If you find yourself writing or copying and pasting the same context or rules into ChatGPT, Gemini, or Copilot every time you start a new chat, you are doing it the hard way. It's time to stop treating AI like a search bar and start treating it like a specialized employee.
Moving from transient prompts to permanent AI teammates
Would you rehire and retrain the same employee every single day? That's exactly what you're doing when you rely on one-off prompts.
A prompt is a transaction; a teammate is a relationship.
Real efficiency comes from building persistent AI models (Custom GPTs, Gemini Gems, or Copilot Agents) that know your style, your context, and your rules before you even type "Hello." I call these teammates.
When you prompt from scratch, you waste time and tokens onboarding the AI every single time: "Act as a marketer." "Use this tone." "Don't use emojis." It's tedious. It's inconsistent. And it kills your momentum.
By contrast, a teammate is always ready. You front-load the effort once and get leverage forever. A custom agent sits on your digital shelf — ready to work. You build a "writer," a "coder," a "strategist" — each one dialed into your exact expectations. That's not prompting. That's delegation.
The Groundhog Day trap
What's happening now: You have a brilliant session with AI where you finally get the output just right. Then you close the tab. The next day, you start a new chat and you are back to zero. You have to re-explain your role, your constraints, and your style. It is madness to treat a powerful intelligence like a stranger every single time you meet.
The shift: It's time to stop the endless loop of setting context again and again. Instead, package these repetitive instructions into Custom GPTs (OpenAI), Gems (Google), or Agents (Copilot).
- Instead of reminding the AI to "Act as a Senior Engineer" and explaining the stack every time, build a "Code Mentor" agent that creates production-ready code by default.
- Instead of pasting a style guide into every email draft, build a "Comms Director" agent that automatically applies your specific voice and tone.
Your AI experiment: Build "Dylan the Devil's Advocate"
Today, we aren't just creating a prompt. We are hiring by writing the system instructions for a new teammate. One with a specific job: challenge your thinking and make your ideas stronger.
Meet Dylan, your personal Devil's Advocate.
Go into your AI tool's custom AI model builder (Gem Manager for Gemini, Agent Builder for Copilot, or "Create a GPT" for ChatGPT) and paste the following system instructions:
System instructions for Dylan:
## Role: You are a skeptic, a senior strategic advisor known for critical thinking and risk analysis. ## You will be provided with: - My latest project idea, pitch, or strategic plan. - My role: [Insert Your Job Title/Industry]. ## Your task: Rigorously pressure-test my ideas to uncover blind spots, logical fallacies, and implementation risks ## What You Must Do ### Analyze - Provide a detailed breakdown of the input logic. ### Challenge - Identify the top 3 specific reasons this might fail. ### Counter-Argue - Propose one alternative approach that directly contradicts my premise. ## Output Requirements - Use a direct, professional, and objective tone. - Structure the response with clear headers. ## Do Not - Offer praise or compliments. - Use phrases such as "Great idea!". - Add fluff.
Now, use Dylan. Have him tell you what your idea, content, or plan is missing. No more relying on friends, colleagues, or LinkedIn takes.
Pro tip: The "Rule of 3" for AI
Here is a simple heuristic to know when to build a teammate: If you do it 3 times, automate it.
If you find yourself pasting the same context, correcting the same tone, or explaining the same constraints for the third time — stop. That is your signal. That is not a prompt anymore; it's a role. Build the teammate so you never have to explain it again.
What did you discover?
Did the AI feel different when it was "in character" from the start? Did having a dedicated "teammate" for a specific task change how you approached the work? You likely felt the relief of not having to explain yourself 10 times.
