Build Your Permanent Claude Workstation

## Onboarding a New Hire Every Morning

[A] "So you sit down at your desk, right? You paste a perfectly crafted campaign brief into Claude, hit enter, and the very first sentence it spits back at you is: 'In today's rapidly evolving digital landscape, it's crucial to delve into whatever.'"

[B] "Oh, yeah. The absolute worst."

[A] "Right. You just immediately roll your eyes and hit delete. The AI knows the client's product perfectly because you just gave it the brief — but it has absolute total amnesia about you. It doesn't know you despise the word 'delve,' or your agency's preferred tone, or even your priorities for this quarter. All that actual context for how you operate still lives entirely locked inside your head."

[B] "It really is the ultimate Groundhog Day scenario for marketers and creators. You are basically onboarding a brilliant, infinitely capable new hire every single morning. They show up eager to work, but they remember literally nothing about how you did things yesterday. They don't intuitively know that you always lead a pitch with hard data, or that your agency is actively trying to pivot away from a certain industry right now."

[A] "So you spend half your day just re-explaining your basic boundaries. Which means you end up in this exhausting loop, just correcting the same mistakes, deleting the exact same robotic adjectives over and over. So today, we are fixing that. We are pulling from a highly practical, step-by-step text guide designed to completely solve this problem — to fundamentally transform the tool from an amnesiac assistant into a seasoned colleague who actually knows your daily workflows, your writing style, and your professional boundaries flawlessly."

## The Three-File Architecture

[B] "Okay, let's unpack this. Because the solution laid out in our guide relies on setting up exactly three text files and making one configuration change. But the first thing that jumped out at me is where the guide says we have to put these files. The location is surprisingly specific — it says to create a folder structure directly in Finder on your computer, living right alongside your co-work interface."

[A] "But wait. My entire workflow — my whole digital brain — lives in a note-taking app like Obsidian. Why does the source insist I can't put this memory system there? Wouldn't a cloud-based note app be way easier?"

[B] "Well, it seems counterintuitive at first, but it comes down to how the specific interface interacts with your local machine. You've been using Claude Desktop throughout this series — and by now, with your vault configured, your skills running, and Claude connected to your local files, you've essentially evolved that setup into what we'll call Claude Cowork: a purpose-built AI workspace for marketing professionals. The core philosophy is the same; the name reflects how far you've come. And at this level, speed of file access really matters. If you bury your core identity files inside a proprietary database like Obsidian or Notion, Claude Cowork often struggles to fetch that data instantaneously."

[A] "Ah, I see. So it's about speed and access."

[B] "Exactly. By putting a dedicated About Me folder right in Finder, you are giving Claude Cowork a direct, unobstructed, local pathway to read your rules the millisecond it boots up. And to be clear — this doesn't replace your Obsidian vault. Think of the vault as your client-facing knowledge base, and this About Me folder as the personal operating layer that runs underneath all of it. We create the main folder in Finder called About Me, and inside that we are going to have three markdown files, plus two subfolders labeled Outputs and Templates."

[A] "But I have to push back on the architecture itself for a second. Why three separate markdown files? If this co-work interface can read my local drive, why not just write one massive master document titled 'How I Work,' dump my entire bio into it, and let the AI sort through it?"

[B] "Because treating the AI like a messy filing cabinet creates a massive cognitive load for both you and the machine. If you dump your core personality, your hated buzzwords, and your current quarterly goals all into one massive document, it becomes a nightmare to maintain. Your core identity doesn't change much year to year, but your active projects change every month. By forcing this into three distinct files, the AI parses different types of constraints separately. And more importantly, you only have to update the specific piece of your workflow that is actually changing."

## aboutme.md: Let the AI Interview You

[A] "Got it. So let's look at what goes into these distinct files. The first one is simply called aboutme.md."

[B] "The source says to keep this under 2,000 tokens — roughly a couple of pages of text. It's the big picture. Your work philosophy, your operational rules, basically how you want the AI to behave as your partner. But the most actionable tip in the entire guide is about how you generate this file. The author explicitly warns against sitting down at a blank screen and trying to write your own bio."

[A] "Because you just freeze up, or leave out the obvious stuff."

[B] "Totally. If you try to list your own operational rules, you will inevitably leave out the crucial, invisible habits that actually define your work. Instead, you turn the tables — you prompt Claude to interview you. You open a fresh chat and say: 'I need to build my aboutme.md file to serve as your permanent memory. Interview me with 20 questions covering my professional identity, my work philosophy, my quality standards, and my operational rules. Ask me one question at a time, wait for my answer, and then ask the next.'"

[A] "That is so smart. I imagine it asking something like, 'When you present a new marketing campaign, do you prefer to start with a creative concept or the underlying data?'"

[B] "Right. And you just casually answer, 'Oh, always the data — start with the problem, then the creative solution.' And boom, that becomes a permanent operational rule. Every single answer is synthesized by the AI into a structured profile that it will carry into every future interaction. It totally removes the burden of introspection from you."

## anti-ai-writing-style.md: The Quarantine Zone

[A] "Once that's done, we move to the second file — the guide calls it anti-ai-writing-style.md."

[B] "I see this as the quarantine zone. As a marketer, I feel like this file alone justifies the entire system. This is where we lock away those obvious AI clichés — phrases like 'harness the power of,' 'delve into,' or the classic adjective stacks where everything is a 'vibrant, dynamic, and multifaceted tapestry.' Those phrases are the hallmarks of a lazy, zero-context generation."

[A] "The brilliance of this second file is that it acts as a living immune system for your workflow. When you are working on a project and Claude spits out a robotic transition phrase you despise, you don't just delete it and rewrite the sentence — you extract the pattern."

[B] "So instead of just getting annoyed, you analyze why it annoyed you. If it writes 'In conclusion, this strategy will...' and you realize you hate obvious transitional announcements — you open the anti-ai-writing-style file and literally add a bullet point: 'Avoid sentence openers that announce what is coming, like in conclusion or first let's examine.' You extract the underlying pattern, name it, and add it to the quarantine file."

[A] "Over a few weeks, this file becomes a highly tuned filter. You essentially build an automated constraint that prevents the AI from ever making that specific stylistic mistake again. Incredible."

## mycompany.md: Your Current Reality

[B] "And then we have the third piece of the puzzle: mycompany.md. The guide is very strict about this one — keep it under a thousand tokens. Unlike the first two files, which are about your stable identity and your permanent dislikes, this file represents your current, immediate reality. It must contain exactly three things: three current goals with real metrics, two active focus areas, and two deliberate no-decisions — basically what you are not doing this quarter."

[A] "And those no-decisions represent a massive tactical advantage. When we brief an AI, we almost always focus on what we want it to do. But outlining what you are explicitly not doing this quarter creates incredibly strong guardrails. If my agency usually does all kinds of tech marketing, but right now we are hyper-focused on B2B software, I would explicitly put in this file: 'We are saying NO to direct-to-consumer e-commerce pitches this quarter.' What does telling the AI what we aren't doing actually achieve on a tactical level?"

[B] "Oh, exactly how the AI processes that constraint is fascinating. By explicitly stating a NO, you are radically shrinking the probability space the model has to search through. It stops the AI from hallucinating irrelevant e-commerce strategies or wasting valuable compute tokens on ideas that just don't align with your immediate priorities. You date this file, and you update it on the first of every month. Your core identity stays stable. Your style guide grows as you find new things to hate. But this third file keeps the AI anchored to what matters today."

## The Silent Bouncer: Global Instructions

[A] "All right. So the architecture is built — three files sitting in our Finder folder. But this brings up a massive logistical problem. Having an employee handbook on the desk is completely useless if the employee refuses to open it. If I have to start every single morning by typing out, 'Hey Claude, please navigate to my local drive, read my About Me file, then read my anti-AI style file, then read my Company file...' haven't we just replaced one frustrating repetitive task with another?"

[B] "If you were doing it manually, yes, the system would collapse under its own friction. But the guide reveals a configuration step within the co-work interface that bypasses manual prompting entirely. You go into the application settings, and you edit what are called the global instructions."

[A] "Ah, okay. So these global instructions act like a bouncer at a nightclub — they stand at the door of every single new chat window and enforce the rules before the conversation is even allowed to start."

[B] "Exactly. In the settings, you paste a specific command string: 'Before every task, you must read the About Me file, the anti-AI writing style file, and the My Company file.' But the most critical part of this command is the behavioral constraint you add at the end — you tell it: 'Apply everything in those files to your responses. Do not summarize what you've read.'"

[A] "Oh. Do not summarize. That is key."

[B] "If you don't explicitly tell the bouncer to keep his mouth shut, he's going to read the entire rulebook out loud to you every time you walk into the club. Without that command, the AI will waste a massive amount of space spitting a summary of your own identity back to you before it even begins drafting the brief you asked for. And the instructions also tell Claude to only access your Outputs or Templates subfolders when you explicitly command it to — you don't want the AI autonomously rummaging through your old confidential client deliverables just because it thought they might be relevant."

[A] "What's fascinating here is that this runs silently. Claude reads your three-part profile before it even processes the first letter of your actual prompt. Every single output, from the very first interaction of the day, is deeply contextualized."

## Whisper Flow and Token Management

[B] "But let's talk about the cost of that context. If the AI is secretly reading 3,000 words of my personal rules, style guides, and company goals before I even type hello, isn't that burning through my usage limits?"

[A] "It is a tradeoff — context absolutely costs tokens. But think about the alternative. Without the system, you might spend five or six iterative prompts arguing with the AI, trying to fix its tone, correcting its assumptions about your goals, and removing the word 'delve.' That back and forth burns far more tokens and far more of your actual time than simply front-loading the rules once perfectly. Zero-shot accuracy — getting it right on the first try — is ultimately more efficient."

[B] "Here's where it gets really interesting. The guide points out that once Claude has this perfect permanent memory of your standards loaded up, your physical typing habits suddenly become the biggest bottleneck in your workflow. We need to completely rethink how we input information. And the guide heavily emphasizes a feature called Whisper Flow — a dictation tool."

[A] "Yeah, most people skip this section because they think they type fast enough. But the argument the author makes isn't just about the physical speed of words per minute — it's about a fundamental cognitive shift. When you type a prompt, you are essentially sending a rushed text message. Your fingers get tired, so you unconsciously abbreviate your thoughts. You leave out the nuance, the strategic background, the why behind the task. Whereas speaking a prompt is like leaving a nuanced, detailed voicemail for a trusted colleague."

[B] "The raw math makes this stark: we type at roughly 40 to 60 words per minute, but we speak at 150 words per minute. If I'm typing, I'm probably going to bang out something like 'write a LinkedIn post about our new Q3 marketing report.' Totally generic. But if I'm holding down a hotkey and just talking, I can pour out the actual context: 'Draft a LinkedIn post about the Q3 report — our target audience is CMOs who are exhausted by fluff, keep the tone sharp, heavily index on that surprising data point about customer retention on page four, and make sure the overall vibe is authoritative but not arrogant.'"

[A] "You use the word 'vibe' — and that is exactly what the spoken prompt captures that the typed prompt misses. You are conveying the strategic posture of the piece."

[B] "Okay, so we are front-loading massive context files and now we are dictating huge, sprawling verbal prompts. Won't the system just slow down or get confused by the sheer volume of text? The guide dedicates an entire section to managing the engine through what the source calls the five-hour rolling token window. Most premium AI tiers give you a set allowance of computational power that replenishes over a rolling five-hour period. Every word you send and every word the AI generates eats into that allowance. If you maintain one single endless chat window all day, the AI has to reread the entire history of that conversation, plus your global instructions, plus your three core files, every single time you ask a new question. The token cost grows exponentially with every message."

[A] "So the guide outlines three specific habits to keep the system running efficiently. The first one is fresh sessions every 20 messages. Before you restart, you tell Claude: 'Extract the final polished campaign idea from this brainstorm, strip out the messy back and forth, and save it as a clean document in my templates folder.' Then you start fresh. The AI drops the heavy transcript of the brainstorm, freeing up your token window, but the valuable output is saved securely."

[B] "The second habit is batching tasks. Because typing is tedious, we usually ask the AI to do one thing at a time — 'write the email,' wait, 'now review it for tone,' wait. Every single microprompt forces the AI to reread the entire context. But because you are speaking your prompts now, you give it a comprehensive sequence up front: 'Read the new client brief. Step one, write three distinct options for the campaign. Step two, draft an email pitch for each. Step three, review them against my anti-AI writing style file. Step four, save the best one to my Outputs folder.' One single massive prompt — and the AI executes the entire batch in one unbroken thought process."

[A] "And the third habit is strategic model selection. Don't just use one model for everything. Use Sonnet for the vast majority of your daily work since it is incredibly fast and highly capable. But when you need complex strategy where absolute quality trumps speed, you toggle over to Opus. And for tiny, repetitive micro steps inside automated workflows, you use Haiku. It's exactly like running a team — you do not hand tedious, repetitive data entry work to your brilliant, expensive senior strategist."

[B] "Managing tokens isn't just about software limits — it's about being an effective manager of your AI workforce, assigning the right brain to the right task. When we zoom out and look at this entire architecture — the Finder folders, the three core files, the silent bouncer in the global instructions, the spoken project management prompts — what we're really doing is stopping the endless, exhausting cycle of explaining ourselves to a machine. You are building a permanent foundation so you can actually start doing higher-level strategic work."

[A] "Your homework to get the system running today: first, run that 20-question interview and let Claude pull your aboutme.md file out of your head organically. Second, go find five pieces of AI-generated text you recently received and absolutely hated — extract the underlying patterns and put them straight into your anti-AI-writing-style quarantine file. Third, define your current reality: write your mycompany.md file with three goals, two focus areas, and crucially, define your two no-decisions for this quarter. Fourth, download Whisper Flow or your system's native dictation tool and force yourself to use it for your next three major prompts. Just compare the results."

[B] "Once you perfectly externalize your writing style, your deep-seated dislikes, your strategic goals, and your boundaries into an automated, permanent system — where does the tool end and your own mind begin? If an AI can perfectly execute your specific, highly nuanced taste while you sleep, it completely redefines what it means to be a creator. It's something to think about the next time you ask a machine to write an email. Are you just using a piece of software, or are you collaborating with a digital reflection of yourself?"