/
Published on 03.03.2026
TLDR: Most people are paying a hidden "copy-paste tax" — using AI in a chat window and then manually shuttling outputs between tools. The real shift isn't picking the right AI tool; it's consolidating your work into environments where AI can read, write, and act directly. The author maps out three levels of AI integration and shares how moving to Claude Code plus Obsidian eliminated the middleman problem.
There is a brutally honest observation buried in this Substack post from AI Maker, and it is one that most AI power users will recognize but few want to admit: you have become a human API. You brainstorm in ChatGPT, copy to Notion. Research in NotebookLM, paste into Docs. Draft in Claude, move to Substack. Generate social posts, copy to a scheduler. You are not thinking or creating — you are routing data between systems that refuse to talk to each other.
The author calls this the "copy-paste tax," and it is a perfect name for a real phenomenon. Every time you Command+C, Command+V an AI output into another tool, you are paying a cognitive toll that has nothing to do with the quality of your thinking. You are doing integration work that software should be handling.
What makes this particularly insidious is that it feels productive. You are using AI every day. Ideas are flowing. Content is getting drafted faster. But if you actually audit where your time goes, a disturbingly large chunk is spent deciding where things should go, how to format them, and how to connect this output to everything else. That is not leverage — that is overhead with extra steps.
The honest question the author eventually lands on is not "which AI tool should I use?" but "where does my AI actually live?" And that reframing matters more than any feature comparison between ChatGPT, Claude, or Gemini.
In Pursuit of Agentic AI Workspace
The article lays out a useful mental model with three levels. Level 1 is chat-based AI: you ask questions, get answers, and manually move them somewhere useful. Every conversation starts from zero. Level 2 is integrated AI: tools like Gemini inside Google Workspace or Claude with MCP connections that can read and write inside your actual work environment. Level 3 is agentic AI: the system chains actions across multiple tools without you orchestrating each step.
The litmus test for Level 3 is sharp: "Can your AI read your meeting notes, draft a follow-up email, update your project tracker, and flag what needs your attention — without you copy-pasting between any of those steps?" If you are still the one connecting dots between tools, you are at Level 2 at best.
Here is what the author is not saying explicitly but is worth pointing out: most people are firmly at Level 1 and do not realize it. Even people who pay for ChatGPT Pro or Claude Pro are overwhelmingly using these tools as sophisticated chat windows. The gap between Level 1 and Level 2 is not about intelligence or willingness — it is about the friction of setup. Connecting MCP servers, configuring file system access, consolidating data into repositories — these are nontrivial steps that require technical comfort.
The article also takes a notable swipe at OpenAI here, saying their lack of strong multi-app integration has caused the author to move away from them. That is a real competitive vulnerability. OpenAI has the brand recognition and the model quality, but if Claude and Gemini own the integration layer — the place where AI meets your actual files and tools — the chat window alone will not be enough.
Notion's recent launch of custom AI agents that run autonomously inside workspaces is another interesting development mentioned here. For teams already living in Notion, this represents a genuine path to Level 3 without switching tools. The question is whether Notion's agents can match the flexibility of a general-purpose tool like Claude Code operating on raw files.
In Pursuit of Agentic AI Workspace
The most concrete part of the article is the author's description of consolidating an entire newsletter operation into a single repository — published posts, drafts, content ideas, social media archives, writing guidelines, performance data, research notes — all as plain files in folders. Then pointing Claude Code at the whole thing.
This is a genuinely powerful pattern and one worth thinking about carefully. When your AI agent can read every file in a repository and understand relationships between them, you stop asking "Claude, help me brainstorm ideas" and start saying "Read my top-performing posts and my writing guidelines, then draft social variants that match my voice." The AI writes outputs directly into the correct folder. No intermediary steps.
The author extended this further by connecting Claude Code to NotebookLM through MCP, creating a pipeline where research feeds directly into drafts without alt-tabbing. And then moved personal productivity — replacing Notion, Todoist, and Google Sheets — into an Obsidian vault connected to Claude Code.
The results described are compelling: Claude flagged that the same project had been deferred for three consecutive weeks. It caught a blocker that was about to collide with a scheduled announcement. It pulled a line from the author's own weekly review about parallel task-switching ruining focus and used it to restructure the next week's plan.
But let me push back on something the article glosses over. This approach has a significant prerequisite: you need to be comfortable with plain text files as your source of truth. Markdown files in a Git repository or an Obsidian vault are fantastic for solo knowledge workers and developers. They are a much harder sell for teams that depend on real-time collaboration, permissions, shared dashboards, and structured databases. The "consolidate into plain files" advice is genuinely transformative for the right person and completely impractical for others.
There is also a dependency risk worth naming. The author's entire workflow now routes through Claude Code. If Anthropic changes pricing, rate limits, or capabilities, that workflow breaks. When your AI is "the glue" instead of you being the glue, you have traded one form of coupling for another. The difference is that the new coupling is faster — but it is still coupling.
In Pursuit of Agentic AI Workspace
The real bottleneck is not AI quality — it is integration. Most people waste more time moving AI outputs between tools than they save from using AI in the first place. Audit where your copy-paste friction actually lives before evaluating new AI tools.
Consolidation beats orchestration. Rather than connecting five tools with complex automations, moving your work into a single environment that AI can access directly (a repository, a vault, a workspace) eliminates the integration problem at the root.
Claude Code's auto-memory creates compounding returns. The system gets smarter session after session because it retains preferences, patterns, and decisions. This is fundamentally different from chat-based AI that starts from zero every time.
Notion's agentic AI is the enterprise counterpart. For teams already in Notion, custom AI agents that run on schedules and connect to Slack, email, and other tools offer a path to Level 3 without abandoning existing workflows.
The "where should my AI live?" question matters more than "which AI should I use?" The tool comparison game is a distraction. The real decision is about where your work consolidates and whether AI has access to that environment.
Start with your highest-friction workflow, not a complete overhaul. The author's system grew from solving one specific pain point (newsletter production). Trying to build an agentic workspace from scratch is the wrong approach — find the workflow that hurts the most and fix that first.