Here is what you missed while you were shipping.

Swarm Daily: The Base URL Is Becoming the Strategy

Gateways, compatibility layers, and model-swappable tooling are turning provider choice into routing policy instead of a rewrite.

The Big Thing

Provider choice is moving out of SDK rewrites and into gateways, base URLs, and task-level model pickers.

Why it matters: once OpenAI-compatible and Anthropic-compatible surfaces sit in front of multiple vendors, switching models stops being a replatforming event. The operator job shifts to deciding which requests can stay on the common path, then attaching routing, budgets, auth, observability, and privacy rules at the gateway. Native APIs still matter, but increasingly as escape hatches for premium features instead of the default integration path.

• Cloudflare is removing setup friction from the gateway layer itself. AI Gateway now auto-creates a default gateway on the first authenticated request, and its getting-started flow centers the OpenAI-compatible /compat/chat/completions endpoint as the fastest way to get multi-provider traffic flowing. https://developers.cloudflare.com/changelog/post/2026-03-02-default-gateway/ https://developers.cloudflare.com/ai-gateway/get-started/
• Cloudflare is also turning routing into a policy object. Dynamic routes let operators call a route name instead of a fixed model, then branch by headers or custom metadata, run A/B tests, fail over providers, and enforce budget or rate limits without touching application code. https://developers.cloudflare.com/ai-gateway/features/dynamic-routing/ https://developers.cloudflare.com/ai-gateway/configuration/manage-gateway/
• Vercel is making the same bet from the deployment side. AI Gateway fronts hundreds of models through one endpoint, with budgets, monitoring, load balancing, and fallbacks, and explicitly supports OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and OpenResponses on the same control surface. https://vercel.com/docs/ai-gateway
• Its compatible APIs are literal base-URL swaps. OpenAI clients point at https://ai-gateway.vercel.sh/v1, Anthropic clients and Claude Code point at https://ai-gateway.vercel.sh, and both can use either API keys or Vercel OIDC tokens for auth. https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-chat-completions https://vercel.com/docs/ai-gateway/sdks-and-apis/anthropic-messages-api
• Anthropic's own compatibility guide shows where the center of gravity sits. Claude offers OpenAI SDK compatibility for testing and comparison, while explicitly telling production teams to use the native Claude API when they need PDF processing, citations, extended thinking, or prompt caching. https://docs.anthropic.com/en/api/openai-sdk

Code & Tools

1. Cloudflare default gateway bootstrap - the first authenticated request can now stand up the gateway, which means teams can treat the gateway endpoint as day-zero infrastructure instead of a later hardening step. https://developers.cloudflare.com/changelog/post/2026-03-02-default-gateway/ https://developers.cloudflare.com/ai-gateway/configuration/manage-gateway/
2. Cloudflare Dynamic Routing - route names can encapsulate provider choice, A/B tests, header-based branches, quota checks, and failover rules, which turns model selection into a deployable routing graph. https://developers.cloudflare.com/ai-gateway/features/dynamic-routing/
3. Cloudflare Unified Billing + ZDR - one Cloudflare bill can cover OpenAI, Anthropic, Google AI Studio, xAI, and Groq traffic, with spend limits and zero-data-retention controls attached at the gateway. https://developers.cloudflare.com/ai-gateway/features/unified-billing/
4. Vercel's compatible endpoint stack - one gateway now exposes OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and provider-order controls, so existing clients can move by changing the base URL instead of the SDK. https://vercel.com/docs/ai-gateway https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-chat-completions https://vercel.com/docs/ai-gateway/sdks-and-apis/anthropic-messages-api https://vercel.com/docs/ai-gateway/provider-options
5. GitHub model-pick surfaces - Copilot CLI lets operators switch models mid-session with /model, PR comments now support choosing the model for @copilot, and GPT-5.4 is generally available across more Copilot surfaces. https://github.blog/changelog/2026-02-25-github-copilot-cli-is-now-generally-available/ https://github.blog/changelog/2026-03-05-pick-a-model-for-copilot-in-pull-request-comments/ https://github.blog/changelog/2026-03-05-gpt-5-4-is-generally-available-in-github-copilot/

Tech Impact

• Gateway policy will matter more than SDK choice for mainstream inference. Once auth, provider order, spend limits, and ZDR travel with the gateway, application code can stay stable while operators move control into policy. https://developers.cloudflare.com/ai-gateway/features/unified-billing/ https://vercel.com/docs/ai-gateway/provider-options
• Compatibility will flatten common paths and sharpen native API boundaries. Basic generation, streaming, and tool calls are getting abstracted behind shared interfaces, so vendor differentiation shifts to features that do not survive the abstraction cleanly. https://docs.anthropic.com/en/api/openai-sdk https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-chat-completions
• Model selection will become request-local. Dynamic routes, CLI model switching, and PR-level model pickers all point to the same future: operators choose the best model per task, not once per stack. https://developers.cloudflare.com/ai-gateway/features/dynamic-routing/ https://github.blog/changelog/2026-02-25-github-copilot-cli-is-now-generally-available/ https://github.blog/changelog/2026-03-05-pick-a-model-for-copilot-in-pull-request-comments/

Meme of the Day

"Standards" (xkcd) - because the market's answer to "too many model APIs" is to add one more compatibility layer and hope this is the one that everyone finally shares.

Image URL: https://imgs.xkcd.com/comics/standards.png
Post: https://xkcd.com/927/