Lots of people speak about AI agents and how you can use them to build your company and scale. Very few talk about how to actually do it and how to think about it.

<div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe10e82ea-23e7-4b86-8c11-0f4481660097_1536x1024.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/bfa70cef93c21a8e89c9732a0cf36b3f.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a></figure></div><p>First of all, as we transition to building companies with agents rather than humans, we need to rethink things from the ground up. This starts with what I call AJDs — Agent Job Descriptions. A standard job description has the following things:</p><ul><li><p>About the company</p></li><li><p>What the role is</p></li><li><p>Responsibility required</p></li><li><p>What an ideal candidate looks like</p></li><li><p>Benefits</p></li></ul><p>However, an Agent Job Description needs to be a little different as we can build this labour, buy it or a combination of these two. While an AJD may be performative in nature, it can help clarify your own thinking about what your needs are for this agent.</p><p>So what does an Agent Job Description look like? The following:</p><ul><li><p>What is the outcome that the agent needs to achieve?</p></li><li><p>What context does it require to succeed?</p></li><li><p>What are the liveliness, security and privacy requirements?</p></li><li><p>What is your budget for the agent to run?</p></li></ul><p>Before jumping into examples, lets explore the landscape of the approaches we can take to build the agent.</p><div><hr></div><h2>The Landscape</h2><p>When deciding how to build your agent, you need to be able to make trade-offs between complexity, capability and context.</p><div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffe265bdf-b347-49c7-9257-c6404aed6a4a_940x788.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/f2f0a951e1b3586402484c3dfa42e642.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a><figcaption>The agent design trade-off triangle: capability, context, and complexity.</figcaption></figure></div><p>Some examples of how you would need to balance these things:</p><ul><li><p>An agent that converts your voice memos to draft blog posts would be:</p><ul><li><p>Low capability, low context and low complexity</p><ul><li><p>It simply converts text → speech</p></li><li><p>It doesn’t need context outside of the voice memo</p></li><li><p>Isn’t very complex with how you construct it</p></li></ul></li></ul></li><li><p>An agent that goes through your Telegram messages and reads things for you would be:</p><ul><li><p>High capability, mid context and high complexity</p><ul><li><p>It’s able to read messages from anyone that wants to reach out to you</p></li><li><p>It knows only what has been written in the chat</p></li><li><p>Requires a local Telegram client to be instantiated (if you don’t want to have to add a bot to every convo)</p></li></ul></li></ul></li><li><p>An agent that reads someones LinkedIn profile and then crafts a personalised message would be:</p><ul><li><p>Mid capability, mid context and low complexity</p><ul><li><p>Being able to make cold emails can be useful</p></li><li><p>Scraping data from LinkedIn to get context is useful</p></li><li><p>It isn’t very complex in how you actually join these pieces together</p></li></ul></li></ul></li></ul><div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15e17df8-3bd8-4c54-b218-678e59c01748_940x788.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/224d96ba2ee027939039fa7eb860ca5f.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a><figcaption>Positioning agents within the trade-off triangle.</figcaption></figure></div><p>I’m sure I’ll evolve this over time but that’s a high level starting point to think about this stuff.</p><p>Whenever you see all these “here’s how I built an AI agent that replaced my sales team” on LinkedIn, it’s 100% full of shit. If your sales team are click-and-drag AI agents then your sales team probably sucks and your company is dying.</p><p>Good agents require careful thought and a logical understanding of when a human should be in the loop.</p><p>Going back to the point, you broadly have three ways you can build agents:</p><h2>Level 0 - Chat Platforms (low capability, variable context, no complexity)</h2><p>This is your existing ChatGPT conversations, Claude projects or Custom GPTs that you’ve built that can be constructed natively and most of us have used. As you add context to these chats/projects their context grows over time which makes them more valuable/sticky. You should not discount these agents as they are still the best to do what you need.</p><div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4bee0eac-1d90-4ccf-9e16-140106fd6ddb_1020x751.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/66107c9492e3355d0a5ecff8a9608fa5.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a><figcaption>Level 0 AI agents compared by capability, context, and caution. More context helps, but memory fades.</figcaption></figure></div><p>Your only downside with these agents is that the more you speak to them the more they can forget their context, so you need to rely on strong system instructions. I like to configure my Claude, Gemini and ChatGPT with different system prompts so they’re different types of agents I can consult.</p><ul><li><p>ChatGPT is great for the memory feature and o3’s researching capabilities.</p></li><li><p>Claude is great when I don’t want context and someone friendly but doesn’t gas me up and tell me I am a visionary genius who will solve world hunger</p></li><li><p>Gemini is easily the most capable model that I give challenging scenarios to compute EVs of situations. Context is fragile with it though!</p></li></ul><h2>Level 1 - Trigger Platforms (mid capability, low context, low complexity)</h2><p>When you think of agents this is most likely where you’ll start from. Your level 1 agents are based on platforms that say:</p><ul><li><p>If X gets hit, then Y will happen (with maybe a few steps in between)</p></li></ul><p>The best example of this is Zapier, Lindy, Operator by ChatGPT or other <em>simple</em> workflow orchestration tools. These platforms can be intimidating for non-technical users but can be easy to get the hang of once you understand them.</p><p>They are most people’s best shot of making an AI agent that is somewhat useful in their life. However they only fit a narrow use case of:</p><ul><li><p>Things that are hard for non-technical people but valuable in some particular instance</p></li><li><p>Things that are very easy for technical people that don’t want extra complexity</p><div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd72fa21e-47e5-407b-bc9e-cc5e488b775e_1024x1024.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/bbacc1f283b359d319409c7f3cac6b81.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a><figcaption>Each card represents a building block in a no-code automation stack.</figcaption></figure></div><p>I have still have Zapier webhooks that I rely on to do miscellaneous things around the place despite having higher levels of capability.</p><h2>Level 2 - Semi-Code Platforms (high capability, mid context, high complexity)</h2><p>This is when you start getting into things that are more valuable. By far, I’d say that n8n dominates this category in terms of the sheer capability. Semi-code platforms require you to have a good understanding of how APIs work, data structures and data pipelines. In return, the capability they give you is much higher than your standard trigger based workflows.</p><p>The only reason they are ranked as mid-context is their execution engines can’t really handle anything beyond 1000s of rows of data. For real context, you will need tens of thousands if not millions of rows.</p><p>Anything where you want serious and proper context, your semi-code platforms will tap you out. Also you have to remember these platforms cannot “always stay on” to do things and get expensive as you scale (hundreds of dollars very quickly).</p><div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe7de4fd6-89b9-4233-a3fd-3f2ec3935adc_1024x1086.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/ba7bf5a732c732206770a00dc0d4b264.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a><figcaption>Level 2 platforms are not for beginners, they’re for users who can handle data pipelines, system logic, and manual setup.</figcaption></figure></div><h2>Level 3 - Custom Code (high capability, max context, high complexity)</h2><p>We end up full circle where the most capable agents are custom code bases that have the full power of the compute they run on, can access any API on the internet, can always stay on and are efficient as you scale up to the maximum scale of their compute.</p><div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Febdd15da-3d97-44db-ac37-58537a7451f0_1177x1180.png" rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/7458a9871dc1a667448a8f8318e700ea.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a><figcaption>As capability increases, so does the engineering effort required. Custom Code sits at the top for those who can handle the setup.</figcaption></figure></div><p>While you can can kind of vibe-code a bit of this, you need real engineering capability to build these things as they have to get the job done and the context window is beyond what most chat applications can handle. Furthermore this also requires infrastructure deployment and the chops to be able to run it consistently. Even as AI gets more capable, it does not solve architectural decisions which you need to think about when building custom code based agent.</p><p>There’s almost a very simple heuristic to think about this:</p><ul><li><p>The low level platforms</p><ul><li><p>Cost a lot more per $ of compute</p></li><li><p>Are limited in what they can do</p></li><li><p>Don’t make you think about what you want to do</p></li></ul></li><li><p>The higher level platforms</p><ul><li><p>Cost a lot less per $ in compute</p></li><li><p>Are unlimited in what they can do</p></li><li><p>Make you think about every aspect of the agent</p></li></ul></li></ul><p>This is why the Agent Description is extremely useful as it helps you size out what level you need for the task at hand. Agents that don’t need a lot of context or scale should be delegated to lower levels. Agents that are critical, need precision and handle scale will need custom builds that no platform can give you out of the box.</p><div><hr></div><h2>Closing</h2><p>Even though influencers on social media make it easy to believe that AI will magically solve all your problems, there is still a lot of thinking that needs to be done to make these things work. I like to best think of them as “mental leverage”. It doesn’t dictate how to steer your mind, but it lets you go 100x faster in whatever direction you do choose. Luckily, you can also learn anything you’d like with these incredible tools. Non-technical → technical is just a few months of prompting now.</p><p>The world is open to those who have the grit, determination and stamina to do hard things with technology that makes learning anything easy. Paradoxical in a beautiful way.</p></li></ul>

Pt 2: Scaling an AI Native Company

많은 사람들이 AI 에이전트와 그것을 사용하여 회사를 구축하고 확장하는 방법에 대해 이야기합니다. 하지만 실제로 어떻게 해야 하는지, 어떻게 생각해야 하는지에 대해 이야기하는 사람은 매우 드뭅니다.

<div><figure><a href="https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpu..." rel="nofollow"><div><picture><source><img src="https://static.fwimg.io/img/feed/bfa70cef93c21a8e89c9732a0cf36b3f.jpg" alt=""></source></picture><div><div><div></div><div></div></div></div></div></a></figure></div><p>우선, 인간 대신 에이전트로 기업을 구축하는 과도기에 접어들면서 우리는 모든 것을 근본부터 다시 생각해야 합니다. 이는 제가 AJD(에이전트 직무 설명서)라고 부르는 것부터 시작됩니다. 표준 직무 설명서에는 다음과 같은 내용이 포함됩니다:</p><ul><li><p>회사 소개</p></li><li><p>역할</p></li><li><p>필요한 책임</p></li><li><p>이상적인 지원자상</p></li><li><p>혜택</p></li></ul><p>하지만 에이전트 직무 설명서는 이 노동력을 구축하거나 구매하거나 또는 이 두 가지를 조합할 수 있기 때문에 약간 다르게 접근해야 합니다. AJD는 본질적으로 수행적일 수 있지만, 이 에이전트에 대한 귀하의 필요를 명확히 하는 데 도움을 줄 수 있습니다.</p><p>그렇다면 에이전트 직무 설명서는 어떤 모습일까요? 다음과 같습니다:</p><ul><li><p>에이전트가 달성해야 할 결과는 무엇인가?</p></li><li><p>성공을 위해 필요한 맥락은 무엇인가?</p></li><li><p>생동성, 보안 및 개인정보 보호 요구사항은 무엇인가?</p></li><li><p>에이전트 운영을 위한 예산은 얼마인가?</p></li></ul><p>예시로 들어가기 전에, 에이전트를 구축할 수 있는 접근 방식의 지형을 탐색해 보겠습니다.</p>

(이하 생략)</p></li></ul></li></ul><p>이것이 에이전트 설명이 매우 유용한 이유입니다. 현재 작업에 필요한 수준을 파악하는 데 도움을 줍니다. 많은 컨텍스트나 규모가 필요하지 않은 에이전트는 낮은 수준으로 위임해야 합니다. 중요하고 정밀성이 필요하며 규모를 처리해야 하는 에이전트는 어떤 플랫폼도 즉시 제공할 수 없는 맞춤형 빌드가 필요할 것입니다.</p><div><hr></div><h2>마무리</h2><p>소셜 미디어의 인플루언서들이 AI가 마법처럼 모든 문제를 해결할 수 있다고 믿게 하지만, 이러한 것들을 작동시키기 위해서는 여전히 많은 사고가 필요합니다. 저는 이를 "정신적 지렛대"로 가장 잘 생각합니다. 어떤 방향으로 마음을 조종해야 하는지 지시하지는 않지만, 선택한 방향으로 100배 빠르게 나아갈 수 있게 해줍니다. 다행히도 이러한 놀라운 도구들로 원하는 모든 것을 배울 수 있습니다. 비기술적인 것에서 기술적인 것으로의 전환은 이제 몇 달의 프롬프팅만 있으면 됩니다.</p><p>기술을 통해 어려운 일을 할 수 있는 끈기, 결단력, 인내심을 가진 사람들에게 세상은 열려 있습니다. 아름다운 방식으로 역설적입니다.</p></li></ul>

2부: AI 기반 기업 확장

연방준비제도가 널리 예상되었던 25베이시스포인트의 금리 인하를 단행한 후 수요일 오후 주요 디지털 자산은 약간 하락했습니다. 첫 번째 year.Bitcoin (비트코인(BTC))는 ... 에 머물고 있습니다.

연준 금리 0.25% 인하 후 암호화폐 시장 폭락

현재 암호화폐 시장의 강세 분위기가 2026년까지 지속된다면, 여러 디지털 통화가 자본 유입을 늘릴 수 있는 잠재력을 보여줄 것입니다.
특히, 일부 자산은 강력한...

2026년 시가총액 500억 달러 달성할 암호화폐 2종

소닉 가격은 주요 가치 영역에서 횡보하며 매집 조짐을 보이고 있습니다. 숏 포지션이 증가함에 따라, $0.40를 향한 랠리와 숏 스퀴즈가 예상됩니다.
요약
- $0.30 지원...