OpenAI Updates API for Agent Skill Enablement

OpenAI Updates API to Enable New Features

OpenAI launches a significant update for its Response API, allowing developers to integrate agent skills and use a full terminal. The changes aim to overcome context limitations in interactions, providing models with longer-lasting memory and tools for complex operations.

The updates include Server Compression, Hosted Shell Containers, and a new standard called Skills. These improvements transform how agents operate, enabling them to perform prolonged tasks with more efficient memory.

Overcoming 'Context Amnesia'

The primary barrier for autonomous agents was the accumulation of long tasks. Each call to a tool or script execution adds to the conversation history, quickly hitting the model's limit.

OpenAI's response is Server Compression, which allows agents to operate for long periods without losing accuracy. Data from the e-commerce platform Triple Whale shows that its agent, Moby, handled 5 million tokens without losing precision.

This allows the model to "summarize" its past actions while retaining useful context and eliminating irrelevant information, transforming it into a persistent system.

Managed Cloud Containers

The introduction of the Shell Tool marks OpenAI's entry into managed computing. Now, developers can choose the container_auto, which provides a Debian 12 environment hosted by OpenAI.

Each agent will have access to a complete terminal, with included native execution environments such as Python 3.11 and Node.js 22, along with persistent storage at /mnt/data.

This framework allows agents to perform complex data transformations without requiring custom ETL infrastructure for each project, optimizing data engineers' workload.

Comparison of OpenAI and Anthropic Skills

OpenAI faces challenges from Anthropic regarding Skills. Both companies use a similar file structure, but differ in their strategies for the future of work.

OpenAI's approach emphasizes an integrated environment for developers, enabling quick creation of complex agents. In contrast, Anthropic focuses on open standards, offering portability and flexibility.

Recently, the enterprise search startup Glean reported significant accuracy improvements by implementing OpenAI's Skills framework, rising from 73% to 85%.

Implications for Decision-Makers in the Industry

The updates offer a significant productivity boost, allowing engineers to focus on rapid deployments and improvements. Skills enable the creation of modular expert knowledge that can be reused in internal projects.

For security managers, granting AI models access to networks is an evolution that demands attention. OpenAI's security strategies include Domain Secrets to ensure secure API calls.

Final Considerations for Companies

OpenAI now offers a complete package: not only an advanced model but also an operating environment and a training manual. The choice becomes clear:

• Choose OpenAI for an integrated and agile environment for long-duration autonomous work.

• Choose Anthropic if your organization seeks portability and an open ecosystem standard.

These updates indicate a shift in the AI landscape, moving from mere chat interactions to a scalable and sustainable infrastructure.