I’ve been a working web developer for 28 years. I’m skeptical by nature. When something gets hyped, I wait. I watched “the metaverse” get a trillion dollars of investor attention and then quietly dissolve. I watched NFTs. I watched blockchain-everything.
AI tools are different. I know this because I’ve been using them daily for over a year now, and I haven’t stopped. That’s the real test for me — not “is this impressive in a demo” but “do I still reach for this six months later.”
So here’s the actual list of AI tools I use in my web development business, what I use them for, and — just as importantly — what I’ve tried and stopped using.
Cursor — My Daily Code Editor
I switched from VS Code to Cursor about ten months ago and I haven’t looked back. Cursor is VS Code under the hood (same extension ecosystem, same keybindings) with AI baked in at a deeper level than any plugin achieves.
What I actually use it for:
- Inline edits. Select a block of code, hit a keyboard shortcut, describe what you want changed. It rewrites just that block, in context. This is faster than any workflow I’ve had before.
- Explaining unfamiliar code. When I inherit a codebase or pick up a project after six months away, I use Cursor’s chat to ask “what does this function do and why does it exist?” faster than reading through it myself.
- Generating repetitive code. Any CRUD operation, any standard API endpoint, any boilerplate — I describe it, review it, done. The boring 30% of every project goes much faster.
What I don’t use it for: architectural decisions. When I’m deciding how to structure a system, I think on paper first. AI will happily suggest an architecture — but it’s optimizing for something that sounds reasonable, not for your specific constraints.
Claude — The One I Think With
Claude (from Anthropic) is my go-to when I need to think through something, not just generate something. The quality of reasoning is noticeably different from other models I’ve used — it’s more likely to push back, point out trade-offs, and say “I’m not sure” when it isn’t sure.
I use Claude for:
- Architecture discussions. “I’m building a multi-tenant SaaS with these constraints, here are two approaches I’m considering — what am I missing?” It’s like rubber duck debugging, but the duck talks back.
- Writing first drafts. Blog posts (including this one, to be honest — I outline, it drafts, I rewrite heavily), client proposals, technical documentation. It’s fast to get to a first draft. The rewriting is still on me.
- Code review. Paste a function, ask “what’s wrong with this?” It catches things I miss, especially in code I’ve been staring at for hours.
Compared to ChatGPT: Claude tends to be more careful, more honest about uncertainty, and better at long-form reasoning. ChatGPT is faster to respond and often better at creative tasks. I use both, but Claude is my default.
GitHub Copilot — Still Useful, Not Essential
I had GitHub Copilot before I switched to Cursor, and it was genuinely useful. Now that Cursor has similar autocomplete built in with a better model, Copilot is redundant for me. I still have the subscription because it’s integrated into my workflow, but I’d say it’s maybe 20% of what I used to lean on it for.
If you’re not ready to switch to a new editor, Copilot in VS Code is a solid entry point to AI-assisted coding. It’s good at completing repetitive patterns and filling in boilerplate. It’s not good at understanding your whole project or reasoning through complex problems.
Perplexity — For Research I Actually Trust
When I need to look something up and I want citations instead of confident hallucinations, I use Perplexity. It’s a search engine that runs AI synthesis over real web results and shows you the sources. For things like “what’s the current best practice for rate limiting in Laravel 11” or “what WordPress hosting providers support Cloudflare cache rules” — it’s faster than reading five Stack Overflow threads.
I don’t use it for anything where being wrong has real consequences. For those, I read primary sources. But for orientation research and quick lookups, it’s saved me probably 30 minutes a day.
What I’ve Tried and Stopped Using
Midjourney / DALL-E for client work: I tried using AI image generation in client projects. The images are impressive until a client actually looks at them — then you see the weird hands, the uncanny details, the complete inability to match their brand. Stock photos with licensing are still more reliable for most commercial use.
AI-generated content that goes straight to publish: Every time I’ve tried to fully automate blog content or product descriptions, the result is technically correct and completely hollow. It reads like AI wrote it — because AI wrote it. Content that represents your business needs your voice in it. AI gets you a draft; you get the byline.
Too many AI tools: I went through a phase of trying every new AI tool that launched. It was a productivity killer. Now I have my toolkit and I stick to it unless something has a genuinely compelling use case that my current tools don’t cover.
The Honest ROI
I’m a solo developer. My time is the constraint on everything I can deliver. AI tools have given me somewhere between 20-40% more effective output per hour, depending on the task. For the parts of the job that are repetitive or research-heavy, it’s even higher. For the parts that require real craft — designing a user experience, making architectural decisions, understanding a client’s business — AI doesn’t move the needle much.
The tools that earn their place in my daily workflow are the ones that give me back time without costing me quality. That’s a shorter list than the internet would have you believe. But it’s a real list, and it’s growing.
If you run a small business and want to know which AI tools might actually help you (or which ones are marketing noise), reach out. That conversation is free and I won’t try to sell you something you don’t need.