Building with AI: speed, clarity, and the limits of craft

Being on sabbatical has given me something I rarely had during intense CEO years: time to explore. Not by reading about trends, but by actually trying tools until I understand what they change in real work.

Recently I decided to test a practical question: can I ship something real with AI, without being a developer?

So I built my own homepage.

The mistaken explanation: "AI makes building easy now"

I used ChatGPT to generate the initial structure and copy blocks. I worked in VS Code to iterate. I deployed via Netlify. And I used AI as my "always-on pair". Not just for generating code, but for debugging, improving layout, and doing quick reviews.

I know the code isn't perfect. A senior engineer or designer could improve the architecture, simplify the CSS, refine the UI details. But for a simple personal site, that wasn't the point.

The point was to get live.

And I did.

It took roughly four days of focused work and a couple of subscriptions. In exchange, I got a functioning website I can keep improving over time. And, importantly, I removed the biggest blocker: waiting.

A common story people tell about this is that AI removes the need for skill.

In practice, what I experienced was different. AI removed some friction. It did not remove the need for clarity. And it did not remove the value of experience.

I'm not a developer, but I'm not starting from zero

I don't have a formal coding background. But I'm not new to technical thinking either.

Years ago, in neuroscience research, I had to write scripts to analyse EEG/TMS data. I was studying consciousness, and you can't do that work without getting your hands dirty in tooling and data. It wasn't "software engineering", but it taught me the logic.

Later, in banking and fintech, I worked close to product and engineering. I led teams building digital processes and core systems. At EstateGuru I was hands-on with product and platform decisions. As a CEO, I often ended up in a practical QA role: testing flows, spotting edge cases, pushing for better UX, asking "why is this step needed?" and "how do we automate this?"

So while I don't write production code daily, I've learned the most important part: how to talk the language and how to judge whether something works.

This matters, because AI does not replace thinking. It amplifies it. If you can think structurally, you get more leverage out of the tools.

What's actually going on: AI increases the value of direction

The biggest surprise wasn't that AI can code.

The surprise was that AI quickly becomes limited by the human giving instructions.

When I had a clear plan, the tools moved fast. When I wasn't clear—when I changed my mind, or didn't know what I wanted—everything slowed down.

This pattern is familiar well beyond software. In companies, execution rarely fails because people don't try. It fails because priorities shift, ownership blurs, and decisions don't have a clear home.

With AI, the same thing happens at the micro level.

AI doesn't remove the need for thinking. It raises the value of thinking. Especially planning. Priorities. Constraints. Taste. Decision-making.

If you don't know what you want, you can generate a lot of output and still not ship.

The mechanics: why it starts to feel like a team

The other realization was how naturally work splits into roles.

One "agent" can suggest design and layout options. Another can refine copy. Another can research examples. Another can implement changes in code. Another can act like a reviewer.

It starts to feel like having a small team—except the team is always available, and the limiting factor becomes your ability to direct it.

This doesn't remove accountability. If anything, it increases it. You still own what goes live. You still choose what matters. You still decide what "good" means.

But it changes speed. It changes iteration. And it makes shipping much more accessible.

In a way, this is why "shipping beats perfect" is not just a motivational line. It is a practical operating principle. Once something exists in the real world, feedback becomes concrete and improvement becomes easier.

What actually works: ship a version one, then earn quality

This is where the tradeoff becomes clearer.

I know the site can be better. I can already see the parts that are held together by persistence rather than elegance. And after a week of testing, it's obvious that a strong senior developer and a strong designer would move faster than I can.

Not because I'm incapable of learning.

Because experience is leverage.

They have pattern recognition. They know what to ignore. They know where things break later. They can keep quality high without slowing down.

AI helps a non-developer ship version one. It does not remove the limits of craft.

So the right goal isn't "replace professionals".

It's "remove the waiting and move the project forward".

Where out-of-the-box wins: when the surface area grows

This experiment also changed how I think about tooling choices.

For a basic homepage, code plus AI was surprisingly workable.

But the moment I wanted an admin page to manage a blog, things got harder. Not because it's impossible, but because the surface area expands quickly: content management, editor experience, permissions, security, maintenance, integrations.

That's where an out-of-the-box solution starts to make sense.

I might still end up using something like Wix or a similar platform for the basics, and use AI inside that environment. The reason is not "lack of ambition". It's that these platforms bundle solutions that remove accidental complexity.

For someone like me, that tradeoff can be rational.

And for professional web engineers, it can also be rational when the goal is speed, stability, and built-in capability rather than full control.

When features get more complicated, the "build from scratch" path often becomes slower unless you are doing it professionally.

Practical lens: what an operator takes from this

I didn't come out of this thinking "everyone should start coding".

I came out of it with a clearer view of leverage.

If you can define outcomes, priorities, and constraints, AI becomes a real accelerant. If you cannot, AI will produce output, but it won't produce progress.

And when complexity rises, you still need good judgment about tools, tradeoffs, and when to bring in experts.

Closing

If you've been hesitating to build something because you "don't have the background", I'd encourage you to test it. Pick a small, real outcome. Something you actually need. Then ship a version one.

You'll learn more in three days of doing than in three months of reading.

In the end, progress belongs to people who move.

Next Steps

Book an execution mapping session — We'll explore how AI is changing your operations and where you can ship faster.

Related reading: 3+: The Meaning (and Why It Works) — The operating framework for managing execution at scale. When clarity is your bottleneck, 3+ shows you how to limit WIP and deliver faster.

Also: Productive laziness and the new operating system for work — The bigger picture on how AI is reshaping operations, and why clarity and governance matter more than ever.