The workspace question

When you first move past ChatGPT, the question that comes up is some version of: where do I do this? In what app? On what computer? With what setup?

This is what I call the workspace question, and it's surprisingly hard. Hard enough that it's the main reason most professionals never make the move — not because they're not capable, but because nobody has packaged the workspace properly for someone who isn't an engineer.

Let me describe what a real AI workspace currently looks like. Then I'll explain why it's the way it is, and what I'm doing about it.

What a workspace actually is

If ChatGPT is a tab in your browser, an AI workspace is closer to a workshop. It's a place where:

• You have one or more AI models (Claude, ChatGPT, Gemini) wired into a single environment
• The AI can see and interact with your files, calendar, email, and the other tools you actually use
• You can run small agents in the background — repeating tasks, watching for events, doing routine work
• Everything stays private to you (or to your team), not floating around in a vendor's chat history

That description sounds simple. The current reality is that setting it up requires some combination of:

• Installing a command-line tool called Claude Code (or Codex, depending on the day)
• Configuring something called MCP servers — small bridges between AI models and your tools
• Editing JSON files
• Running terminal commands
• Generally being comfortable with the kind of things engineers do all day

This is not a small ask of someone who became a doctor or a lawyer or a CEO precisely because they were good at things that aren't engineering.

Why it's like this

The honest answer: AI tooling is still being built by engineers, for engineers. The current generation of workspace tools is powerful, flexible, and quite well-designed — but for a developer audience.

This is normal. Almost every transformative technology starts in a developer's hands and slowly gets repackaged for everyone else. The web was a Unix curiosity for years before it had a Geocities. The personal computer was a hobbyist kit before it had a Macintosh.

We are roughly at the powerful but unfriendly stage with AI workspaces. The capability is there; the package isn't.

What you can actually do today

If you're willing to push through the friction once, the current generation of tools is genuinely powerful. You can:

• Install Claude Code and use it as your AI workspace
• Connect it to your file system, calendar, email, Notion, and various business tools through MCP servers
• Build small agents that handle real work for you in the background
• All of this without writing more than a handful of lines of configuration

I'll walk through specific setups in upcoming posts — one for a small medical practice, one for a solo lawyer, one for a small business — because the right shape changes a lot depending on the work. The principles are the same; the integrations and the rules of the road are not.

Two paths through the wall

Most of the people I work with one-on-one want the workspace, but don't want to spend a weekend learning to install command-line tools. Reasonably so. Their time is their actual asset.

That points to two paths. One is the hand-built workspace described above — visible, controllable, every component yours. The right fit for the people who want to see and shape the underlying machinery, and who get value from understanding it.

The other path is a packaged workspace: tools connected through a normal web interface, agents and integrations already wired, no machinery to manage. Less control, far less friction. The shape of products in this category is still being worked out, but it's where most non-technical professionals will eventually meet AI — not through a terminal, but through a workshop they walk into.

Neither path is wrong. They serve different people. The point is that both paths exist now, and the wall between most professionals and the deeper layer of AI is finally lower than it has ever been.

The takeaway is this:

The workspace question is the real wall between most professionals and the deeper layer of AI. The wall is being lowered, slowly, but it's still there. If you want to climb over it the hand-built way, the tooling exists. If you'd rather wait for the door, that one is being built too.