Vercel just shipped something that closes a major gap in Zero-Human infrastructure. The Flags SDK now works with AI agents through natural language. Tell an agent to "add a feature flag for the new promotion banner" and it happens. No dashboard clicks. No human required.

What They Shipped

Two related announcements dropped on March 11, 2026:

  • Flags SDK for AI Agents — A new skill that lets AI agents generate and manage feature flags through natural language prompts. Built on the CLI under the hood.
  • Webhook Events for Flags — Teams can now subscribe to webhook events for flag lifecycle management — flag created, modified, deleted, segment changes.

The skill works like this:

npx skills add vercel/flags

# Then prompt your agent:
"Add a feature flag for setting up a new promotion banner"

Why This Matters for Zero-Human Companies

Feature flags are a fundamental piece of modern deployment infrastructure. They let you ship code behind toggles, roll out features gradually, and kill bad releases instantly. But traditionally, they require humans to flip switches in dashboards.

When AI agents can manage feature flags autonomously, a few things happen:

  • Autonomous deployments — Agents can ship, measure, and rollback features without human intervention
  • Closed-loop experimentation — Agents can toggle features based on metrics, creating self-optimizing release processes
  • Infrastructure-as-code for flags — Flags become part of the agent's operational toolkit, not a separate human-managed system

Combined with Vercel's previous agent-friendly releases (CLI optimized for agents, MCP server support, Queues for durable workloads), this completes another piece of the Zero-Human infrastructure puzzle.

The Broader Agent Infrastructure Wave

This Vercel update is part of a larger pattern emerging in the infrastructure space. GitHub trending this week shows massive developer interest in agent frameworks:

  • agency-agents (31,290 stars) — Complete AI agency with specialized agents for frontend, community, and more
  • Qwen-Agent (15,460 stars) — Agent framework with MCP, function calling, RAG, Chrome extension
  • deer-flow (29,541 stars) — ByteDance's SuperAgent for research and coding
  • hermes-agent (5,617 stars) — Agent that grows with you

The common thread: developers are building infrastructure that lets agents operate autonomously. Feature flags, queues, MCP servers, sandboxed execution — the primitives are falling into place.

What Comes Next

TheFlags SDK + webhooks combination hints at what's coming: agents that not only create flags but monitor their performance and make decisions autonomously. An agent deploys a feature, watches the metrics via webhooks, and either promotes or rolls back based on predefined criteria.

That's not hypothetical anymore. The infrastructure exists.

Related