Copilot became generally available in June this year. I turned it on the same week and I never turned it off. Six months is enough time to have a real opinion, not a hot take. So here it is, the good, the bad, and the part nobody wants to talk about, which is money and risk.
The good part is boring, and that is the point
Copilot is amazing at boring code. Boilerplate, mappers, DTOs, test setup, that one function that converts a date format you wrote a hundred times in your life. It just types it for you. You press tab and you move on with your life .
Tests are where it shines the most for me. I write the first test in a file, with a good name and a clear arrange, act, assert structure. Then Copilot suggests the next test cases and it is often right. Edge cases I would maybe skip because I was tired, it offers them for free. My test coverage did not go up because of discipline. It went up because the cost of writing one more test dropped close to zero.
Let me try to put a number on it. On a normal day Copilot saves me maybe twenty to thirty minutes of typing and small lookups. That sounds small. But twenty five minutes a day is around two hours a week, almost a full work week per year, per developer. The subscription costs ten dollars a month. If your developer costs the company fifty dollars an hour or more, the math is not even close. As a pure typing assistant, it pays for itself in the first morning of the month.
The dangerous part
Now the other side. Copilot is confident in a way that humans are not. When a junior developer is not sure, they write hesitant code, they ask questions, they leave a comment. Copilot writes wrong code with the same smooth style as right code. It will happily suggest a function that handles nine cases out of ten and looks complete.
I got burned once. It suggested a date range check that looked perfect. It compiled, the happy path tests passed, the code review passed too because it looked like code I would write. The bug was in how it handled the boundary day. We caught it before customers did, but only because someone was testing something else nearby. That is luck, not process.
Here is the business math from the other direction. One subtle bug that reaches production can cost a day of investigation, a hotfix, a postmortem, and some trust from the customer. Call it twenty hours of team time if you are lucky. That is months of “minutes saved” gone in one incident. The tool is positive on average, but the variance is real and you have to manage it.
So my rule is simple. Copilot writes, I review. Every suggestion gets the same eyes I would give to a pull request from a new team member. Fast eyes for boilerplate, slow eyes for anything with logic, money, dates, or permissions. Kent Beck has been writing about how tools change which work is cheap and which is expensive, and I think that is the right frame. Copilot made typing cheap. It did not make thinking cheap.
Who should use it
If you are a senior developer, use it today. You have the judgment to filter the bad suggestions and you will collect the time savings with low risk.
If you are a junior, I am more careful. The tool will let you produce code you do not understand yet. That feels like progress but it is debt. My advice is to use it, but force yourself to explain every accepted suggestion out loud. If you cannot explain it, delete it and write it yourself.
For teams, treat it like any other dependency. Decide as a group how you review AI suggested code, especially in the risky areas. Do not let it be a silent personal choice, because the bugs it creates are team bugs.
One last thing. Two weeks ago OpenAI launched ChatGPT and everyone is talking about it, my Twitter feed is nothing else right now. I am still forming an opinion and I will write about it soon. But it makes one thing clear. Copilot was not a gadget, it was the first product of a wave. Better to learn how to work with these tools now, with judgment, than to pretend the wave is not coming.
Six months in, my verdict is simple. Copilot is a great typist and a terrible engineer. Pay for the typist. Keep doing the engineering yourself.
Pax et bonum.