OpenAI Declares 2025 “The Year of the Agent”
Yesterday, Mar 11, 2025, OpenAI announced new agentic tooling, and declared 2025 “the Year of the Agent.” Two separate but complementary things were announced, the Responses API, and the Agents SDK.
Tool Use Comes to the API
The Responses API is a replacement and expansion of the Completions API that we’ve been using to build applications with OpenAI models. Though Completions will continue to work, many of the other new fancy features that were announced are only available in the Responses API.
In particular, the following tools that ChatGPT users have come to know and love, will now finally be available in the Platform API, so we can use them to build our own applications:
Web Search - the agent handling your response can now make web searches to enrich its response
File Search - hookup knowledge sources, and let ChatGPT power your RAG applications
Computer Use - the agent can use a browser on your behalf
Again all these things were available already in ChatGPT, however for developers using OpenAI’s platform APIs to build their own applications, all we got were the raw models themselves, and perhaps a little memory. Now those fancy ChatGPT features are available in the API, and AI-driven applications will be far easier to build.
Agents SDK: Orchestration, Guardrails and Observability
The second major announcement was the Agents SDK. This may prove to be the Kubernetes of the Gen AI agent-driven application space. Just as Kubernetes became the de facto standard for hosting websites and other workloads, LangGraph and Crew AI had been competing to be the tool to orchestrate and guard agentic bots and applications. However, now OpenAI has put their own competitor in the race, and even open sourced it!
The Agents SDK can be installed with PIP (the common package manager for Python; Node.js support is coming soon), and provides clean conventions for defining multiagent interaction, tool use, and more. It includes Tracing and Observability so enterprises building internal AI agentic applications can audit what the agents are doing, where data is going (in particular if it leaves the enterprise, called data exfiltration), where there are performance bottlenecks, and more.
The Future is Now
While this may all sound jargony, the sum of it is that OpenAI is putting out a robust set of tools for moving beyond mere chatbots, and enabling the tech industry to build full on agents like Jarvis from Iron Man, Samantha in the movie Her, and the computer in Star Trek. This will be the year that generative AI tools transition from the chatbot proof of concept we’ve been testing out the past two years to the artificial coworkers and hybrid intelligence of the late 2020s. Strap in cause blast off is imminent!
Read Other Posts
