Developer tooling goes enterprise

The software that lets an artificial intelligence agent use tools is starting to look like enterprise infrastructure, not a demo. OpenAI added production controls to its Agents software kit on April 15, while Anthropic has begun charging Claude users by usage for third-party agent tools. (openai.com) (techcrunch.com) An agent software development kit is the plumbing behind systems that can inspect files, call tools, run commands, and pass work between specialized assistants. OpenAI said its updated Agents software development kit now includes a “model-native harness” for working across files and tools on a computer, plus native sandbox execution for running that work in a controlled environment. (openai.com) OpenAI’s documentation says the kit already included tracing for visualizing and debugging agent workflows, handoffs for delegating work between agents, guardrails for validating inputs and outputs, and human-in-the-loop controls. The company now describes the package as a “production-ready upgrade” of its earlier Swarm experimentation project. (openai.github.io) OpenAI framed the April 15 release around a problem developers hit when they move from prototypes to production. The company said model-agnostic frameworks can miss model-specific capabilities, while managed agent application programming interfaces can limit where agents run and how they access sensitive data. (openai.com) Anthropic is making a parallel shift on the business side. TechCrunch reported on April 4 that Claude Code subscribers would no longer be able to use subscription limits for third-party harnesses including OpenClaw, and would instead need a separate pay-as-you-go balance. (techcrunch.com) International Business Times Australia reported that the Anthropic change took effect at noon Pacific on April 4 for tools like OpenClaw and is expanding to other third-party harnesses. The report said Anthropic offered a one-time credit equal to a subscriber’s monthly plan cost, redeemable by April 17 and valid for 90 days, along with discounts of up to 30% on prepaid usage bundles. (ibtimes.com.au) Anthropic’s explanation was capacity and cost. Boris Cherny, the head of Claude Code, said on X, as quoted by TechCrunch, that subscriptions “weren’t built for the usage patterns of these third-party tools,” while TechCrunch also reported his statement that the change was “more about engineering constraints” than hostility to open source. (techcrunch.com) The tension is clearest in the reaction from OpenClaw’s creator. TechCrunch reported Peter Steinberger said Anthropic had copied popular features into its own harness and then “lock[ed] out open source,” while Anthropic said it still supports access through its own interfaces, including claude.ai and Claude Code, under existing subscription limits. (techcrunch.com) (ibtimes.com.au) The result is a more familiar enterprise pattern: tighter controls on how agents run, and metered billing for heavy use. OpenAI is selling safer deployment primitives, while Anthropic is drawing a clearer line between consumer-style subscriptions and the compute costs of automated agent workloads. (openai.com) (ibtimes.com.au)