Teamday
← Back

OpenAI DevDay 2026: Dots, Always-On Agents in ChatGPT

OpenAI
agentsautomationenterprisefuture-of-workknowledge-workbusinessgptnews

Why OpenAI's Dots Turn ChatGPT into an Always-On Agent Workforce

OpenAI's DevDay 2026 keynote, opened by Sam Altman, leads with a product for delegation rather than a model. "Dots" are always-on agents in ChatGPT that handle work in the background and check in when something needs approval. The rest of the day covers the platform underneath: models, a cloud Codex, an agents API and a marketplace.

Dots as delegates, not chatbots: Altman says you can "delegate ambitious pieces of work the way you would to a high agency engineer or a chief of staff." The examples are standing responsibilities: monitor incoming bug reports and draft PRs, run the budget planning cycle, or get the app off a legacy API before it shuts down. Each dot has its own cloud computer and browser, follows the user's access, works across more than 4,000 connected apps, and can be reached by text or voice. Users can set boundaries on app and computer use and custom instructions for its actions.

Spaces for people and their agents: ChatGPT Spaces are shared pages where teammates and dots work together. You can tag a dot in a comment, give a page standing instructions such as updating findings from a Slack channel daily, and build charts, sheets and prototypes in place. OpenAI's own team says people began bringing their dots into group chats as delegates. Engineers' dots fix "dozens of bugs a day" from bug reports dropped into a channel.

Specialist dots for enterprise: In preview, companies can set up virtual teammates for accounting, marketing and legal. Teams "give them goals and contexts, review what they produce, give them feedback to improve, and that'll be shared across your whole company." OpenAI is working with Microsoft to manage them through Agent 365. Dots and Spaces ship to Pro, Business Premium and Enterprise plans, with dot conversations not counting against usage.

Speed and price as levers for agents: GPT-6.1 Sol offers near-Astra intelligence at about a fifth of the price, with cached input at a 95% discount, which makes long agent runs cheaper. "Ultrafast" gives eight times the speed at six times the price (about 300 tokens per second). A Decisions API lets a small model pick from predefined options (route, classify, choose the next step) in a fraction of a second. Pro 500 adds 25 times Plus usage, and Pro 200 reopens.

The build layer: The open source Codex harness now runs in the cloud, so tasks continue when a laptop closes and move between phone, browser and desktop. A Codex Security cloud finds vulnerabilities and prepares verified fixes. The agents API adds memory, multi-agent controls and computer use, and Bedrock managed agents bring OpenAI models to AWS. OpenAI also previewed private intelligence with zero data retention for sensitive workloads.

Distribution: "Sign in with ChatGPT" lets users spend their own plan tokens in third-party apps. Plugin extensions place full applications inside ChatGPT, with Figma and Adobe among launch partners. An OpenAI marketplace with 30-plus partners lets enterprises spend existing OpenAI commitments on partner products.

5 Takeaways from OpenAI DevDay 2026 for AI-Powered Work

  • Agents become a default feature - A per-user, always-on agent is now bundled into ChatGPT plans, which moves delegation from early adopters to the mainstream.
  • Identity, access and boundaries matter - Each dot has its own identity and computer, follows the user's permissions, and can be limited by the user. Governance is the product.
  • Specialist agents are the enterprise wedge - Goal-driven role agents for accounting, marketing and legal, with feedback shared company-wide, is the employee model applied to functions.
  • Cost per agent-hour is falling on purpose - Cheaper Sol, 95% cached-input discounts and the cloud harness all target long-running agent work.
  • The platform play is distribution - Sign in with ChatGPT, plugin extensions and the marketplace aim to make ChatGPT where agent products are found and bought.

What OpenAI's Dots Mean for AI Employees in Organizations

OpenAI is productizing the same idea TeamDay is built on: persistent agents with identity, shared context and human approval points, working alongside teams. That validates the category and raises the bar for specialists. The differentiation moves to domain depth, multi-agent coordination, and what a dot can actually be trusted to do. Altman closed by rejecting the industrial-revolution metaphor, arguing AI should give people "more power over their own lives" rather than make them cogs.