Tag: AI

Get Them Out

~ David Willis ~ Leave a comment

7-40 Challenge | Round 4, Day 32

I went for a walk this afternoon with a small headache and vague tension in my chest — the kind that comes from knowing you have something to get out but not being able to see it clearly yet. By the time I got done, I had a complete framework for the communication course I’ve been circling for months. Ten principles. A product structure. A content engine. None of it existed in any organized form before I started walking.

The ideas were already in my head. They just needed out.

That’s the part most people skip. They sit with ideas swirling, waiting for the moment when it all clicks into place internally before they start. But it doesn’t click inside. It clicks when you get it outside — onto a page, into a voice recording, onto a whiteboard, into a conversation. The act of externalizing is what organizes the thinking, not the other way around.

I’ve written a blog post every day this year. The best ones didn’t come from sitting down with a clear idea. They came from starting with a half-formed thought and watching it take shape as the words came out. The writing did the thinking for me.

I left on the walk this afternoon with tension. I came back with details fleshed out. The only difference was getting it out of my head and into the air.

I Told the AI to Edit My Book

~ David Willis ~ Leave a comment

7-40 Challenge | Round 4, Day 23

Earlier this year, I finished my first novel — 105,000 words of a YA superhero story set in the 1990s. It needed editing. I had Claude. I figured this would be straightforward.

I said, essentially: let’s edit this.

The AI started rewriting my story. Not editing — recreating. It changed plot points. It rearranged material. It put scenes out of order and stopped tracking what had happened in previous chapters. It was hallucinating its way through my manuscript, and the output was getting further from my story with every pass.

So I stopped and changed how I talked to it.

Instead of “edit this,” I said: read this chapter. Read the chapters before it. Tell me what works and what doesn’t. Point out the parts that are heavy, the parts that don’t explain enough, the parts that slow down. Do not make any edits. Just show me the problems.

And it worked.

The AI became a sharp, tireless reader who could point out structural issues I was too close to see. I made the decisions about what to change. I did the rewriting. But I had a partner who could read my 105,000 words without fatigue and tell me where the story was dragging, where a character’s arc was inconsistent, where I was telling the reader something the scene had already shown.

That manuscript lost nearly half its weight through editing. Every cut made it better. And the AI didn’t make a single one of those cuts — I did.

The difference between the first attempt and the second was entirely in how I defined the problem. “Edit this” is not a problem statement. It’s a wish. “Read this and tell me what’s wrong without touching it” is a problem statement with boundaries, criteria, and a clear role for each party.

The AI didn’t get smarter between attempt one and attempt two. I got clearer.

The Drift You Don’t Notice

~ David Willis ~ Leave a comment

7-40 Challenge | Round 4, Day 15

Week one, you push back on everything AI gives you. You check the output. You question the reasoning. You verify the facts. You’re in charge and you know it.

By week ten, the checking feels redundant. The tool has been right so many times that pushing back seems like wasted effort. So you stop. Not all at once — you just skip a verification here, accept a suggestion there. And somewhere between week one and week ten, you’ve abdicated without ever choosing to.

That’s the trap. You don’t abdicate by decision. You abdicate by trust accrual.

I use AI every day — for writing, for data work, for thinking through problems. It is the most powerful tool I’ve ever worked with. And the more powerful it gets, the more dangerous the drift becomes.

Because it gets worse as the tool gets better, not better. A sharper tool makes abdication more tempting. The output looks cleaner. The reasoning sounds tighter. The errors get harder to spot — not because they’re smaller, but because they’re wrapped in fluency that makes you want to believe them.

Here’s what I’ve learned from the chair: AI is a reasoning engine, not a truth source. It doesn’t know anything. It processes what it’s given and returns the most plausible-sounding result. If the truth isn’t in what you’ve supplied or what it’s been trained on, it starts on the wrong foot and builds confidently from there.

My edge is whatever only I can supply — my intent, my standards, my domain knowledge, my ability to say “that’s wrong” when the output sounds right.

The thing nobody tells you is that AI doesn’t erode your ability to reason. It erodes your exercise of it. The muscle is still there. You just stop using it because the tool made it feel unnecessary. And by the time you need it — the day the output is confidently, fluently wrong — the muscle hasn’t been worked in months.

I have one rule that doesn’t bend: if I ship it, it’s mine. Not AI’s fault. Not the tool’s limitation. Mine. I signed off on it. My name is on it.

The signature got cheap. The responsibility didn’t.

Go Deep

~ David Willis ~ Leave a comment

I ran a demo today. Asked AI a question in plain English. It wrote a SQL query in real time. I asked it to convert the output to R. Done. Less than a minute.

Three years ago that could have taken me a few hours. Minimum.

Everyone in the room was impressed, and I don’t blame them. It is impressive. But the part that mattered most isn’t the part that got the reaction.

The SQL it produced was good. It took the natural language prompt I gave it and created what I wanted. However, I still had to verify the SQL to make sure my demo was successful. I was able to do that because I have been doing this kind of work for almost twenty years. I didn’t have to look it up. I just knew.

And that’s the thing more people need to talk about.

AI is going to flatten surface-level knowledge. If all you bring to the table is the ability to do something the machine now does in thirty seconds, that’s a problem. But if you can evaluate whether what the machine produced is actually right — that’s a different conversation entirely.

I told the room: build your context architecture. Know every piece of your workflow. Know how the levers get pulled. Know what right looks like before you ask the machine to produce it. Because without that architecture, AI doesn’t help you. It just runs your bad assumptions faster.

The people who thrive through this won’t be the ones who learned the tool fastest. They’ll be the ones who went deep enough to know when the tool got it wrong.

I am thankful that I have had the last twenty years to learn the data. Today that investment is paying returns I didn’t expect.