It doesn't necessarily have to be harmful for it to be stigmatized by society.
1,639 karma · joined October 24, 2025
Blog: orchidfiles.com
It doesn't necessarily have to be harmful for it to be stigmatized by society.
Thanks for reading! When writing this essay, I drew solely on my own experience. I’ve often noticed that startups post job listings with misleading job descriptions, especially in stigmatized industries. It’s only after the interview that they reveal what the work will actually entail. Perhaps you simply haven’t noticed such job listings.
The main reason this exists is that Cursor can talk to a custom OpenAI-compatible endpoint, but its backend cannot reach localhost, so the extension manages the proxy + tunnel + provider auth flow for you.
Known limitation: this only works reliably with custom model IDs configured through the Ungate dashboard. Cursor’s built-in model IDs can bypass the custom base URL.
But the decision has been made. And if AI remains in databases after the AI hype, the next generation of developers will no longer know databases without AI. That's why I said it will become the norm. But that's just my guess. Unfortunately, I can't see the future. I don't know what it will actually be like in a few years.
The blog has no monetization or advertising. Right now, I am only investing my time in creating content and looking for founders to talk to. My plan is to build an audience over the next few years and then offer consulting services.
But other programmers who see this content immediately realize that it was generated and stop reading it.
This problem can be solved not only by automating the process. For example, one solution is to simply stop doing all of this. You don't need to analyze your competitors every morning; it's enough to do it once a month or even once a quarter. Reading news about AI is one way to procrastinate.
Your precious morning time should be spent on truly important tasks.
Many business founders are willing to spend entire days on meetings, even though everything can be discussed asynchronously via text, and even faster.
I know that most developers dislike meetings and conference calls, and as a team lead, I consider it my duty to allow them to work without meetings.
When I launch my startups, I always organize the work of the entire team at the very beginning so that it can function without a single conference call. As a founder, I have never needed to call anyone on the team to resolve any issues.
In Auto-Run in Sandbox mode, the Command Allowlist isn’t applied. Sandbox replaces per-command approvals with filesystem and network restrictions, but the UI doesn’t explain this well.
Workaround: enable Legacy Terminal Tool in Cursor Settings > Agents > Inline Editing & Terminal. With that, the allowlist works as expected, and git commands will require approval again.
I also use similar questions for different situations, and different ones for different projects. So, to avoid writing these questions every time, I created commands for each question. There is also a "/capture" command that analyzes the dialogue and suggests what to save from it, what commands and rules to create. The "/insights" command not only analyzes the current dialogue, but also suggests ideas for creating new commands and rules. If you have asked the same question several times in the dialogue, these commands will suggest creating new commands specifically for these questions.
This is strikingly different from development. In development, AI increases my productivity fivefold, but in texts, it slows me down.
I thought, maybe the problem is simply that I don't know how to write texts, but I do know how to develop? But the thing is, AI development uses standard code, with recognized patterns, techniques, and architecture. It does what (almost) the best programmer in their field would do. And its code can be checked with linters and tests. It's verifiable work.
But AI is not yet capable of writing text the way a living person does. Because text cannot be verified.
But I think this is individual.
Another example. I asked gpt-5.2-codex to add an array of 5 values to the script and write a small piece of code. Then I manually deleted one of the values in the array and asked the agent to commit. But the model edited the file again and added the value I deleted to the array. I deleted that value again and asked the agent to “just commit.” But the agent edited the file again before committing. This happened many times, and I used different commands, such as “never edit the file, just commit.” The model responded that it understood the command and began editing the file. I switched to gpt-5.2, but that didn't help.
I switched to sonnet-4.5, and it immediately committed on the first try without editing the file.
Google Search Console shows the user's query if the query is popular enough and your website is in the search results. Bing shows all queries, even if they are not popular, and if your website is in the search results.
But if AI recommends your website when answering people's questions, you cannot find out what questions the user discussed, how many times your website was shown, and in what position. You can see the UTM tag in your website analytics (for example, GPT adds utm source), but that is the maximum amount of information that will be available to you. But if a user discussed a question with AI and only got your brand name, and then found your site in a search engine, you won't be able to tell that they found you with the help of AI advice.