PATTERN Cited by 1 source
Agent-first infrastructure API¶
A platform API design pattern where the infrastructure provisioning surface is designed for AI agent consumption as a first-class citizen — not as an afterthought bolted onto a human-facing UI. Chat, external services, and agents all access provisioning through a single MCP (or equivalent protocol) interface.
Structure¶
- Single MCP server as the canonical control point for all provisioning operations
- Natural-language-compatible input — agents describe intent, API maps to templates
- Multi-step workflow support — agents can chain provisioning + deployment + data loading + configuration in one sequence
- Skill-file extensibility — adding a
.mdskill file enables new agent capabilities without API changes
Rationale¶
When AI agents can provision infrastructure programmatically, multi-step workflows that previously required human coordination become trivial: "spin up a new workspace, deploy several DABs, upload S3 data, run a hydration script" — all from a single agent command. The MCP abstraction ensures the same logic serves human chat, external service integrations, and autonomous agent workflows.
Benefits¶
- Eliminates the UI-as-bottleneck for agent-driven workflows
- Hyper-personalised environment setup that would be impractical manually
- Single point of governance — all provisioning (human or agent) goes through one auditable path
- Skill-file extensibility means new capabilities don't require API changes
Vision¶
The endpoint of this pattern: an engineer describes a customer situation in plain language → an agent provisions an appropriately configured environment → retrieves relevant context → reproduces the issue or builds the demo → returns a summary ready for the customer.
Seen In¶
- sources/2026-07-23-databricks-self-serve-infrastructure-vending-machine — MCP at FEVM's core; "adding a simple .md skill file allows you to ask for a new environment on the command line in plain english and log in minutes later"