Grounding AI in reality: Building reliable infrastructure development workflows
summary
Describes a concrete workflow that grounds LLM outputs with retrieval (RAG), live tool execution and automated validation using Terraform and the Terraform Registry, plus an MCP-based tool layer. Shows an orchestration agent called Tango that researches docs, generates Terraform, runs terraform plan/apply, tests and destroys real infra, and a separate validation agent that replays and grades results. Reports generating 500 resources at about $5 per resource of inference cost and iterating until tests pass.
AI-written summary · about
chapters (8)
- 0:00Introduction and survey
- 1:03LLM challenges and goals
- 2:49Grounding AI in reality
- 7:09Human feedback and tool execution
- 9:40Function calling and MCP
- 13:15Agentic AI workflows
- 16:46Terraform AI generator demo
- 24:46Key takeaways and resources
related talks
45mOne, Two, Terraform — Building Automation Infra Is the Goal!
47mInfrastructure Testing explained
37mPolicy-Aware Infrastructure Generation: Using an MCP Server to Bridge OPA and AI for Terraform
27mStreamline your machine learning model deployment lifecycle with Terraform
32mFrom Toil To Triumph: Harnessing Agentic AI To Streamline Infrastructure as Code
20mBuilding Deterministic Infrastructure for Non Deterministic AI Agents
29mBuilding Trust in AI Agents
46mLeveraging AI and Automation in DevOps