The Five Stages of Being Replaced by Your Autocomplete

In 1969, the psychiatrist Elisabeth Kübler-Ross described five stages of grief. They start with denial and end with acceptance. Companies later turned them into the "change curve" to explain how employees react to layoffs and restructurings.

Software developers have been going through five stages of their own since 2021. It started with an autocomplete that helped you write code. Since Opus 5.5 came out, it looks like the AI can do the whole job on its own. That's a problem when the job is yours.

1. Curiosity

In June 2021, GitHub announced Copilot, and you joined the waitlist. When you finally got in, you typed function isPrime( and it wrote the rest. You hadn't needed to check for prime numbers since university, but it was still impressive.

You showed it to a coworker. You both agreed it was a fancy autocomplete that would never understand a codebase with more than one file. Then you went back to Stack Overflow.

2. A crush

After a few weeks you started to like it. If you wrote a comment, Copilot would autocomplete the whole function. So for the first time in your career you wrote good comments, because now someone was reading them.

Then came Cursor, and you switched editors, which you had sworn you would never do. It was mostly VS Code, so it didn't count. You asked for a change in a chat window and it edited the file for you. You still had to fix a lot of what it wrote. That felt good, because it meant you were needed.

In December 2024, I published a post called AI Tools Are Making Developers Stronger, Not Obsolete. It lists what AI coding tools couldn't do, including "Difficulty working across multiple files" and "Inability to fix their own errors." The post is still online. I'm leaving it there.

3. The high

In 2025 came Claude Code and Codex, which opened your files and changed the code themselves. Then they ran the tests to see if everything worked.

Work got so much easier, and you got more done in a week than you used to get done in a month. You finished your side project from 2019 on a Saturday afternoon. You ran four agents in four terminals and started calling them your team. Your GitHub contribution graph turned solid green.

The only thing that slowed you down was the usage limit, so you bought a second subscription. You told everyone it paid for itself. It did, just not for you.

When people asked if AI would take your job, you said the code was never the hard part. The hard part was figuring out what the client wants and how the system should work, and AI couldn't do that.

4. The realization

Then on September 22, Anthropic released Opus 5.5, and it turns out it can do that part.

You paste in the client's vague email, and it asks the questions you would have asked. Then it asks one you hadn't thought of. You ask it to review your architecture, and it finds the problem you were hoping nobody would notice. It works across multiple files, and it fixes its own errors. Both were on my list.

Anthropic's announcement mentions a tester who did a 680,000-line code migration with it in less than a day. The announcement calls that "work that would have taken an engineering team weeks." You are the engineering team in that sentence.

In the evening, you look at what you did today. You approved two plans and wrote "looks good" on three pull requests. Claude Code asks for permission before it runs a command, unless you start it with --dangerously-skip-permissions. You started doing that months ago.

Like the original five stages, these don't always come in order. A lot of developers are in stage 3 and stage 4 at the same time, sometimes in the same pull request.

5. Acceptance

For a few years, AI was a tool that helped you do your job. Now it does the job, and somebody only has to tell it what to build. Your product manager has found out that they can do that themselves.

Your company pays your salary, and it pays for the AI subscription. It has started to notice which of the two does the work.

HR has a chart for this. It's the change curve, and it tells managers what people need at each stage. People in denial need reassurance. People who have reached acceptance need concrete next steps.

Your calendar has a new meeting on Friday. It's called "Quick sync," and someone from HR is invited.

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com