At its DevDay event on September 29, 2026, OpenAI unveiled Dots, an agent designed to run continuously in the background rather than wait for a prompt. Instead of a single back-and-forth conversation, Dots is built to stay connected to a user's linked apps and keep working toward a goal over time, on a dedicated cloud environment rather than a local session. Access is rolling out in stages, starting with Pro and Business Premium users, with enterprise, education, and healthcare organizations able to enable it in beta through their administrators.

Codex moves further into "do the work," not just "suggest the code"

Alongside Dots, OpenAI expanded Codex, its coding agent, so it can now run tasks in the cloud rather than only on a developer's local machine — meaning someone can kick off a coding task from one device and check on it from another. The update also added an "/agents" interface for delegating and tracking multiple coding tasks at once, voice input for the command line, and a Codex Security Cloud feature that scans repositories and new commits, flags and investigates issues, and prepares fixes rather than just reporting them.

The API is catching up to the demo

OpenAI also extended its Agents API to support computer use — letting an application operate software through its graphical interface the way a person would, rather than only through structured API calls. It separately announced a partnership with Amazon to run OpenAI-built agents inside AWS through a new Bedrock Managed Agent offering, a sign that agentic AI is being built for enterprise infrastructure, not just consumer chat apps.

Why this matters beyond the product announcements

Each of these updates is a small, concrete step in the same direction: from an assistant you talk to, toward a system that keeps working on tasks across multiple apps and sessions with less supervision. That shift is exactly what analysts have flagged as the next phase for enterprise AI adoption — not better chat answers, but agents trusted to execute multi-step work and report back.

What it means for how you work

• Developers: Coding agents that run in the cloud and persist across sessions change day-to-day workflow — tasks can be kicked off and checked later rather than babysat in real time, and reviewing agent-generated changes becomes its own skill worth building.
• Knowledge workers broadly: An "always-on" agent model means more of your job may involve defining tasks clearly and checking outputs, rather than doing every step yourself — a different skill set than using a chatbot for one-off questions.
• Anyone evaluating AI-related roles: Job descriptions mentioning "agent orchestration," "AI workflow design," or "agent oversight" are likely to become more common as more companies plug tools like Dots and Codex's cloud agents into actual business processes rather than pilots.

None of this means autonomous agents are reliable enough to run unsupervised on anything that matters yet — staged rollouts and enterprise admin controls suggest OpenAI itself is being cautious here. But the direction is clear, and it's worth tracking regardless of which specific vendor's agent your employer eventually adopts.

Source: OpenAI, DevDay 2026 Recap