Artificial Intelligence (AI)

OpenAI Launches New Tools to Build AI Agents

Ken Metral

Ken Metral

— 2 min read
Cosmico - OpenAI Launches New Tools to Build AI Agents
Credit: OpenAI, Inc.

OpenAI has unveiled a suite of new tools aimed at bringing practical, scalable AI agents into the hands of developers and enterprises. At the core of this release is the Responses API, which gives builders direct access to the underlying components powering OpenAI's most advanced agentic technologies—including web search, file analysis, and even keyboard/mouse control.

These tools mark a shift from hype to utility in the AI agent space. For years, “AI agents” have been a buzzword in tech circles, but the gap between demo and real-world performance has remained wide. With the Responses API, OpenAI hopes to close that gap.

“It’s pretty easy to demo your agent,” said Olivier Godemont, OpenAI’s API product head. “To scale an agent is pretty hard, and to get people to use it often is very hard.”

What the Responses API Can Do

The Responses API allows developers to build their own versions of OpenAI’s Operator and deep research tools—agent-based systems that can browse websites and compile reports. This includes:

  • Web Search with Source Citations: Using GPT-4o search and GPT-4o mini search, agents can fetch real-time answers from the web, citing sources in their responses. Notably, these models outperformed even GPT-4.5 on factual benchmarks, scoring 90% and 88% respectively on OpenAI’s SimpleQA test.
  • File Search Utility: Enterprises can scan across internal files to extract insights without sending data back to OpenAI’s servers for training.
  • Computer-Using Agent (CUA): This model can simulate keyboard and mouse activity to automate tasks like data entry or navigating workflows. It’s the same core tech behind OpenAI’s Operator and can even run locally for enterprise customers.

However, these tools are not without limitations. Despite improved factual accuracy, GPT-4o search still gets 1 in 10 answers wrong and struggles with short queries like “Lakers score today.” The CUA model, while groundbreaking, is still in its early stages and not yet highly reliable for operating system-level tasks.

Goodbye Assistants API, Hello Agentic Future

The Responses API is set to replace OpenAI’s Assistants API, which will be sunset in the first half of 2026. It reflects a broader vision: that AI agents, not just chatbots, are the real future of productivity.

To support this shift, OpenAI is also releasing an open-source Agents SDK. The SDK allows developers to integrate AI agents into their systems, implement safety mechanisms, and monitor agent behavior for debugging and optimization. It builds on Swarm, OpenAI’s multi-agent orchestration framework released last year.

Why It Matters

The launch comes amid growing scrutiny of the “AI agent” concept. Earlier this week, Chinese startup Butterfly Effect went viral for overpromising on its agent platform Manus—a reminder of how quickly enthusiasm can turn into backlash when tools fall short.

OpenAI is aware of the stakes. CEO Sam Altman has declared that 2025 will be “the year AI agents enter the workforce,” and these new releases are a concrete step in that direction.

Instead of offering a prebuilt agent platform, OpenAI is choosing to empower others to build agents tailored to their own workflows, tools, and datasets. This modular, developer-first approach could be key to turning agent hype into real-world impact.

Bottom line: OpenAI is no longer just showing off what agents could do. It’s handing developers the tools to build what agents should do. Whether 2025 becomes the year of the AI agent remains to be seen—but OpenAI just raised the bar.

Read more