Using ChatGPT for the First Time as a Developer

Last weekend, OpenAI opened ChatGPT — a conversational AI tool — to the public. I spent a few evenings putting it through its paces. This is not a usage guide; I haven’t spent nearly enough time with it to write one. It’s simply the honest first impressions of a developer after a few days.

First session

My first test was a small one: I asked it to generate a date-range function for a Laravel project. It returned working, readable code in seconds. Frankly, it was better than I expected. In my second test, I pasted in an error message; it correctly diagnosed the problem and explained the fix.

During those first few hours I felt a clear sense of curiosity mixed with a little unease. Having a tool answer what you ask — rather than merely searching for it — is an unusual experience.

Where it works

After a few days, I found the tool added the most value in these situations:

• Reconstructing things I know but can’t recall. Describing a regex pattern or a boilerplate I’ve written hundreds of times before, and getting it back instantly.
• Getting initial bearings in unfamiliar territory. Asking “give me an outline of this” on a topic I don’t know well produces a useful starting map.
• Cleaning up text or comments. Bringing order to a disorganized explanation I’d written hastily.

All three are really variations of the same thing: accelerating reasoning I already possess. Recognizing this clarified both what I can use the tool for and what I can’t.

Where it isn’t reliable

This is the important part. The tool states things it doesn’t know with the same confident tone it uses for things it does. It suggested a library method that doesn’t exist; it described incorrect behavior as correct. And it did so with exactly the same self-assurance it shows when it’s right. You cannot tell from the output alone whether the output is accurate.

At one point I ran a deliberate experiment: I introduced a false premise on a topic I know well, with subtle but consequential details, and waited to be corrected. It wasn’t. It accepted the false premise and built on top of it. This is expected behavior — but seeing it live made an impression.

That led me to a rule I formed on day one: this tool is an assistant, not an authority. I review everything it produces the way I’d review a pull request from a junior developer — I don’t accept a single line without reading it, understanding it, and testing it. In areas I know, it speeds me up; in areas I don’t, it can confidently walk me in the wrong direction.

Where I’m placing it for now

I’ve been doing this for fourteen years, and in that time I’ve seen plenty of tools that were supposed to “change everything.” Some genuinely did; most were forgotten. I’m not in a position to say which group ChatGPT will fall into. But I’ve noticed this: the places where it adds value are exactly the places where I’m already capable of reasoning. The moment I delegate the reasoning to it, my work deteriorates.

An analogy: a good search engine does the same thing — it speeds up access to information, but leaves the job of evaluating that information to you. I think of ChatGPT in the same frame; the difference is that it presents its answers in a far more persuasive form. That persuasiveness is itself a risk.

In the coming weeks I plan to try using it more systematically. For now, the note I’ve written down is this: the tool’s speed is real, but that speed only has value when you can verify the accuracy yourself. As long as I don’t forget that, it looks like it will make a useful desk companion.