Cloudflare has released cf, a new CLI that replaces Wrangler and expands its command surface from roughly 280 operations to over 3,000, covering the entirety of Cloudflare's API. The tool is designed primarily for AI agents, not humans — a design choice backed by usage data showing agents now account for 48% of Wrangler usage, up from single digits a year ago. Agents use almost twice as many distinct commands per day and are four times more likely to chain six or more commands together. The architectural enabler is Forge, Cloudflare's unified API generation pipeline that auto-generates CLI commands from OpenAPI schemas. Instead of hand-building each command with per-team conventions — the approach that produced Wrangler's inconsistent terminology across d1 info, hyperdrive get, and workflows describe — cf treats the API schema as the single source of truth. Annotate the schema once, and the CLI, documentation, and SDKs all derive from it. This is a genuine efficiency unlock: it eliminates the coordination tax of dozens of product teams building bespoke CLI experiences independently. The agent-first design philosophy runs deep. JSON is the default output format, not tables. A natural-language search command (cf cli search) lets agents find the right operation among 3,000 routes without bloating context windows. The new cloudflare.config.ts configuration format replaces TOML with TypeScript, giving agents LSP-powered type-checking and autocomplete. Cloudflare reports that some internal Wrangler configs of 5,000+ lines were condensed by 40% through programmatic environment generation. The honest framing here is that Cloudflare is building infrastructure for a world where the developer is one step removed from the tool. Their own language is explicit: "You as the human customer of this CLI are, in reality, one step removed from using it." This is a platform play dressed as a developer tool. The more surface area agents can access, the deeper the lock-in — and the harder it becomes to migrate away from Cloudflare's ecosystem when your entire deployment pipeline is agent-mediated through cf. The generative case is real. Expanding from 280 to 3,000 operations genuinely creates new capability: agents can now set up workers, deploy, monitor, configure WAF, buy domains, and manage access policies from a single tool. The TypeScript config format with bindings helpers (bindings.kv, bindings.d1, bindings.r2, bindings.queue, bindings.ai, bindings.vectorize) provides a discoverable surface that makes Cloudflare's platform more accessible to both agents and developers. Vite becomes the default local development server, and a migration path from Wrangler is provided. The extractive risk is platform capture through abstraction layers. When agents mediate between developers and infrastructure, the platform that controls the agent interface controls the relationship. Cloudflare's intention to expand cloudflare.config.ts beyond Workers to manage "Cloudflare as a whole" signals a configuration monoculture strategy. Every product, every trigger, every binding flows through a single TypeScript file and a single CLI. That's powerful for usability and dangerous for portability. The broader signal matters more than the product: CLI design is now agent-first by default. Cloudflare's data — 48% agent usage, 2x command breadth, 4x command chaining — is the clearest public evidence yet that AI agents are becoming the primary consumers of developer infrastructure. The companies that recognize this fastest will capture the agent-mediated developer workflow. The companies that don't will find their tools bypassed entirely.
