AI Champion Blueprint

A free handout: how to store keys safely, what data you can give AI, how to introduce AI at work and become a champion on your team.


Why this guide

One data leak cancels out any time you save. This Blueprint puts in one place what everyone bringing AI into their job should know: how not to lose keys and passwords, what must never go into a model, how to discuss AI with your manager, and how to turn a personal skill into value for your team.

It is written for people without a technical background. Terms are explained in plain words.

Safety is covered in depth in the offline AI basics module and in the first live session. This is the short version to keep at hand.


Secrets and keys: how to store them safely

A secret is any password, token or access key (an API key is a "password" your program uses to call a service). A secret is also money: with your key someone else can spend your limit.

  • Never paste a key into an AI chat or write it in the instruction files for your agent.
  • Locally, the key lives in a .env file. That file must be listed in .gitignore - the list of things that are not sent to the repository (GitHub). Then the key does not travel to the internet with your project. A good AI will remind you: "put the key in .env".
  • For a published project, keys are stored on the server in the service settings (on Vercel it is Environment Variables), not in the code.
  • One key, one job. Give minimal permissions and revoke a key if it may have leaked.
  • A key in a plain file on your computer is the bare minimum. It works, but if the file leaks, you are in trouble.

Final check. Ask the AI: "Check whether the secrets in this project are stored safely and whether anything should be fixed".


What data you can give AI and what you cannot

Rule zero: ask your management. Before uploading work data to AI, ask your manager, security team or compliance what is allowed. Every company has its own rules. No answer means no, and you work on test data.

Not without explicit permission:

  • Personal data of customers and employees: names, emails, phones, addresses, passport numbers, salaries (the company is legally responsible, for example under GDPR).
  • Passwords, tokens, access to systems.
  • Trade secrets and NDA materials: contracts, private financials, unpublished plans.
  • Regulated data: medical, financial, biometric.
  • A personal AI account for work data, if the company has not approved it.
  • Connecting an agent to production systems: test first, then under supervision.

Do it right:

  • Use a company plan if there is one: business plans usually do not train models on your data. Ask IT about the retention policy.
  • Give minimal access: a separate test folder beats access to everything.
  • Do not switch on "allow everything" without looking.
  • If in doubt, do not send.

How to make safe test data

Most tasks can be practiced without real data. Two ways:

  1. Anonymize the real thing. Ask the AI: "Replace all names, emails and amounts with fictional ones, keep the structure".
  2. Generate from scratch. "Make 20 sample records in this format [example row], realistic but entirely fictional".

Rule: always prototype and demo on test data. Move to real data only when the tool is approved and management has said yes.


How to introduce AI at work and talk about it

  • Start with the rules question, not the tool. "What data am I allowed to use with AI?" is hard to refuse and shows you are responsible.
  • Take a tiny case. Not "let's move to AI" but "let's try it on one small task".
  • Speak in results. Not "AI is great" but "it was 2 hours a week, now it is 15 minutes".
  • Show, do not tell. A 5-minute demo beats a presentation.
  • Do not push. "We must adopt AI" breeds resistance. Interest works: "Want me to show how this works for me?"
  • Say honestly where AI is not needed. It raises trust in everything else.

How to become an AI champion: 6 steps

An AI champion is not the person who knows the most about AI but the one who brings the team measurable value and teaches others.

  1. Find a pain, not an excuse to use AI. A boring regular task that eats time.
  2. Solve it and measure. Before and after. Without a number it is a toy, with one it is an argument.
  3. Show it live. Let people press the button themselves.
  4. Hand it out. A template, an agent in a shared folder, a short how-to. A champion is someone after whom everyone around can do it.
  5. Get a mandate. With numbers and a live case, go to your manager: "I automated X, it saves Y. Give me N hours a week to do this systematically".
  6. Teach others. Short workshops, answers to questions, a shared channel with tips.

Do not: be an evangelist without results, automate everything at once, hide how it works, or wait for permission instead of results.


How to tell whether a use case pays off

  1. Time without AI. Hours per month spent by hand.
  2. Time with AI. Hours now. The difference is your saving.
  3. Cost of AI. Subscriptions plus token usage for this case.
  4. Result: saving minus cost. Positive - scale it. Near zero or negative - say honestly that AI is not needed here.

Example for your manager: "It was 8 hours a week, now 1. Saving about 28 hours a month. AI costs about 30 euros a month. I suggest rolling it out to the team".


Your 90-day plan

Write three lines and keep them:

  1. In 30 days I ______ (what works for me every day).
  2. In 60 days I ______ (what my colleagues use).
  3. In 90 days I ______ (the value and number I can show my manager).
AI Sandbox Lab

Want to build your own system with support

In the 6-session program you build a project, your own agents and a plan for adoption at work. Not sure the format fits? Book a free 15-minute call on Telegram.

After the program you can book a personal consultation of up to 1 hour: we go through your role, how to introduce AI at work, and build a basic plan.

See the program

← All materials

Ask a question