Most people chase the best AI model. They switch from Claude to ChatGPT to Gemini every few weeks, rebuilding their setup each time. This video from Grace, a digital growth consultant, flips that completely: the model barely matters, and the thing you should actually be building is the workspace around it.
I watched this one twice because the framing clicked something into place for me. The creator makes the case that frontier models are converging in quality, so the real risk isn’t picking the wrong one. The real risk is being locked into one.
The key idea: model vs harness
She borrows Jensen Huang’s definition of an agent: what separates an agent from a chatbot is the loop. It takes an action, checks the result, keeps going until the job’s done.
So here’s the split that matters:
- The model is the raw intelligence. Swap Claude for GPT and this changes.
- The harness is everything that makes the agent yours. Your brand voice, your campaign data, your workflows, your rules.
Swap the model and the marketing ability shifts a bit. Everything that makes it useful to you stays put, as long as you built the harness properly. That’s the whole thesis.
Old way vs new way
Old way: you set up prompts, custom instructions, and context inside one platform’s interface. Everything lives in that vendor’s account. When something better ships, you start over and re-explain your brand from scratch.
New way: your agent workspace is a folder on your own machine. Plain files. You point whatever platform you like at that folder. The platform becomes disposable. Your work doesn’t.
The five components of a workspace
The expert breaks a portable workspace into five parts:
- Knowledge and context: what the agent knows about you, your brand, your products, your customers.
- Instructions/config files: how the agent should behave, plus the rules it follows in this workspace.
- Agent skills: repeatable workflows the agent can trigger (copywriting, data viz, page audits).
- Sub-agents: specialized role definitions you call when needed, like a data analyst or a growth specialist.
- Tools: the connections to your real systems: analytics, CRM, Search Console.
The three portability buckets 🧳
This is the part I found most practically useful. Once you see the workspace as components, you stop asking “how do I rebuild everything” and start asking “which pieces actually move.”
- Bucket 1, fully portable: your context and knowledge. These are just files and documents. Grab and go, works anywhere.
- Bucket 2, needs light translation: instructions, skills, sub-agents. Same content, but each platform names and stores them slightly differently.
- Bucket 3, needs reconnecting: your tools. MCP connections and integrations have to be re-authorized when you move.
That’s it. Two of the three buckets are basically free to move.
How it looks in practice
The creator walks through three real setups:
- Marketing team, one brand: one shared context folder so every agent pulls from the same brand truth, then separate project folders per channel (paid, content, email, SEO). Skills and sub-agents shared across the team.
- Functional team, say SEO: identical shape, but split by type of SEO work instead of by channel. Goes deeper rather than wider.
- Solo consultant, many clients: all clients in one workspace, each with their own context folder. Skills and agents sit at the top level and get inherited by default. A client only gets their own version when they genuinely need something different.
Inherit by default, override only what’s different. Clean.
The migration demo
Here’s where it gets concrete. She builds a workspace in Claude Code with a data analyst sub-agent, asks it how the site performed in SEO over two weeks, and watches the loop run: pull Search Console data via MCP, call the data visualization skill, generate charts, check them against the brand designer skill, save to the right folder.
Then she moves the whole thing to Codex. Loading the same project folder, Codex detects the Claude configuration and offers a one-click import, mapping skills and config into its own structure. Skills carry over. MCP connections need a manual check.
On the Codex side she builds a growth specialist agent and runs a full homepage audit. Her take: Codex is noticeably faster and more accurate at browser control, and it screenshots the page at multiple resolutions to check UX across devices. Output is a formatted Word doc with findings, fixes, and an action list.
Then back to Claude, using an agent migration skill she built that handles the two-way conversion. It scans the project, proposes a migration plan, offers a backup, and syncs new skills and agents into the Claude structure. She recommends installing it at the user account level so every project can reach it.
One honest caveat she flags: the migration skill doesn’t yet sync agent memory files. Her workaround is pointing both platforms at a single memory source of truth.
Practical steps to start 🛠
- Create one local folder for your marketing workspace.
- Put your brand context, product info, and voice guidelines in a shared context subfolder.
- Add a project instructions file at the root.
- Create a skills folder and drop in reusable workflows.
- Build sub-agents that reference those skills.
- Connect your tools last, since they’re the part you’ll reconnect anyway.
- Always back up before running any migration.
The closing line stuck with me: stop obsessing over model intelligence, and build a foundation that compounds instead. Worth watching the full video for the on-screen folder structure and the migration walkthrough.