LowKeyDevs@LowKeyDevs
For the past few days I've been thinking about a problem and been researching and implementing stuff related to it.
The problem infrastructure is a domain where we're still very reluctant to use AI agents to do stuff. Obviously the usage isn't zero, but the AI workflow hasn't been pushed into that domain properly.
The main reason for this, I think, is trust. We can't trust agents to do these things, even though models nowadays are very much capable of handling it.
So I thought to build a platform that could help solve this. First I built something like an "agent for infra" basically just calling AWS APIs with some extra policies and a trust layer imposed on top.
Then I realized this didn't make sense, because nobody does stuff directly in AWS or other infra platforms. Every company or org has their own workflow to handle their infra. What if I could find a way to push agents into their workflow instead?
This led me to think why not build something for each component of the workflow (PRs, observability, cloud, etc.)? But this also didn't make sense, because I'd be building something already built by other orgs and working well (CodeRabbit, Greptile, etc.).
After this, I thought why not build something monolithic that can be used by the whole workflow? The first thing that came to my mind was an MCP server, because it's the closest thing that can fit into an infra workflow.
So yeah! This is what I'll be working on for a long time and will keep posting updates here.
^ the first implementation i mentioned
