Why is verification so important? Because, if you tell an agent to do something that it can't independently verify, then—just like a human developer—the best they can do is guess. And because agents work really fast, each action based on a guess is quickly succeeded by an even more tenuous guess. And then a guess of a guess of a guess, and so on. Very often, when I return to my desk after 30 minutes and find that an agent made a huge mess of the code, I come to realize that the AI didn't suddenly "get dumb," but rather that an application server crashed or a web browser stopped responding and the agent was forced to code speculatively and defensively.

The upshot here is that when there's no platform you can trust to be real, the next best thing is a platform where you can trust everything is fake.

I had not thought of it this way but I love the idea and it is very true and feels right. I often create posts that are just links to stuff I read on the internet. Typically I don't write any comment at all. I simply quote the relevant part of the link.

Here's how to leave a comment on this web site:

1. Read a post
2. Think, "I want to comment on this"
3. Draft a post on your blog
4. Add a hyperlink to my post
5. Paste an excerpt to which you want to respond
6. Write your comment
7. Hit publish

A lot of developers are feeling scared and hopeless about the changes being wrought by all this. Yes, AI is being used as an excuse by executives to lay people off and pad their margins. Yes, how foundation models were trained was unethical and probably also illegal. Yes, hustle bros are running around making bullshit claims. Yes, almost every party involved has a reason to make exaggerated claims about AI. All of that can be true, and it still doesn't matter. Your job as you knew it is gone.

Get serious about learning and using these new tools. You will, like me, recoil at first. You will find, if you haven't already, that all these fancy AI tools are really bad at replacing you. That they fuck up constantly. Your new job starts by figuring out how to harness their capabilities anyway. You will gradually learn how to extract something that approximates how you would have done it yourself. Once you get over that hump, the job becomes figuring out how to scale it up. Three weeks ago I was a Cursor skeptic. Today, I'm utterly exhausted working with Claude Code, because I can't write new requirements fast enough to keep up with parallel workers across multiple worktrees.