Back to Lab
/
AI PROCESS TEACHING

AI Is a Hammer (And You've Never Seen One Before)

What I learned teaching a nonprofit team how to actually use AI — not as a genie, but as a process.

Last Thursday I spent two hours with the team at a nonprofit doing extraordinary work fighting for human rights in one of the most challenging environments on earth.

They asked me to help them figure out how to use AI.

The Genie Problem

Here is the pattern I see everywhere:

Someone opens ChatGPT. They type “Write me a blog post about X.” They get 1,500 words of polished, soulless nothing. They think: AI is overhyped.

But the problem was never the tool. The problem is that we treat AI like a genie.

“I want a chair.” Okay — leather La-Z-Boy or plastic folding chair? You didn’t say.

If your input is vague, your output is garbage. That is not an AI problem. That is a process problem.

The Hammer Metaphor

Imagine handing someone a hammer for the first time. They have never seen one. They might grab it by the head. They might throw it. They might use the handle to poke something.

That is where most of us are with AI right now.

We are all beginners. The models change weekly. There is no “right way” — but there are better ways. And better ways start with understanding what you are trying to build before you pick up the tool.

Process First, Tool Second

The mistake is jumping to “Which AI should I use?” before answering “What is my actual workflow?”

With the team, we started with their content creation process — not the technology. What does their blog pipeline actually look like?

  1. Research — What is trending? What connects to their mission?
  2. Aggregation — Where are the trusted sources?
  3. Drafting — Who writes? In what voice?
  4. Editing — How do you catch the “AI-ness”?
  5. Analysis — What performed? What didn’t? Why?

Once you map the process, then you assign the tools. Perplexity for research. NotebookLM for grounding your sources. Claude or ChatGPT for drafting — with a specific persona, not a blank prompt. A separate editor agent that strips out the em-dashes and robotic transitions.

Each step is a link in a chain. Each link is simple. The power comes from connecting them.

The Soundboard Principle

I used this analogy with the team: a professional audio mixer looks terrifying. Hundreds of knobs and switches. But every single row is identical — it is just Channel 1, Channel 2, Channel 3. The complexity is an illusion created by repetition.

AI workflows are the same. Break the process into the smallest possible pieces. Get each piece working. Then chain them together.

Do not try to build the soundboard. Build one channel. Then copy it.

Start Manual

The instinct is to automate everything immediately. Build the perfect pipeline. Connect all the APIs.

Don’t.

Go through the process manually first. Use Perplexity for research, copy-paste the output into ChatGPT, iterate with an editor agent, review the result. It is clunky. It is slow. It works.

Because until you have gone through the loop three or four times, you do not actually know what your process is. You think you do. But you will discover that Step 3 should come before Step 2, or that you need a sub-step you never considered.

Build the MVP. Refine it. Then automate.

The Deeper Lesson

Teaching this session reminded me of something I keep coming back to:

The people who win with AI are not the most technical. They are the most process-oriented.

They are the ones who can look at a messy, manual workflow and say: “This is actually seven discrete steps. Let me separate them.”

That is not an AI skill. That is an operations skill. A systems-thinking skill. The kind of skill that has always mattered — it just matters more now, because the tools can finally keep up with the thinking.

Related: