Everyone's talking about AI replacing developers. I've been using AI coding tools daily for over a year and my actual experience is more nuanced — and more interesting — than that narrative.
I'm a full-stack developer with 9+ years of experience, currently working at Conga and taking freelance projects on the side. AI tools have genuinely changed my workflow. But not in the way the hype suggests.
What I actually use
GitHub Copilot is my daily driver, baked into VS Code. It handles the repetitive scaffolding I used to type on autopilot — boilerplate DTOs, test stubs, repetitive utility functions. For this kind of work, it's genuinely useful: fast, contextually aware, and correct most of the time.
Cursor I use for bigger refactors and when I want to ask questions about an unfamiliar codebase. The ability to highlight a chunk of code and ask "what's the intent here?" or "what edge cases am I missing?" is surprisingly useful when joining a legacy project.
LLM APIs directly (mostly OpenAI) I use when I'm building AI features for clients — integration work, not research.
Where AI genuinely helps me
Boilerplate and scaffolding. Generating a new NestJS module with controllers, services, and DTOs? Copilot handles 80% of it. I review, adjust, and ship. Saved time: real.
First-pass unit tests. I write the function, ask Copilot to generate tests, then add edge cases it missed. Faster than writing from scratch, and it forces me to read the tests critically rather than produce them mechanically.
Documentation drafts. I write code comments and API docs the way I think about the code. Copilot often fills in the prose faster than I would. I edit, but the starting point is useful.
Exploring unfamiliar libraries. Instead of reading docs linearly, I ask Cursor "how would I implement X using this library?" and get a working example I can interrogate. Faster than guessing from examples.
Where I don't delegate to AI
Architecture decisions. Which service owns this domain? Is this the right abstraction? Should this be async? These require understanding the system, the team, the client's roadmap. No AI has that context.
Debugging production issues. AI is bad at "why is this failing in prod but not locally?" because that requires reasoning about infrastructure, state, timing, and environment — things that live outside the file you're editing.
Security-sensitive code. Auth flows, permission checks, data validation at boundaries — I write these myself and review them carefully. AI-generated security code is often subtly wrong in ways that don't show up in tests.
Anything touching client business logic. When the correct behaviour comes from a domain conversation with a client, not from a pattern in training data, I write it myself.
The judgment gap
Here's what I tell clients when they ask about AI and my rates: AI is a force multiplier for output, not for judgment. The more senior you are, the more value you can extract from AI tools — because you know which suggestions to accept, which to reject, and which tell you that you're solving the wrong problem.
Junior developers using Copilot often accept plausible-looking code that has subtle problems. Experienced developers use it to move faster through the parts of the work that don't require judgment, so they can spend more time on the parts that do.
That gap — the judgment gap — is where the real engineering value lives. And it's not going away.
Practical upshot for freelance clients
Using AI tools means I can scope projects more tightly, move faster through scaffolding phases, and spend more of the engagement on the hard problems. For clients on fixed budgets, that's a better return. For clients who want senior-level thinking on architecture and design, those tools don't change what I bring to that conversation.
If you're evaluating a developer and they claim not to use any AI tools in 2026, I'd be curious why. The honest answer is: everyone uses them. The question is whether they use them with judgment or without.