Quick answer
To keep your AI agent setup when you switch tools, store the know-how outside the tool. Keep one plain-text setup file with the agent’s job, rules, steps and the corrections you have made over time. Save a few examples of good output and a short note on which apps the agent connects to and why. Update it whenever you correct the agent. When you move, paste the file into the new tool, run the same task side by side and fix the gaps. As Nate B Jones points out, raw data usually moves easily; it is the instructions, corrections and skills that get left behind.
Key takeaways
- Data moves; know-how often doesn’t. Instructions, corrections and skills are usually tied to one tool.
- Keep one setup file per agent in plain text, in a folder you own.
- Log corrections the same day so they reach next month’s work.
- Save two or three examples of good output as a quality benchmark.
- Note each connection: which app, what the agent reads or does, and who approved it.
- Test the move side by side on the same task before you switch fully.
📘 Main article: This guide supports Agentic Control Plane: Why Microsoft, Salesforce and OpenAI Want to Run Your AI Agents. It gives you a practical way to protect the work you put into an AI agent, so changing provider doesn’t mean starting again.
In this guide
After a few months with an AI agent, it starts to feel like it knows how you work. That knowledge is valuable, and it is easy to lose.
This guide answers one question: how do you keep your AI agent setup when you switch tools? It builds on our main article, Agentic Control Plane: Why Microsoft, Salesforce and OpenAI Want to Run Your AI Agents, and on the recruiter story in Nate B Jones’s video.
What gets lost when you switch AI tools?
Mostly the things you taught the agent, not the data it used. In Nate’s example, a recruiter taught her agent how she screens candidates, what each hiring manager wants and who actually got hired. When her company considered changing provider, the candidate data could move. Her instructions, corrections, skills and tool connections would not.
| Part of your setup | Where it often lives | Risk when you switch |
|---|---|---|
| Records and files | Your CRM, drive or database | Low |
| Instructions and rules | Inside the agent’s settings | High |
| Corrections over time | Scattered across old chats | Very high |
| Skills and routines | In the platform’s own format | High |
| Connections to apps | Platform integrations | Medium to high |
How do you write an agent setup file?
Use one plain-text document per agent with five short sections. Plain text works almost anywhere, so it moves with you.
- Job: one or two sentences on what the agent does and for whom.
- Rules: what it must always and never do, including actions that need your approval.
- Steps: the order it works in, written as a numbered list.
- Definitions: what your key terms mean, such as “active customer” or “revenue”.
- Corrections log: dated notes on mistakes it made and the fix.
Definitions matter more than they look. Nate notes that an agent can use the right number with the wrong meaning. Writing your terms down stops that happening in a new tool. If you already use an AGENTS.md file, see how to replace an AI skill with one line in AGENTS.md for a related approach.
💡 Tip: Store setup files in a folder you control, such as your own cloud drive, not only inside the AI tool. Name them clearly, for example “weekly-sales-summary-agent.txt”.
How do you capture corrections as you go?
Write a one-line note every time you correct the agent. Nate warns that last month’s correction doesn’t automatically make it into this month’s brief. Someone has to maintain it.
- Date it. “2 Nov: stopped including cancelled orders in the total.”
- Say what to do instead, not just what went wrong.
- Move repeat fixes into the Rules section once they happen twice.
- Review the log monthly and delete anything no longer true.
This habit also makes the agent better where it is now, because you can paste updated rules back in whenever they change.
Choosing a new platform? Use our guide on how to choose an AI agent platform without getting locked in, and read Agentic Control Plane: Why Microsoft, Salesforce and OpenAI Want to Run Your AI Agents for why this matters now.
Which examples and connections should you save?
Save two or three examples of output you were happy with, plus a short note on each connection.
- Good examples give the new tool a clear target and give you a benchmark to compare against.
- One “bad” example with a note on what was wrong can be just as useful.
- Connection notes: the app, what the agent reads or changes, which permissions it has and who approved it.
⚠️ Warning: Never paste passwords, API keys or private customer details into a setup file. Note which connection is needed, then set up access fresh in the new tool using its own secure settings.
How do you test your setup in a new tool?
Run the same task in both tools before you switch fully.
- Paste the setup file into the new tool’s instructions or skill area.
- Reconnect only the apps it needs, with the same approval rules.
- Run one real task in both tools using the same input.
- Compare against your saved examples. Note what is missing or different.
- Fix the setup file, not just the new tool, so the fix travels with you next time.
- Switch over once the new tool matches your benchmark for a few runs.
🔎 Unconfirmed / rumour: Some platforms say they can import skills or settings from other tools. Formats are changing quickly and aren’t standard yet. Don’t rely on an import feature until you have tested it with your own setup.
What mistakes should you avoid?
- Keeping know-how only in chat history. Old chats are hard to search and often don’t export cleanly.
- Copying every old instruction across. Clean up first; out-of-date rules cause errors.
- Skipping definitions. The new tool will guess what your terms mean.
- Switching in one go. Run both side by side for a short period.
- Not telling your manager what is involved. In Nate’s story, the manager didn’t know how much work the setup held.
Frequently asked questions
Can I export my AI agent’s instructions?
Often you can copy them out manually, even if there is no export button. The safest habit is to keep the master copy in your own document and paste it into the tool.
What is an agent setup file?
It is a plain-text document that describes an agent’s job, rules, steps, key definitions and past corrections, so you can recreate it in any tool.
How often should I update it?
Add a line whenever you correct the agent, and review the whole file about once a month.
Do skills transfer between AI platforms?
Not reliably yet. The ideas transfer if you have written them down, but the platform format may need rebuilding.
Is this worth doing for one agent?
Yes. It takes a few minutes, makes the current agent more consistent and saves hours if you ever move.
The bottom line
The most valuable part of an AI agent is the know-how you put into it, and that is the part most likely to be left behind when you change tools. Keep a plain-text setup file, log corrections as you go, save a few good examples and test any move side by side. Your next step: open a new document today and write the Job and Rules sections for the agent you use most.
Related posts
- Agentic Control Plane: Why Microsoft, Salesforce and OpenAI Want to Run Your AI Agents
- How to Choose an AI Agent Platform Without Getting Locked In
- Claude Skills Clean-Up: Which Skills to Keep, Merge or Delete
[…] How to Keep Your AI Agent Setup When You Switch Tools — how to save your instructions, corrections and skills so they move with you […]