OpenAI DevDay: Model Providers Take Over the Agent Runtime Too — Convenience or Lock-In?

At its DevDay conference in San Francisco on September 29, OpenAI unveiled more than 20 new products and features. The common thread: OpenAI no longer wants to supply only models, but also the environment agents run in.

Key announcements

• Dots: agents that keep working on your goals in the background, even when you're not prompting them. Each runs on GPT-6 Astra, has its own cloud computer and browser, and can connect to more than 4,000 apps.
• Agents API (public beta): a managed agent runtime. OpenAI handles sessions, orchestration, context compaction and recovery; the developer defines only the tools and environment.
• With the Agents API, agents can use an OpenAI-hosted browser, connect to MCP servers, run parallel subagents and keep durable session state.
• Decisions API: uses Luna, OpenAI's cheapest model, for questions with a fixed set of answers.

Background: enterprise revenue is growing fast

According to Axios, OpenAI's annualized revenue has grown more than 70% since the start of the third quarter to nearly $70 billion, with business-to-business revenue more than doubling over the same period. These products are largely aimed at enterprises.

The cost of convenience: what lives where?

A managed agent runtime makes building agents much easier. But at the same time, an agent's session history, memory, tool connections and rules start being kept on a single provider's platform. If you later want to switch to another model or provider, what has to move is no longer just an API call but the agent's entire way of working.

Four questions for enterprises

• Whose infrastructure, and which country, holds the agent's memory and session history?
• Are security and permission rules defined in the provider's console or in a layer you control?
• Are tool connections built on open standards like MCP, or provider-specific?
• How much work would it take to run the same agent on a different model?

The remot angle

remot helps you use managed agent runtimes while keeping control inside the company: model and tool calls, security policies and logs are managed in a single, provider-independent layer. That way running an agent in OpenAI's environment today and on another model tomorrow is possible without rebuilding the rules and records.