This isn’t how it happened.

I want to be able to speak openly about this because the discussion around AI coding tools has become extremely polarized at this point. On one side is a group claiming that developers don’t matter anymore, that anyone can create anything, which is nonsense. The other side claims that none of these applications amount to anything, that the whole ecosystem is a sham.

The Surface Looks Easy. The Middle is Where It Gets Real.

Open Lovable or Base44 for the first time and the pitch is intoxicating. Describe what you want, watch a working UI appear, and click deploy. For the first hour, you feel like a magician. For the second hour, you feel like a founder. By hour ten, you feel like someone who has just discovered why developers get paid what they get paid.

The break point is always the same. You ask for a custom feature that touches data, auth, or an integration the AI doesn't have a clean pattern for, and the whole house starts wobbling. You ask the AI to fix one thing, and it ends up rewriting three others. You revert. You re-prompt. You revert again. Suddenly, your "I'll ship this in a weekend" project is a Tuesday-night ritual of trying to figure out why the dropdown menu broke when you asked it to change a button color.

This is the part that does not show up in the demos.

You Need To Know What You Are Telling The Tools

Here is the thing nobody selling these platforms wants to say out loud: the more you actually understand code, the better the tools work for you.

When the AI generates a broken component, a person who knows what state management looks like can read the output and say "you forgot the dependency array on this hook." A person who has never written a line of code stares at the same output and just asks the AI to "fix it" again, which sometimes works and sometimes spirals.

I am not a senior engineer. I am a marketer who codes in an intermediate fashion, and that was good enough for keeping AIOverview running while the AI was struggling. If I were less skilled, then I would have abandoned the whole thing by the second week. If I were more skilled, then the product would already be twice as good as it is now.

It is a real difference and I am not going to deny that. An engineer who truly excels at his craft, if given the same concept and the same technology, would have created something that works far better. More efficient design. Superior performance. Advanced features. Fewer corner cases that I am only beginning to uncover with actual users. I recognize this difference. To conquer it requires owning it first.

Debugging is The Actual Job

The skill these tools demand, and that almost nobody talks about, is course correction. You are not really "coding." You are conducting. You are reading output, spotting drift, redirecting, undoing, re-scoping, and constantly translating between what you want and what the model thinks you want.

If you are the kind of person who gives up the first time something breaks, these platforms will eat you. If you are the kind of person who treats every bug as a small puzzle worth solving, you are going to learn an incredible amount, almost by accident. You end up indirectly learning to code. Not because anyone is teaching you, but because every time you debug something, you absorb a little more of how the underlying system actually works.

That is the sleeper benefit of all this. Six months of building with AI tools is a more practical coding education than a lot of bootcamp curricula, as long as you stay engaged with the why and not just the what.

Where These Tools Genuinely Shine

I'm being critical, but I do think this category is going to keep improving and matter more, not less. The honest read on Lovable, Base44, and the rest of the wave:

They are ideal for starting from zero to one. For quickly validating a hypothesis. For individual efforts, internal tools, prototyping, minimum viable products that do not require scaling to millions of customers right away. They cater to a legitimate market of operators, marketers, independent entrepreneurs, and folks within corporations with an idea, but without any engineering manpower to implement it.

This isn’t negligible. This represents a substantive change. I simply don’t believe it supplants the engineer behind the keyboard with a comprehensive vision of what needs to be created. The constraint has shifted. It was previously “can you write the code?” It is now “do you understand which code matters?”

SEO and Marketing are The Parts Nobody Warns You About

Here is something I did not expect to spend so much time on: getting an AI-built site to actually behave properly in search and AI visibility.

The platforms will happily generate you a site. They will not, by default, generate you a site that has clean schema markup, a proper llms.txt file, sensible internal linking, semantic HTML structure, decent Core Web Vitals, or any of the dozens of small technical things that determine whether your site shows up in Google's AI Overviews, ChatGPT search, Perplexity, or Gemini.

If you are building a tool about AI brand visibility, and your own site is invisible to AI search, you have already lost the argument before anyone gets to use the product.

This is where my background at TBR genuinely made the difference. Working at TheBestReputation, I spend my days thinking about how brands surface across search engines, AI answer engines, third-party platforms, and citation graphs. I knew what schema to ask for. I knew what an llms.txt should look like. I knew that the AI was not going to volunteer any of this unless I prompted it specifically.

If you are coming into these tools without a marketing and SEO foundation, you are going to ship a product that nobody finds. That is a worse outcome than not shipping at all.

AIOverview exists because of TBR

I want to be direct about this. aioverview.com is a free GEO and AI brand visibility tool that I built and that TBR supports. It is genuinely useful. I use it. Our clients use it. People in the industry have started referencing it.

But it exists because I work at TheBestReputation, not in spite of it. Years of doing reputation management, SEO, brand placement, and AI search optimization for clients gave me the domain knowledge to even know what a useful AI visibility tool should look like. The AI coding platforms gave me the leverage to ship it without waiting six months for a dev cycle. Both halves of that mattered.

If I had only had the coding tools and no domain expertise, I would have built something generic. If I had only had the domain expertise and no coding tools, I would still be writing a spec doc for an engineer who has not been hired yet.

How I'm Going To Keep Improving It

Honest answer: not by prompting Lovable or Base44 in tighter and tighter circles.

I'm going to improve AIOverview by paying attention to what the rest of the industry is doing. What other tools are being built. What features are gaining traction. What AI search behaviors are changing month over month. What new platforms (Perplexity, Gemini, AI Overviews, ChatGPT search) are surfacing and how. 

Then I bring those insights back to the build. The AI tools are the execution layer. The strategy layer still comes from being a person in the industry, watching the industry, talking to people in the industry. You cannot prompt your way to relevance. You have to earn it by paying attention.

Final Advice: If You Are About To Try This

Have fun. Be creative. Do not be scared of bugs. Bugs are not a verdict on your ability. They are a normal part of building anything that is worth building. You will fix them. You will get there. The act of fixing is the part that makes you better.

Do not view AI coding assistants as an easy way out. Use them as a backup plan in case you cannot ship your product because you are unable to hire a development team, do not have enough money to pay for an existing one, or want to test your idea without spending too much. They are valuable when used within their scope. They are lethal when forced to go beyond it.

AIOverview is far from flawless. It does contain some bugs. It lacks some features that I wish it had. It also needs some parts of its source code to be rewritten. So what? I did not build it for perfection. I built it to be useful, and then keep improving it until perfection came.

That is the part the AI cannot do for you. Caring whether it gets better. Showing up the next day. Doing the research. Knowing your industry. Knowing what you are actually trying to build, and why.

The tools are getting better. The people behind the tools still matter a lot more.