OpenAI has introduced new tools aimed at helping businesses and developers create AI-powered agents, marking a significant step towards advancing autonomous AI applications.
Introducing the Responses API
On Tuesday, OpenAI launched the Responses API, a new framework designed to enable enterprises to develop AI agents that can browse the web, scan company files, and automate tasks. This API replaces OpenAI’s Assistants API, which will be phased out by mid-2026.
The launch comes amid a growing race to develop truly autonomous AI agents, which can execute complex workflows with minimal human intervention. While AI agents have gained significant traction, recent launches—such as Manus AI by Butterfly Effect—have failed to meet expectations, underscoring the challenges in delivering functional AI-powered automation.
How OpenAI’s AI Agents Work
The Responses API gives developers access to OpenAI’s latest AI models, including GPT-4o search and GPT-4o mini search, which power ChatGPT Search. These models help AI agents fetch real-time information from the web, generate responses with source citations, and analyze documents in a company’s database.
According to OpenAI, GPT-4o search has a 90% accuracy rate on the company’s SimpleQA benchmark, outperforming previous AI search models. However, OpenAI acknowledges that hallucinations and incorrect responses remain a challenge, as AI-generated search results are not always reliable.
Expanding AI Agent Capabilities
The Responses API also includes:
- A file search tool to extract relevant data from internal company databases.
- The Computer-Using Agent (CUA) model, which automates keyboard and mouse actions for enterprise software, streamlining processes like data entry and workflow execution.
Companies can host the CUA model on-premise, while the consumer-facing version will remain limited to web-based actions through OpenAI’s ChatGPT Operator tool.
Challenges in AI Agent Development
Despite the promise of AI agents, significant hurdles remain. OpenAI itself admits that its CUA model is still unreliable for automating operating system tasks, meaning human oversight is still necessary. Additionally, while AI search models improve accuracy, GPT-4o search still gets 10% of factual queries wrong.
AI agents also struggle with short, navigational queries such as live sports scores, and ChatGPT’s citation reliability has been questioned in recent reports.
A Step Towards Enterprise AI Adoption
Alongside the Responses API, OpenAI has also released the Agents SDK, an open-source toolkit designed to help developers integrate AI models with internal systems, implement safeguards, and monitor agent activities for debugging and optimization.
OpenAI’s head of API product, Olivier Godement, emphasized the company’s goal of moving from flashy AI agent demos to real-world enterprise adoption. CEO Sam Altman had earlier predicted that 2025 would be the year AI agents enter the workforce, making them a mainstream part of business operations.
Also read: AI-First Startups to Dominate 2025: Google Cloud Report
Shaping the Future of AI Agents
As AI-powered automation gains momentum, OpenAI’s latest tools could pave the way for more capable, enterprise-ready AI assistants. However, with ongoing challenges in reliability and accuracy, the road to fully autonomous AI agents remains a work in progress.
(This article was first reported by TechCrunch.)
