ ์ค๋ช ํ๋ ์ฃผ์ ์์ธ์ด๊ธฐ๋ ํฉ๋๋ค.
์ฌ์ผ์ ๊ฒฉ๋ ฌํ ๋นํ๊ณผ ๋ด๋ถ ๊ท ์ด
๋ฒ๋ฌด์ฅ๊ด ๋ํ ํ ๋ ๋ธ๋์ฌ๋ ํ์์ผ ๊ตญํ์์ฌ๋น์์ ์ด๋ฆฐ ๋ฒ๋ฌด๋ถ ์์ฐ ์ฒญ๋ฌธํ์ ์ถ์ํ์ฌ ์ด ๊ธฐ๊ธ ์กฐ์ฑ์ ์นํธํ์ต๋๋ค. ๊ทธ๋ โ์ด๊ฒ์ด ์ด๋ก์ ์ธ ๊ฒ์ ์ฌ์ค์ด์ง๋ง, ์ ๋ก๊ฐ ์๋ ๊ฒ์ ์๋๋คโ๋ผ๊ณ ์ฃผ์ฅํ๋ฉฐ, ๊ธฐ๊ธ์ด ํน์ ์ ์น ์ฑํฅ์ด๋ 1์ 6์ผ ์ฌ๊ฑด, ๋๋ ํน์ ์ธ๋ฌผ์ ๊ตญํ๋์ง ์๊ณ โ๋๊ตฌ๋ ์ ์ฒญํ ์ ์๋คโ๊ณ ๊ฐ์กฐํ์ต๋๋ค.
๊ทธ๋ฌ๋ ์ฒญ๋ฌธํ๋ ๊ณง ๋ฏผ์ฃผ๋น ์์๋ค์ด ์ด ๊ธฐ๊ธ์ โ์ ์น์ ์์ง๋(slush fund)โ์ด๋ผ๊ณ ๋งน๋น๋ํ๋ฉฐ ๋ธ๋์ฌ ๋ํ์ ์ง์ค์ ์ผ๋ก ์ถ๊ถํ๋ ์ฅ์ด ๋์์ต๋๋ค. ํฌ๋ฆฌ์ค ๋ฐด ํ๋ (Chris Van Hollen) ์์์์ ๋ฑ ๋ฏผ์ฃผ๋น ์์๋ค์ โ์ด๊ฒ์ ํฐ๋ฌด๋์๊ณ ์ ๋ก ์๋ ์์ง๋โ, โ๊ณต๊ณต ์๊ธ์ ์์ํ ์ ๋โ, โ์ธ์ค์ ์ด๊ณ ๋ถ๋ฒ์ ์ด๋ฉฐ ๋ถํจํ ์ ํ-๋๋ง ๊ณํโ์ด๋ผ๊ณ ๊ฐํ๊ฒ ๋นํํ๋ฉฐ, 1์ 6์ผ ์์ฌ๋น ๊ฒฝ์ฐฐ๊ด์ ํญํํ ์ฌ๋๋ค๋ ์ด ๊ธฐ๊ธ์ ์ํ์๊ฐ ๋ ์ ์๋์ง ๊ฑฐ๋ญ ์ง๋ฌธํ์ต๋๋ค. ๋ฐด ํ๋ ์์์ ๋ธ๋์ฌ ๋ํ์ด โ๋ฒ๋ฌด์ฅ๊ด ๋ํ์ด ์๋ ๋ํต๋ น์ ๊ฐ์ธ ๋ณํธ์ฌ์ฒ๋ผ ํ๋ํ๊ณ ์๋คโ๊ณ ๋น๋ํ๋ฉฐ, โ์ด๋ฌํ ์๋ชป๋ ํ์๊ฐ ๊ณ์๋๋๋ก ํ ์ด๋ ํ ์๊ธ๋ ์ง๊ธํ๋ ๊ฒ์ ์ ๋นํํ๊ธฐ ์ด๋ ต๋คโ๊ณ ๋ง๋ถ์์ต๋๋ค.
๋๋๊ฒ๋, ๊ณตํ๋น ์์๋ค ์ฌ์ด์์๋ ์ด ๊ธฐ๊ธ์ ๋ํ ๋ฏธ๋ฌํ ๋ถํธํจ๊ณผ ์๋ฌธ์ด ๊ฐ์ง๋์์ต๋๋ค. ์์ ๋ค์๋น ์๋ด๋ํ์ธ ์กด ํ (John Thune)์กฐ์ฐจ ๊ธฐ์๋ค์๊ฒ โํ์ ๋ถ๊ฐ ๋ตํด์ผ ํ ๋ง์ ์ง๋ฌธ์ด ์์ ๊ฒโ์ด๋ผ๊ณ ์ธ๊ธํ๋ฉฐ ์ฐ๋ ค๋ฅผ ํํ์ต๋๋ค. ์ด๋ ํธ๋ผํ์ ๋น๋ด ์ธ์ฌ๋ค์กฐ์ฐจ ์ด ๊ธฐ๊ธ์ ๋ํด ์ด๊ดํ๊ฑฐ๋ ์๋์ค๋ฌ์ํ์ง ์๋๋ค๋ ์ ํธ๋ก ํด์๋ฉ๋๋ค.
์ฌ์ง์ด ํธ๋ผํ ํ์ ๋ถ ๋ด๋ถ์์๋ ๊ท ์ด์ ์กฐ์ง์ด ๋ํ๋ฌ์ต๋๋ค. ์ด ๊ธฐ๊ธ ์กฐ์ฑ์ ๊ด์ฌํ ์์ ์ด์๋ ์ฌ๋ฌด๋ถ์ ์ต๊ณ ๋ณํธ์ฌ๋ ๊ธฐ๊ธ ๋ฐํ ์งํ์ธ ์์์ผ์ ๊ฐ์์ค๋ฝ๊ฒ ์ฌ์ํ์ต๋๋ค. ์ด๊ธฐ ๋ณด๋์ ๋ฐ๋ฅด๋ฉด, ๊ทธ์ ์ฌ์์ ์ด ๊ธฐ๊ธ ์กฐ์ฑ์ ๋ํ ๋ฐ๋ฐ ๋๋ฌธ์ธ ๊ฒ์ผ๋ก ์๋ ค์ก์ต๋๋ค.
ํธ๋ผํ ๊ฐ์ธ์ ์ด๋๊ณผ ํฅํ ์ ๋ง
๊ทธ๋ ๋ค๋ฉด ํธ๋ผํ ์ ๋ํต๋ น ๋ณธ์ธ๋ ์ด ๊ธฐ๊ธ์ ์ํ์๊ฐ ๋ ๊น์? ๋ฒ๋ฌด๋ถ๋ ํธ๋ผํ์ ๊ทธ์ ๋ ์๋ค, ๊ทธ๋ฆฌ๊ณ ํธ๋ผํ ์กฐ์ง์ ์ด ๊ธฐ๊ธ์์ ์ง์ ๋์ ๋ฐ์ง ์์ ๊ฒ์ด๋ผ๊ณ ๋ฐํ์ต๋๋ค.
๊ทธ๋ฌ๋ ํ์์ผ์ ๋ฐํ์ง ๋ ๋ค๋ฅธ ํฉ์ ์กฐํญ์ ๋์ฑ ์ถฉ๊ฒฉ์ ์ ๋๋ค. IRS๊ฐ ํธ๋ผํ์ ๊ทธ์ ๊ฐ์กฑ, ๊ทธ๋ฆฌ๊ณ ๊ด๋ จ ์ฌ์ ์ฒด์ ๋ํ ๋ชจ๋ ์ธ๋ฌด ๊ฐ์ฌ(audit)๋ฅผ ์ค๋จํ๊ธฐ๋ก ํ ๊ฒ์ ๋๋ค. ์ด๋ ๋ํต๋ น์๊ฒ ์ ๊ณต๋๋ ๋งค์ฐ ์ด๋ก์ ์ด๊ณ ํน๋ณํ ์๋ณด์ ๋๋ค. ํ์ฌ ํธ๋ผํ๊ฐ ์ด๋ค ๊ฐ์ฌ๋ฅผ ๋ฐ๊ณ ์๋์ง ๋ช ํํ์ง ์์ง๋ง, ์ด ์กฐํญ์ ๊ทธ์๊ฒ ์์ฒ๋ง ๋ฌ๋ฌ์ ๋ฌํ๋ ์ง์ ์ ์ธ ์ฌ์ ์ ์ด์ต์ ๊ฐ์ ธ๋ค์ค ์ ์์ต๋๋ค. ๋ง์ฝ IRS๊ฐ ๊ทธ์๊ฒ ๋ ๋ง์ ์ธ๊ธ์ ๋ถ๊ณผํ ๊ณํ์ด์๋ค๋ฉด, ์ด ํฉ์๋ก ์ธํด ๊ทธ๋ ๊ทธ ์ธ๊ธ ์ฒญ๊ตฌ๋ฅผ ํผํ ์ ์๊ฒ ๋ ๊ฒ์ ๋๋ค. ๋ด์ํ์์ค ๋ณด๋์ ๋ฐ๋ฅด๋ฉด, ๊ณผ๊ฑฐ ํธ๋ผํ๋ IRS๊ฐ ์น์ํ์ ๊ฒฝ์ฐ 1์ต ๋ฌ๋ฌ ์ด์์ ์ถ๊ฐ ์ธ๊ธ์ ๋ฉ๋ถํด์ผ ํ ์๋ ์๋ ๊ฐ์ฌ๋ฅผ ๋ฐ์ ๋ฐ ์์ต๋๋ค. ์ด๋ฒ ํฉ์๋ IRS๊ฐ ํธ๋ผํ ์ผ๊ฐ์๊ฒ ๋ฒ๋ฅ ์ ๋ฐ๋ผ ๋ฉ๋ถํด์ผ ํ ์ธ๊ธ์ ์๊ตฌํ ๊ธฐํ๋ฅผ ๋ฐํํ ๊ฒ์ด๋ ๋ค๋ฆ์์ต๋๋ค.
์ํ๋ ๋ค๋ฅธ ๊ธฐ๊ด์ด ์ด ๊ธฐ๊ธ ์กฐ์ฑ์ ๋ง์ ์ ์์๊น์? ํ๋ฒ์ ๋ฐ๋ผ ์ํ๋ โ์์ฐ ์งํ๊ถ(power of the purse)โ์ ๊ฐ์ง๊ณ ์์ต๋๋ค. ์ด๋ ์ด๋ก ์ ์ผ๋ก ์ํ๊ฐ ๋ฒ๋ฅ ์ ํฉ์์ ์ฌ์ฉ๋๋ ์๊ธ ์ง์ถ ๋ฐฉ์์ ๋ณ๊ฒฝํ๊ฑฐ๋, ์ด ํน์ ์ํฉ์ ํด๊ฒฐํ๊ธฐ ์ํ ๋ฒ๋ฅ ์ ํต๊ณผ์ํฌ ์ ์์์ ์๋ฏธํฉ๋๋ค. ๊ทธ๋ฌ๋ ์ด ๊ถํ์ ํธ๋ผํ ํ์ ๋ถ ํ์์ ์๋น ๋ถ๋ถ ์นจ์๋์์ผ๋ฉฐ, ์ค์ ๋ก ์ํ๊ฐ ์ด๋ฌํ ์กฐ์น๋ฅผ ์ทจํ ์ง๋ ๋ถํ์คํฉ๋๋ค.
๋๋ค์์ ์ผ๋ฐ ๋์ค์ ๋ฒ๋ฌด๋ถ๊ฐ ์ด๋ฌํ โํ๊ฒฐ ๊ธฐ๊ธโ์ ๊ฐ์ง๊ณ ์๋ค๋ ์ฌ์ค์กฐ์ฐจ ์์ง ๋ชปํ์ ๊ฒ์ ๋๋ค. ์ด ์๊ธ์ด ํ์ฌ์ ๊ฐ์ ๋ฐฉ์์ผ๋ก ์ฌ์ฉ๋๋ ๊ฒ์ ๋ํด ์ฐ๋ฆฌ๋ ์ด๋ป๊ฒ ๋ฐ์๋ค์ฌ์ผ ํ ๊น์? โ๋ ๋ฐ์ผ๋ฆฌโ์ ์ค๋ ๋๋ ๊ธฐ์๋ ์ด๋ฅผ ํธ๋ผํ๊ฐ ์ฌ์ง๊ถ ์ ์ด๋ฃจ๊ณ ์ ํ๋ ๊ด๋ฒ์ํ ๋ชฉํ์ ๋งฅ๋ฝ์์ ๋ด์ผ ํ๋ค๊ณ ๋งํฉ๋๋ค. ๊ทธ์ ์ฃผ์ ๋ชฉํ ์ค ํ๋๋ ์์ ์๊ฒ ๋ฐ๋ํ๋ ์ฌ๋๋ค๊ณผ ๊ธฐ๊ด์ ๋ํ ๋ณด๋ณต์ด์์ต๋๋ค. ์ด ๊ธฐ๊ธ์ ์ฌ์ฉ์ ํธ๋ผํ ํ์ ๋ถ๊ฐ ๋จ์ํ ์ ์ ์ฒ๋ฒํ๋ ๊ฒ์ ๋์ด, ์ด์ ๊ทธ์ ๋๋งน์๋ค์๊ฒ ๋ณด์์ ์ ๊ณตํ๋ ๋ค์ ๋จ๊ณ๋ก ๋์๊ฐ์์ ์๋ฏธํฉ๋๋ค.
์ผ๋ถ ์ค๋ฆฌ ์ ๋ฌธ๊ฐ์ ๊ฐ์ ๋จ์ฒด๋ค์ ์ด๋ฅผ โ๋ฏธ๊ตญ ์ญ์ฌ์ ๊ฐ์ฅ ๋ถํจํ ํ์โ๋ผ๊ณ ๋น๋ํ๊ณ ์์ต๋๋ค. ๋ฒ๋ฌด๋ถ๋ฅผ ์ ์ ์ฒ๋ฒํ๋ ๋ฐ ์ฌ์ฉํ๋ ๊ฒ์ ๋์ด, ๋๋งน์๋ค์๊ฒ ๋ณด์ํ๋ ๋ฐ ์ง์ ์ ์ผ๋ก ์ฌ์ฉํ๋ ๊ฒ์ด๊ธฐ ๋๋ฌธ์ ๋๋ค. ํนํ, ๋ํต๋ น์ ์ ๊ฐ์ธ ๋ณํธ์ฌ์ธ ํ ๋ ๋ธ๋์ฌ๊ฐ ์๋ช ํ 5๋ช ์ ์์๋ค์ด ์ด ๋ง๋ํ ์๊ธ์ ๋๊ตฌ์๊ฒ ์ผ๋ง๋ ์ง๊ธํ ์ง์ ๋ํด ๊ฑฐ์ ๋ฌด์ ํ์ ์ธ ์์ ๋ฅผ ๊ฐ์ง ๊ฒ์ด๋ผ๋ ์ ์ ์ฌ๊ฐํ ๋ฌธ์ ์ ๋๋ค. ์ด๋ ๋ฏธ๊ตญ ๊ตญ๋ฏผ์ด ์ธ๊ธ์ผ๋ก ๋ฉ๋ถํ๊ณ ์ ๋ถ๊ฐ ์ฐ๋ฆฌ๋ฅผ ๋์ ํ์ฌ ์ฐจ์ ํ ๋์ด ๋จ์ํ ๋ํต๋ น์๊ฒ ์ถฉ์ฑํ๋ ์ฌ๋๋ค์๊ฒ ์ง๊ธ๋ ์ ์์์ ์๋ฏธํฉ๋๋ค. ์์ฒ, ์๋ง, ์์ญ๋ง, ์ฌ์ง์ด ์๋ฐฑ๋ง ๋ฌ๋ฌ๊ฐ ํ๊ธ์ผ๋ก ์ง๊ธ๋ ์ ์๋ ๊ฒ์ ๋๋ค. ์ด๋ ์ ๋ง ๊ธ์ง์ ์ด๊ณ ์ถฉ๊ฒฉ์ ์ธ ์ํฉ์ ๋๋ค.
๊ฒฐ๋ก
ํธ๋ผํ ์ ๋ํต๋ น์ IRS ์์ก ์ฒ ํ์ โ๋ฌด๊ธฐํ ํผํด ๋ณด์ ๊ธฐ๊ธโ ์กฐ์ฑ์ ๋จ์ํ ๋ฒ์ ํฉ์๋ฅผ ๋์ด, ๋ฏธ๊ตญ์ ๋ฒ์น์ฃผ์์ ์ ๋ถ ๊ธฐ๊ด์ ๋ ๋ฆฝ์ฑ์ ๋ํ ์ฌ๊ฐํ ๋์ ์ผ๋ก ํ๊ฐ๋ฉ๋๋ค. ๋ฉ์ธ์์ ๋์ด ์ ์น์ ๋ณด๋ณต๊ณผ ๋๋งน์์ ๋ํ ๋ณด์์ด๋ผ๋ ์ฌ์ ๋ชฉ์ ์ผ๋ก ์ฌ์ฉ๋ ์ ์๋ค๋ ๊ฐ๋ฅ์ฑ์ ๋ฏผ์ฃผ์ฃผ์์ ๊ทผ๊ฐ์ ํ๋๋ ์ํํ ์ ๋ก๊ฐ ๋ ์ ์์ต๋๋ค. ์ด ๊ธฐ๊ธ์ ์๋ ๋ฐฉ์๊ณผ ์ํ ๋์์ ๋ํ ํฌ๋ช ์ฑ ํ๋ณด, ๊ทธ๋ฆฌ๊ณ ์ํ์ ๊ฐ์์ ๊ฐ์ ์ด ์ ์คํ ์๊ตฌ๋๋ ์ํฉ์ ๋๋ค. ๊ณผ์ฐ ๋ฏธ๊ตญ์ ์ด ์ ๋ก ์๋ ๋ ผ๋ ์์์ ๋ฒ์น์ ๋ฏผ์ฃผ์ ๊ฐ์น๋ฅผ ์ํธํ ์ ์์์ง, ์ ์ธ๊ณ์ ์ด๋ชฉ์ด ์ง์ค๋๊ณ ์์ต๋๋ค.