AI Tools Are Pandora's Box: A Web Developer's Honest Take

The Box Is Open

Here's a thing about Pandora's box that most people forget: after all the chaos escaped into the world, one thing stayed behind. Hope. It was right there at the bottom, waiting.

I think about that a lot these days. I use AI tools every single day. I build websites with them, I coach clients on them, I troubleshoot the messes they make. And I'm going to be honest with you in a way that most content about AI tools avoids: I love these tools AND I hate them.

That "and" is doing a lot of work in that sentence. Because the reality of AI in web development isn't a hot take or a horror story. It's complicated. AI is a powerful tool for uncovering research and automating the stuff that burns your brainpower up. It doesn't replace the experts. It still takes someone who knows what to ask for and, most importantly, when the result is definitely wrong.

This isn't an AI tools review dressed up as thought leadership. It's an honest reckoning from someone who lives in the middle of it: the AI website builder limitations, the genuine breakthroughs, the environmental cost, and the three things I wish every business owner understood before they dove in.

I Love These Tools AND I Hate Them

Let me lay out the tension, because I think it matters.

On the good days, AI saves me hours. It handles the cognitive grunt work so I can focus on strategy, design judgment, understanding what a client really needs versus what they're asking for. For small businesses figuring out whether they should use AI for their website or content, the honest answer is: parts of it, absolutely.

On the bad days, I'm debugging code that an AI wrote confidently and incorrectly, or explaining to a client why the "finished website" their AI builder produced doesn't actually work. Eighty-four percent of developers use AI daily now, but only 33% trust it. That gap tells you everything.

And here's the thing that gets lost in the hype cycle: human-generated content still receives 5.44x more traffic than AI-generated content. The AI website pros and cons aren't theoretical. They show up in your analytics.

So I hold both. The tools are genuinely powerful AND genuinely dangerous when nobody's watching the output. Both are true at the same time.

The AI App Builder That Built Three Login Pages

Let me tell you what happened when I tested an AI app builder.

I gave it a straightforward project: an application with three user types. What I got back was three separate login pages. Not one unified login system that handled different roles. Three completely independent login pages. Nothing really worked. The interface was there and things existed. Everything looked like an app. None of it functioned like one.

As a developer, I could diagnose what went wrong. But here's where it gets dangerous: a small business owner or nonprofit admin without a development background might not catch any of that. They'd see a login page (three of them, actually) and assume the app was working.

And it gets worse than just disorganized code. Forty-five percent of AI-generated code introduces security vulnerabilities. That's not my opinion. That's Veracode's finding after analyzing over 100 large language models across 80 coding tasks. In July 2025, a critical authentication bypass vulnerability was discovered in Base44, a Wix-owned "vibe coding" platform, meaning unauthenticated attackers could access private applications on the shared infrastructure. Real users. Real data. Real risk.

AI code needs to be babysat. Even with days of planning upfront, you still usually need to babysit the output. The productivity gain is real but smaller than the marketing suggests, and the failures are often invisible without expertise.

The One Technique That Actually Makes AI Useful

Okay, so if I'm going to tell you what's broken, I owe you something that works. This is the single most useful piece of AI advice I give to every client I coach.

Instead of telling AI what you want and then yelling at it as it continues to mess up, tell it roughly what you want and then ask it to interview you one question at a time to refine your goals.

That's it. That's the technique. And it changes everything about how AI content for business actually works.

Why it works

The typical AI workflow is "generate, then correct." You type a prompt, AI produces something generic, you try to fix it, it gets worse, you start over. It's frustrating and it produces output that sounds like everyone else's. The interview technique inverts that. Instead of AI generating and you correcting, AI extracts YOUR knowledge and voice through targeted questions. Your expertise stays central. AI becomes a facilitator, not a creator.

I've coached clients through this for content strategy, grant proposals, fundraising letters, website planning, basically any situation where you have expertise but you're staring at a blank page. It works because AI is genuinely good at asking follow-up questions. It helps you think about things you wouldn't have considered otherwise, and it's way easier than trying to brainstorm everything upfront.

This isn't just me, by the way. Professional ghostwriters at the 2025 Gathering of the Ghosts independently arrived at the same conclusion: the one thing they don't use AI to do is write directly. Content strategists at Edify Content and purpose-built tools like Leaps AI are building entire methodologies around this approach. It produces exactly the kind of experience-led, authentic content that Google's E-E-A-T framework rewards.

It's an approach worth trying the next time you sit down to work with AI, whether you're a small business owner or running a nonprofit.

AI Design: Follows the Rules but Misses the Soul

I've looked at a lot of AI-generated designs at this point. They're decent. They're usable. They follow the rules.

I've yet to see one I'd call inspired.

A machine spitting out remixes of what it knows of design won't produce anything brave, daring, or risky. It can't. It's working from the average of everything it's seen, and great design, by definition, is not average. AI follows the established principles perfectly but has no idea why those principles exist in the first place. And when you don't understand the why, you can't make the creative judgment call to break the rules when it matters.

I'm not alone in seeing this. Eighty percent of designers say AI undermines originality, according to a Dezeen survey. Canva's 2026 design trend report is literally titled "Imperfect by Design," a deliberate pushback against years of AI-polished, algorithmic sameness. Even the design platforms that integrate AI most deeply are acknowledging that something gets lost.

Where that leaves you

Here's the practical framework: if the soul of your organization doesn't need to be in the design (a quick internal document, a placeholder layout, routine collateral), AI is fine. If you want something that represents who you actually are, that captures your organization in ways that make people feel something, that pushes edges, you need a professional designer who understands your vision outside the framework of the average of all designs.

That's not anti-AI snobbery. It's just where the AI web design limitations are today.

The Environmental Cost We're All Quietly Ignoring

We've almost all had our fun doing dumb little things with AI: the silly images, the joke conversations, the "make me a poem about my cat" prompts. I've done it too. But at this point we know that's burning a lot of energy. We also know that if we refuse to use AI, many of us will fall behind.

That's a real tension, and pretending it doesn't exist isn't helping anyone.

Here's the number that changed how I think about it: image generation uses roughly 58x more energy than text generation. Every time you churn through AI image revisions ("make it bluer, no more blue, actually go back"), you're burning significantly more resources than text-based tasks. AI systems are projected to generate CO2 equivalent to the entire city of New York, annually.

Efficiency is improving. Google reports a 33x reduction in energy per prompt compared to 2024. But aggregate impact grows with adoption, so individual responsibility still matters.

My compromise

Use it carefully. Be specific in your first prompt so you don't need five follow-ups. Learn to preserve tokens. It saves energy and it saves you money. Don't ask AI to do things a quick search handles. Choose the right model for the task. And the interview technique I mentioned earlier? It's token-efficient by nature. Better results in fewer iterations because the AI is working with your actual expertise instead of guessing.

I'm not telling you to stop using AI. I'm saying be thoughtful about when and how.

What This Means for Web Development (and What It Means for You)

Eventually it will be more about knowing best practices than knowing how to write any particular coding language. That's where this field is headed. AI is changing what developers do, not eliminating the need for them. The question isn't "AI vs. web developer." That's a false binary. It's about where AI handles the routine and where you need a human who knows what they're looking at.

Because AI doesn't automatically make things secure. That 45% vulnerability rate is real. And AI doesn't automatically make things accessible. Automated tools catch only about 30% of accessibility issues. For nonprofits and public entities facing the ADA Title II deadline, these aren't abstract concerns. They're legal and ethical obligations.

Junior developer hiring has dropped 25-70% across organizations. The field IS changing. But judgment and design thinking are now more important than sheer coding speed. The developers and designers who thrive will be the ones who understand security, accessibility, strategy, and human communication, not just syntax.

What that means for you, the nonprofit administrator or small business owner reading this: you don't need to become an AI expert. You need to know what AI is good at, what it's not good at, and when to bring in someone who can tell the difference.

Where This Leaves You

The box is open. AI is out in the world, changing everything, and there's no putting it back. But remember, hope was in there too.

Here's what I want you to walk away with:

1. Try the interview technique. Tell AI roughly what you want, then ask it to interview you one question at a time. You'll get better results in fewer attempts, and the output will actually sound like you.
2. Be mindful of resource usage. Image generation uses 58x more energy than text. Small choices add up. Be specific, be intentional, and don't use AI for things a quick search handles.
3. Know what AI can't do. Security, accessibility, genuine creative vision, and knowing when the output is wrong: these all require a human who knows what they're looking at. If you can't evaluate the output, the output could be hurting you.

I'll keep using these tools every day. I'll keep coaching clients on how to use them well. And I'll keep being honest about what they can and can't do, because I think that honesty is more valuable than hype right now.

Speaking of things AI doesn't automatically handle, we just published a four-part series on website accessibility that's worth your time, especially with the ADA Title II deadline now extended to 2027 and 2028.