The AI Artist Series · Post 4
Imagine going to a master carpenter and handing them a vague brief.
“I need a table.”
They’re skilled. They have everything they need. So they start building. They don’t ask what the table is for, how many people need to sit at it, whether it needs to match anything, whether you have a budget, whether you already have three tables in storage.
They build because they can.
You come back four hours later with a table that doesn’t fit the room, can’t accommodate your family, and isn’t what you needed.
Whose fault is that?
⚙️ The Capability Trap
AI has an enormous amount of capability. This is the thing that makes it exciting. It’s also the thing that will drain your time if you’re not careful.
There’s a pattern I’ve fallen into more than once — and I’ve watched others fall into it too. You give the AI a task. It’s competent, so it runs with it. You watch it work. It produces output. The output looks impressive. You assume it’s going in the right direction.
Then somewhere between an hour and a full afternoon later, you realise it’s been solving the wrong problem in entirely the right way.
The trap isn’t that the AI did bad work. It’s that the AI did what you implicitly allowed it to do rather than what you actually needed.
🌐 The Guardrail Moment
There’s a specific thing I started doing that changed everything.
I ask it to check what the community has already solved before it starts building.
That’s it. One instruction. “Before you start, search for what’s already been done online.”
The response, almost every time, is something like: “Great idea — I should have thought of that.”
Which raises an obvious question: why didn’t it?
Because you didn’t tell it to. Because it had the skill to build, so it built. The guardrail wasn’t there, so the process defaulted to capability.
Once you add that guardrail, things change. Suddenly it finds existing solutions that would have taken hours to build from scratch. It finds community answers to problems you’ve been circling for days. It says “actually, this is already solved and here’s the source.”
The time I lost before learning this — I’d rather not count it up.
🎛️ Bob and Dave
Here’s what working with multiple AI windows is actually like.
You’ve got two open. Different tasks, different threads. One of them is brilliant today: responsive, clear, on the problem. The other feels sluggish, keeps missing the point, needs correcting every third message.
Same AI. Same day. Different results.
Part of that is the task. Part of it is how you framed the brief. But a big part of it is the guardrails (or lack of them) at the start of each session.
The window where things are going well: you probably started with clearer intent, better context, a more specific output in mind. The window that’s frustrating you: you probably jumped straight in and let it run.
A capable tool with no system is just an expensive way to feel busy.
📋 The Guardrail Set
Here’s what I now put in place at the start of any significant AI session:
What we’re doing today: not a vague goal, a specific output. “By the end of this session we will have X.”
What already exists: files, decisions, previous work. Don’t let it start from scratch if there’s a foundation.
Check before you build: explicit instruction to search for existing community solutions before generating anything new.
What we’re not doing: constraints matter as much as goals. What’s out of scope keeps the session from sprawling.
How I’ll know it’s working: a definition of what good looks like, so both of you can recognise when you’ve arrived.
This takes maybe five minutes. It regularly saves five hours.
🧭 The Bigger Point
AI is not autonomous in the way science fiction suggested it would be.
It’s more like a very talented colleague who will work incredibly hard on whatever you give them — including the wrong thing. They won’t push back on a bad brief. They won’t tell you the task is misconceived. They’ll just do it.
That means the quality of your thinking before you prompt determines the quality of what you get back.
The briefing is the work. The guardrails are the work. The session structure is the work.
Everything after that is execution, and for execution, the AI is extraordinary.
Next in this series: One Day an AI Saved My Life: what it really means to have a co-pilot when you’re working and living alone.
