Every few weeks, someone in our FlipWork™ community, a friend, a colleague, or a sprint participant, asks me a version of the same question.
"Monica, am I using AI right? Because it feels like I am barely scratching the surface."
They are not wrong. They are using AI every day. But they are using it like a brilliant new hire they never onboarded. No context. No access to their systems. No understanding of how the work gets done. Just questions in a chat window.
That image stayed with me. Because most of the leaders I work with are doing exactly the same thing. Not because they are not smart. They absolutely are. But because no one has shown them that the AI tool they already use has a whole architecture behind it that most people never touch.
Dharmesh Shah, co-founder and CTO of HubSpot and one of the sharpest voices I follow on AI, wrote about this recently in his simple.ai newsletter. He calls it the LLM harness. And the way he describes it changed how I think about what AI fluency really means.
Here is the core idea. The AI model, whether it is ChatGPT, Claude, CoPilot, or Gemini, is like the CPU inside a computer. It is powerful on its own. But a CPU alone cannot store your files, show you a screen, or connect to the internet. It needs an operating system to unlock all of that. The harness is that operating system. It is everything that wraps around the model and fills in the gaps. And it is not one thing. It is a stack of layers, each one adding more capability. He describes it as a ladder.
My mother used to say, "Dime con quién andas, y te diré quién eres." Tell me who you walk with, and I will tell you who you are.
The same is true for your AI. Tell me what context you have given it, what systems it can reach, and what processes you have taught it, and I will tell you exactly what results you are getting.
Most professionals are asking their AI the right questions. They have just never introduced it to the right people.
THIS WEEK'S INSIGHTS
The model is not what is holding you back. The harness is.
1. Most professionals are stuck on the first rung, and it is costing them more than they realize.
The ladder has seven rungs: prompts, custom instructions, skills, plugins, tools, MCP servers, and APIs. Most professionals, even daily AI users, never climb past the first one. A 2024 McKinsey report on AI productivity found that organizations capturing AI's full value go well beyond surface-level use. They configure their tools, integrate them into workflows, and teach the AI how their work actually gets done. The difference is not which tool they chose. It is how intentionally they set it up.
Think of it like this. You hired a brilliant consultant, one of the best in her field. But on day one, you gave her no office, no login to your systems, no introduction to your team, and no explanation of your priorities or communication style. Then you are surprised when her outputs miss the mark. That is most people's relationship with AI right now. The talent is there, but the infrastructure is not.
2. Custom instructions are the fastest upgrade most professionals have never set up.
Every new conversation you open with an AI tool starts as a blank slate. The model does not automatically know who you are, what you do, or how you think. Custom instructions change that. They are reusable context you write once, and the AI references every time. Your role, your communication style, your priorities. You tell the AI who it is working with, and it performs like it already knows you.
A senior leader who takes 15 minutes to set up her custom instructions gets a completely different AI experience than one who does not. Same tool. Same model. Entirely different output. That 15 minutes is the highest-return setup investment available to any professional right now.
3. Skills and MCP servers are where AI stops answering and starts doing.
Above custom instructions sits something even more powerful. Skills, in Claude's framework, are reusable playbooks. You teach the AI how your team does a specific kind of work once, and it follows that process every time without you having to re-explain. Think of it as writing an SOP, a standard operating procedure, but for your AI instead of your team. Above that is MCP, the Model Context Protocol, a new standard that lets AI connect directly to the tools you already use: your calendar, your email, your files, HubSpot, Slack, Notion, and more.
Dharmesh compares MCP to USB-C. Before USB-C, every device needed its own proprietary cable. Now one standard works across everything. MCP does the same for AI. Build the connection once, and any MCP-compatible AI can use it. This is where AI stops being a tool you talk to and starts being a system that works for you.
3 MYTHS TO REFRAME
These are the beliefs I hear most often from smart, capable professionals who are using AI every day but not getting close to what it can actually do.
Myth #1: "Better prompts are the answer."
Why we believe it: Prompt optimization is everywhere. Everyone is selling prompt templates, cheat sheets, and five-word formulas to unlock better AI output.
Reframe: Prompts matter, but they are the first rung on a seven-rung ladder. What my work with leaders confirms: the bigger shift in AI over the last two years has moved from optimizing how you ask to optimizing what the AI has access to when it thinks. A great prompt handed to an AI with no context about you, your work, or your systems still produces generic output. The prompt is the match. Context is the kindling. Without the second one, the first one does not catch.
What to do: Write one paragraph describing who you are, your role, and how you prefer to communicate. Paste it into your AI's custom instructions this week. That one move will improve your outputs more than any prompt template.
Myth #2: "All AI tools are basically the same."
Why we believe it: On the surface, they look identical. A chat window, a blinking cursor, a text box. They all seem to do the same thing.
Reframe: The surface looks the same. The architecture underneath is different. And what you configure on top makes them perform very differently. Claude Projects, ChatGPT's custom GPTs, and Google Gemini's workspace integrations each offer distinct ways to climb the ladder. But here is the part most people miss: a well-configured version of any major AI will outperform an unconfigured version of the same tool, every single time. The tool you use matters less than how well you have set it up.
Think of it like two cooks given the same kitchen. One walks in and starts grabbing whatever is on the counter. The other spent an hour before she started: laying out every ingredient, organizing her tools, and arranging her station exactly the way she works best. Same kitchen. Same ingredients. Completely different dinner.
What to do: Pick one AI you use daily. Spend 20 minutes this week setting up its memory, custom instructions, or project context. Compare the outputs before and after. The difference will be immediate.
Myth #3: "MCP and APIs are for developers, not leaders."
Why we believe it: The vocabulary sounds technical. MCP servers. APIs. CLIs. That belongs in the IT department, not on a leader's desk.
Reframe: The tools are becoming accessible to anyone who decides to use them. MCP servers now exist for HubSpot, Gmail, Google Calendar, Slack, Notion, and dozens of other platforms that most non-technical professionals already rely on every day. You do not need to write a single line of code to connect them. You need to know they exist and decide to set them up. That decision is a leadership decision, not a technical one. And the leaders making that decision now are building a capability gap that will be very hard to close later.
What to do: Visit the MCP server directory for your AI tool of choice. Identify one platform you use daily. Find the connector. The setup is often one click. This is the rung most professionals do not even know exists.
TOOLS TO EXPLORE
This week, try the following tools and Prompts to Steal to start climbing your ladder.
Claude Projects as your AI home base
Prompts to steal:
"I am going to give you context about my role, my priorities, and how I like my work done. Read everything before you respond to anything. Ask me clarifying questions if you need more detail before you begin."
Power tip: Set up a Claude Project the way you would onboard a new team member. Write out who you are, how you think, what you are working on, and what your standards look like. Upload relevant documents. Load in any templates or processes you use repeatedly. Every conversation inside that Project launches with all of that context already loaded. The setup takes 30 minutes. The payback starts immediately.
ChatGPT Custom Instructions for daily users
Prompts to steal:
"Before we work on anything, tell me what you know about me and how I prefer to work. I want to confirm your context before we begin."
Power tip: Treat your custom instructions like a living document. Update them every time your role, priorities, or communication preferences shift. The AI can only reflect who you are today if you have told it who you are today.
MCP Connectors for your existing tools
Power tip: Start with one connection that matters: your calendar or your email. Once you have one MCP server running, the concept becomes concrete, and the next connection gets much easier. You do not need to connect everything at once. One rung at a time. Anthropic's MCP documentation walks through the setup in plain language.
TRY IT THIS WEEK (Micro Actions)
These are designed to move you from knowing to doing. Share these with the leaders in your circle.
1. Write your AI onboarding document. In one paragraph or less, describe who you are, what you do, and how you like AI to communicate with you. Paste it into your AI's custom instructions or Project settings. This is the single highest-return 10-minute investment you can make in your AI fluency right now.
2. Name the rung you are on. Look at the full ladder: prompts, custom instructions, skills, plugins, tools, MCP, APIs. Be honest about where you operate today. You do not need to be at the top. But you should know where you are and what the next rung looks like from where you are standing.
3. Teach your AI one process you repeat every week. Pick a workflow you do on autopilot: a weekly status update, a meeting recap, a client check-in. Write out the steps. Teach your AI how you do it. Save it as a reusable instruction set. That is a skill. It is yours to use every week from this point forward.
POWER TIP
Most professionals use AI as a chat window. The leaders getting the highest-quality results have done one thing differently. They treated their AI like a team member who needed onboarding, not a search engine that needed better keywords. Write context before you write prompts. Teach the process before you test the output. Set up the harness before you expect the model to perform. That sequence changes everything.
👉 Which rung are you on today? And what is the one thing you will set up this week to climb to the next one?
Closing Thought
The model is not what is limiting you. The harness is.
Every AI tool you are already using has more capability than you are currently accessing. Not because the features are locked away. Because most people never set them up.
You do not have to climb every rung. A lot of real, valuable work happens in the first three or four. But knowing the full ladder is what lets you keep growing once you outgrow the rung you are on. And the professionals who know where the ladder goes are the ones who keep building an advantage.
"Dime con quién andas, y te diré quién eres." Tell me who you walk with, and I will tell you who you are.
Your AI becomes what you surround it with. Give it context. Teach it your process. Connect it to your work. And then watch what it can do.
Leaders who build with intention stay sharp and stay ahead.
♻️ Share this with a leader who is still using AI like a blank chat window.
¡Hasta la próxima, un abrazo fuerte! (Until next week, a big hug!)
Monica Marquez
🔔 Follow me for more on LinkedIn
How did you like today's newsletter?
Your opinion matters!
🔥 Loved it! It was insightful.
Responses