I’ve been exploring LLM-driven developer workflows and built a demo where Claude Code acts like a Replit-style frontend for infrastructure.
– You write a prompt in Claude
– Claude generates a workload spec
– Humanitec receives the spec and deterministically deploys infra to GCP (via Terraform)
– App is up in under 1 minute — no pipelines, no DevOps tickets
The backend is policy-enforcing and built for enterprise use cases (used by large orgs like Apple, ABB, etc.), but the flow feels like magic for devs or agents.
demo: https://www.youtube.com/watch?v=jvx9CgBSgG0
Repo: https://github.com/DemoKaspar/vibe-coding-template
Curious how others are thinking about agent-first infrastructure. What would you add/change?
Comments URL: https://news.ycombinator.com/item?id=45513819
Points: 4
# Comments: 0
Source: news.ycombinator.com
