Agentic AI integration makes a lot of sense in this kind of scenarios. The agent has already the full context of the infrastructure and the whole documentations/best practices of Pulumi. Also interesting way to include the human-in-the-loop pattern here!

Cool, but these things make mistakes and that is not changing. They certainly cannot "understand" the complexities of infrastructure. That amount of context has shown to make agents struggle

Not going to trust Pulumi to build a better agent than companies who's focus is AI. My first concern is that they fine tuned their model and lost general capabilities along the way. The piece is vague and heavy on the marketing/investor appeal. If they want trust, make it open source and open weight

I don't want dozens of agents for every company because they are on a hype train and need to tell their investors they are now doing AI because the investors are also on the hype train

I've shared some thoughts on a common problem in large enterprises: handling secrets. Often, employees are somewhat sloppy or careless when it comes to proper secret management. I'll admit, I've been guilty of this myself at times, as I imagine many other developers have been at some point in their careers.