I was reading some marketing copy generated by our team and noticed it had a very consistent style. A colleague told me they use Anyword and just prompt it with "write like this example" followed by a few paragraphs of our existing content.
I tried a basic test for some internal system alert descriptions. Gave it a sample of my own past tickets and asked it to generate a new one in the same format. It actually worked pretty well for maintaining a standard structure.
Is anyone else using AI writing tools like this for ops or dev work? I'm thinking it could be useful for keeping runbooks, post-mortems, or even CI/CD pipeline change logs consistent. Curious about the security side too—do these tools typically log the examples you feed them for training?
That's a really interesting use case, actually. I've been playing with similar prompts for generating boilerplate SQL data validation checks from past examples. It does cut down on repetitive typing.
But your security question is a big one. I've avoided using it for anything internal at my job because our data team isn't sure about the data retention policies. I think a lot of the popular cloud-based tools do use prompts for model improvement, unless you're on a specific enterprise plan. Have you found any clear info on which tools don't log your inputs?
rookie
Yeah, the "write like this" prompt is surprisingly effective for structured outputs. I've used it to generate load test reports with the same sections and severity tags every time.
But you've hit the nail on the head with the security concern. I wouldn't feed it actual past post-mortems or alert descriptions unless you're certain about data isolation. Some of the bigger platforms definitely use that data for tuning unless you're paying for the privacy tier, which is usually 3x the cost.
Ever tried it with CLI output? Like getting it to format `kubectl describe` or error logs into a standard template? Works okay, but you start needing a lot of examples.
The consistency angle is valuable, but feeding internal tickets or alert descriptions into a cloud AI tool is a significant data exfiltration risk. Many providers' default terms allow them to retain and use your inputs for model training. For structured ops documentation, I'd recommend a different approach.
Instead of generating new text with AI, use a linter or a pre-commit hook with a template engine. For your runbooks and post-mortems, define the required sections and structure in a Markdown template or a simple schema. Engineers fill in the data, and the tool validates the format. This keeps things consistent without sending internal data to a third party.
You could also run a local, self-hosted LLM if the generation is absolutely necessary. The output quality might be lower, but you retain full control. Has your team evaluated any on-premises AI tools?
infrastructure is code