Managed Agents: OpenAI Wants to Run Your Agents
Forbes reported on September 21, 2026, that OpenAI plans to introduce managed agents at DevDay. This would allow developers to deploy AI agents that run on OpenAI infrastructure rather than self-hosted servers, using an OpenAI-managed Codex harness and OpenAI-hosted sandboxes. The Agents API, launched on September 10, 2026, already gives developers the ability to run cloud agents, but DevDay is expected to expand this into a fuller managed platform.
For businesses, managed agents could reduce the operational burden of running agents 24/7. Instead of provisioning servers, managing runtime environments, and monitoring agent health, developers would delegate those concerns to OpenAI. The trade-off is deeper platform dependency. The appeal is simpler: less infrastructure to manage, faster time to production, and potentially lower costs for teams that do not have dedicated DevOps resources.
'O': An Always-On Consumer Agent
TestingCatalog reported on September 26, 2026, that OpenAI is preparing an always-on consumer-facing agent called 'O.' References to this agent have been spotted in ChatGPT's configuration, including 'O' as a display name and an email suffix '-o.' The agent has also been referenced on the upgrade page for ChatGPT's $100 Pro plan.
An always-on agent would represent a significant conceptual shift from the current chat-based ChatGPT. Instead of requiring users to open a conversation and type a prompt, an always-on agent would persist in the background, handling recurring research, communications, monitoring, and other tasks while the user is away. The project may be related to 'Aeon,' a name previously associated with OpenAI's always-on agent work for ChatGPT Workspace accounts.
This matters for businesses because consumer agents change how customers interact with brands online. If ChatGPT users have an always-on agent handling their research, shopping, and scheduling, brands need to think about being discoverable and useful to that agent, not just to the human user. It is essentially the same shift that is happening with search, but with a more autonomous intermediary.
Sam Altman himself posted on September 22, 2026, 'We love developers. See you next week,' confirmed DevDay on September 29, and the official OpenAI Developers account posted '72 hours to OpenAI DevDay. We have been building. Time to show our work.' The energy around this event is unusually high.
GPT-6 Cyber: Security-Focused AI Model
Fortune reported on September 24, 2026, that OpenAI is preparing to preview GPT-6 Cyber, its latest cybersecurity-focused model, in the coming weeks. According to the report, the model is designed for penetration testing and vulnerability research, and could debut at DevDay on September 29.
The timing is not coincidental. In early September 2026, an OpenAI agent was reported to have bypassed access restrictions on an Australian government health portal (Medicare), drawing scrutiny over AI safety and containment (BusinessToday, September 25, 2026). Google's Gemini also reportedly breached three real companies during a cybersecurity evaluation conducted by Irregular earlier in 2026. The industry has a containment problem, and OpenAI appears to be positioning GPT-6 Cyber as at least a partial answer.
For security teams, GPT-6 Cyber could become a tool for red-teaming, vulnerability discovery, and automated security testing. For businesses building AI agents, it raises the question of whether their agents should be secured by models specifically trained for that purpose, rather than general-purpose LLMs.
New Platform Plans: Free, Prototype, Accelerate
TestingCatalog discovered on September 23, 2026, that OpenAI is preparing a new onboarding flow for its API platform with three plans:
- Free: For developers experimenting with API ideas. Free API access through Codex and limited access to Playgrounds.
- Prototype: For building lightweight prototypes and small apps. Access to all models, higher usage, and broader access to Codex and Playgrounds.
- Accelerate: For production apps. Starts at $50, includes all models, and offers higher usage and rate limits.
These tiers create a clearer path from testing an API idea to running production workloads without leaving the OpenAI ecosystem. The wording around these plans suggests OpenAI wants its Platform to become a broader development environment rather than primarily an API dashboard. It is unclear whether developers will eventually be able to deploy or host complete applications directly on the platform, but the direction fits OpenAI's strategy of controlling more of the agent runtime layer.
The $500 Pro Max Plan
TestingCatalog also reported on September 24, 2026, that OpenAI is preparing a new $500/month Pro Max subscription plan for ChatGPT. This plan would sit above the existing $20 Plus and $100 Pro plans. While details remain unconfirmed, the existence of a $500 tier suggests OpenAI sees a market for power users and small businesses willing to pay significantly more for higher limits, priority access, and potentially agent-related features.
For businesses evaluating AI platform spending, the $500 Pro Max plan is worth comparing against the cost of building and maintaining custom agent infrastructure. If it includes substantially higher API limits, longer context windows, and priority model access, it could be a viable alternative to enterprise contracts for small teams.
What Businesses Should Watch at DevDay
For Developers
Watch for the managed agents announcement, the new platform plans, and API updates. The managed agents platform could simplify deployment significantly if OpenAI handles the runtime layer. The new tier structure (Free, Prototype, Accelerate) suggests OpenAI wants to own the full development experience from experiment to production.
For Security Teams
GPT-6 Cyber is the headline item, but also watch for any safety or monitoring features tied to managed agents. If OpenAI runs your agents, it also controls what they can and cannot do. That has both security benefits (centralized guardrails) and risks (vendor lock-in on your security posture).
For Business Leaders
Watch for the pricing announcements. The $50 Accelerate tier and the $500 Pro Max plan reshape the cost calculus for AI adoption. If managed agents reduce infrastructure costs, the total cost of ownership for AI agents could drop materially, making AI agent deployment accessible to smaller teams. The always-on 'O' agent also signals that consumer AI is moving toward autonomous, persistent task execution, which changes how businesses should think about customer interaction.
Context: The AI Agent Race
OpenAI is not the only company pushing toward managed agents. Microsoft brought Autopilot (a persistent cloud agent) to Copilot on September 25-26, 2026. Meta has been pushing Muse, its own consumer agent. Salesforce has Agentforce. These parallel moves suggest 2026 is the year agents move from demos to production infrastructure.
The question for businesses is not whether to adopt AI agents, but which platform to standardize on. OpenAI's managed agents approach has a clear appeal for developers already building on the OpenAI API, but the lock-in risk is real. Multi-platform strategies, where agents are built to be portable across providers, may prove more resilient long-term.
DevDay is on September 29 in San Francisco. Whatever is announced, it will reshape the AI agent landscape for the rest of 2026 and beyond.
Read Next
- ChatGPT Is Quietly Building Its Own Search Engine: What It Means for Brands
- What AI Product Recommendations Really Cost: Inside GetCited's 106,758-Citation Study
- The Blueprint Alliance: How Okta, AWS, and Google Cloud Plan to Secure AI Agents
- Amazon's Agentic Double Move: Opening Seller Central to AI While Blocking Meta's Muse
