In early 2024, I decided to rebuild my corporate website. Wix was slow, painful to manage, and useless for SEO. I wanted control. I also wanted speed.
AI seemed perfect for this.
I used Cline with OpenRouter. Picked Next.js. Shared content. Gave guidelines. Let it run.
At first, it felt like progress. Then I ran PageSpeed.
- SEO score: 54.
- Page load: Poor.
- Sitemap: “Generated,” but Google Search Console showed errors everywhere.
- Structure: Blog posts were landing inside services pages.
New Next.js features like the <Image /> component weren’t even used—the model didn’t know they existed.
This wasn’t a “small cleanup” problem. The foundation was wrong.
What I Thought I Delegated
I thought I was delegating execution. In reality, I had handed over a core decision:
Decide the right website structure for SEO so I don’t have to think about it.
That was the mistake.
The Real Failure Mode
This wasn’t about bad code. It was structural failure.
The AI created a website structure that:
- Didn’t match how my business actually works.
- Didn’t reflect how I want customers to understand me.
- Optimised for “generic good SEO,” not my positioning.
It made reasonable choices. They were just wrong for me.
The Line I Crossed
Here’s the clean admission:
I delegated key website structure decisions without freezing my judgment on how the business should be represented.
Once that happens, you’ve already lost control. AI will always optimise for the middle of the bell curve. Operators rarely live there. If your business sits even two steps away from average, AI won’t know that unless you force it to.
The Operator Realisation
AI doesn’t fail at execution. It fails when you ask it to infer intent, boundaries, and non-negotiables.
Those don’t emerge from prompts. They must be frozen before execution starts.
Once I saw that, I stopped “prompting” and started designing control.
What Actually Fixed It
I scrapped the build and restarted. This time, I did three boring things:
1. Wrote a PRD (for AI, not humans) I defined the “Why” and “What” before writing a line of code.
2. Locked the Constraints Pages, hierarchy, SEO intent, and non-negotiables were decided by me, not the model.
3. The “One Feature” Loop I stopped asking for the whole site. I built one small feature per session. Every session ended with a summary, which I fed back into the PRD.
The PRD became Context, Memory, and Project Manager. AI stopped deciding what mattered. It only executed what was already decided.
Time Reality (No Romance)
The “AI will just do it” phase: 3 weeks lost. A broken site. The controlled system: 1 week to design. 5 weeks to execute.
Longer on paper. Faster in reality. Because I stopped fixing surprises.
The Boundary
Here’s the rule I operate by now:
If you’re using AI to build anything important and you haven’t documented your why, how, and what, stop. You’re not delegating. You’re gambling.
Strike Now
Open one AI-built system in your business today. Ask one question:
What decision have I delegated here without explicitly freezing my judgment?
If you can’t answer that in one sentence, the system is already drifting. That’s not an AI problem. That’s an operator problem.