OpenAI wants to simplify how developers build AI agents

OpenAI is releasing a set of tools and APIs designed to simplify agentic AI development, the company has revealed.

While the company has already added new capabilities to support agentic workflows, OpenAI said customers have found it challenging to use these capabilities in production, owing to extensive prompt iteration and custom orchestration logic.

The firm's new ‘Responses API’ aims to solve these problems, along with an ‘Agents Software Development Kit (SDK),’ new search tools, and new integrated observability tools, all aimed at supporting agent development.

The Responses API integrates with OpenAI’s existing ‘Chat Completions API’ and ‘Assistants API’ to assist in agent construction, while the Agents SDK helps users orchestrate both single and multi-agent workflows.

Built-in search tools include a web search and file search function, as well as a computer use function, while users can now take advantage of observability features.

“These new tools streamline core agent logic, orchestration, and interactions, making it significantly easier for developers to get started with building agents,” OpenAI said.

The firm also plans to release additional tools and features soon aimed at simplifying and speeding up agentic development on the OpenAI platform.

How do the new OpenAI tools work?

Responses API is designed to work with the search functions so that it can draw from the real world and more effectively complete entire tasks.

Its focus is primarily on third-party integration and is aimed at developers who want to bring OpenAI’s technology and agentic capabilities into their applications without having to use several APIs or external vendors.

The API can draw on up-to-date public context from the internet via the web search tool or on internal proprietary business context using the file search tool.

While the web search function has been useful in instances such as the creation of shopping assistants, research agents, and travel booking agents, OpenAI said the file search tool could be deployed in customer service and legal capacities.

Web search is priced at $30 and $25 per thousand queries for GPT‑4o search and 4o-mini search respectively. File search is available at a usage price of $2.50 per every thousand queries.

Similarly, file storage is priced at $0.10 per gigabyte (GB) per day, though the first GB is free.

Through the computer use function, Responses API can be used to create agents that perform functions on a computer using interactions like mouse clicks and keyboard strokes. This works in tandem with the model that enables OpenAI’s ‘Operator’ agent.

Computer use function usage costs $3 per each 1 million input tokens and $12 per each 1 million output tokens.

To manage agents once they’ve been developed, the new Agents SDK gives users ‘handoff’ functionality to transfer control between agents, as well as guardrail features for safety and validation checks.

Agents SDK also brings with it tracing and observability functionality so that developers can see how agents are performing and executing tasks, allowing them to debug and optimize accordingly.

OpenAI jumps on the ‘DIY’ agent trend

OpenAI is the latest in a string of firms to bring users the tools they need to build agents and agentic functionality, building on a market trend towards ‘do-it-yourself’ agent development.

Earlier this month, Databricks rolled out new tools to assist businesses in the deployment, monitoring, and integration of agentic tools.

The offering included centralized governance for AI model management, tools for simplified integration of chatbots and custom applications in third-party tools, and an agent evaluation tool.

Similarly, earlier this year Workday released its ‘Agent System of Record’ designed to lessen the burden of managing agentic AI systems.

The tools allow IT leaders to on-board agentic workers, define the roles of those agents, manage their access to business data and systems, and gain visibility into agent activity.

Some of the bigger tech firms have made similar moves, with AWS putting agentic building capabilities into its Amazon Bedrock platform as part AWS re:Invent 2024.

‘Amazon Bedrock Agents’ gives customers the tools to not just develop their own agents, but also orchestrate whole teams of agents to complete complex workflows.

MORE FROM ITPRO

• Elon Musk’s $97 billion flustered OpenAI
• OpenAI inks $12bn CoreWeave deal in latest move away from Microsoft
• AI might not replace software engineers just yet