OpenAI is shifting its agent-building strategy towards multi-step, "chain of thought" processes and multi-agent "swarms" to handle more complex tasks, supported by a new Agents SDK.
Learning from the high barrier to entry of the Assistants API, OpenAI's new Responses API is designed to be simpler, starting with a single stateless call and allowing developers to progressively add complexity.
A key belief is that current AI models are significantly underutilized; the most critical skill for developers is now orchestrating models and tools effectively, rather than waiting for more capable models.
The development of "Computer Use" models, which can operate software and browsers, is seen as a major future growth area, currently in its early stages (analogous to GPT-1/2) with applications emerging in medicine and research.
12 quotes
Concerns Raised
The high complexity of previous APIs (like Assistants API) created a barrier to entry for developers.
AI has not yet delivered a step-change in the speed of scientific research, an early hope for the technology.
The process of orchestrating models, prompt engineering, and maintaining evaluation sets is still extremely difficult for application builders.
Opportunities Identified
Building applications that better leverage the underutilized capabilities of current foundation models through sophisticated orchestration.
Developing domain-specific agents in verticals like medicine, climate tech, and travel.
Extending the runtime of AI agents from minutes to hours or days is expected to unlock significant new capabilities.
Creating third-party AI infrastructure companies that provide specialized environments and tools for agentic workflows.