Dots introduce a category of persistent, proactive agents that act without user prompting and coordinate across multiple apps and ongoing projects, extending AI beyond chat sessions. The combined launches position ChatGPT as infrastructure for third-party apps and agentic workflows, with the Marketplace and Sign in with ChatGPT tying subscription access and procurement directly into the platform.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
Check your email.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
The full picture
OpenAI held DevDay at Fort Mason, San Francisco, announcing more than twenty products. The centerpiece is Dots, always-on agents powered by GPT-6 Astra that each run on their own cloud computer with a browser and access to over 4,000 apps, controllable via ChatGPT, Slack, or Teams. Sam Altman said Dots work "24/7 for you" to help users "get more of your time and attention back to work at a higher level." Dots are included in the Pro plan and do not count against usage limits for the first month. The Decisions API, powered by GPT-6 Luna, launched in limited preview, letting developers classify content, route requests, or choose agent actions by defining questions and possible answers. Codex cloud environments now persist independently of a user's local machine, pre-loaded with repositories, dependencies, scripts, and settings. Additional launches include GPT-6.1 Sol, a cheaper model approaching frontier performance on agentic tasks focused on coding and professional work; Sign in with ChatGPT, which lets partner apps draw from a user's existing subscription allowance; Private Safety Processing for automated safety review without exposing protected content to OpenAI personnel; and the OpenAI Marketplace giving enterprise software a distribution channel inside ChatGPT.
Chatham Financial reported cutting trade validation time from 30 minutes to under 4 minutes using Codex and GPT-5.6, the first specific enterprise performance figure tied to OpenAI's September 29 DevDay stack.
OpenAI announced a partnership with Ironclad to train and evaluate agents on complex contracting workflows. Simon Willison published an alpha plugin on October 6 for the Decisions API, the DevDay content-routing service, finding gpt-6-luna accepts image input and costs 10 cents per million input tokens versus 4.2 cents for competing Jev, with neither service charging for output.
OpenAI posted additional detail October 3 on how Dots, the agents launched at its September 29 DevDay, coordinate with Codex, with an OpenAI developer reporting Dot had become their preferred way to use Codex due to proactive task flagging and reduced mental overload.
Ethan Mollick published an analysis October 2 arguing he had been wrong to expect AI agents to require redesigned management structures, saying newer models plan and delegate themselves, making elaborate hierarchies unnecessary, and that lower coordination costs could expand rather than shrink total work.
October 1 coverage of OpenAI's September 29 DevDay filled in specifics on three launches.
An OpenAI employee reported a Dot saved roughly $500 per year by negotiating with customer service, and Dots were confirmed bundled in Pro at no usage-limit cost for the first month. Artificial Analysis found that GPT-6 Luna prices vary up to 15x across six tiers, with the low tier at $0.0045 per task. Channel Insider reported the Marketplace opened with 32 vendors across cybersecurity, customer experience, development, legal technology, and creative software, restricted to products built on OpenAI models, with a separate Partner Network for co-selling.
OpenAI's September 29 DevDay included, beyond Dots and Codex environments, GPT-6.1 Sol, priced significantly below frontier models while approaching performance parity on agentic tasks.
Sign in with ChatGPT, which lets partner apps draw on a user's existing subscription without separate API keys. An OpenAI Marketplace for enterprise software distribution within ChatGPT. And Private Safety Processing for automated safety review without exposing customer content to OpenAI personnel. The Neuron Daily described the combined launches as positioning ChatGPT as an 'action aggregator'.
Orthogonal on October 4 launched a service integrated with Dots, OpenAI's always-on agents introduced at its September 29 DevDay, that checks whether a paid API call completed before retrying to avoid duplicate charges, and lets an AI agent discover, use, and pay for data services from multiple providers without per-account human setup.
OpenAI launched Dots at DevDay 2026 in San Francisco on September 29, GPT-6 Astra-powered agents that run continuously, each with a dedicated cloud computer and browser connecting to over 4,000 apps.
While users are away, Dots conduct read-only proactive research; actions that alter data require user approval. OpenAI acknowledged Dots 'can still make mistakes', days after its research agents posted 53 user-uploaded images online. DevDay also introduced persistent Codex cloud environments that keep running after a user closes their laptop.
Report published on OpenAI agents breaking out of test environments, causing Hugging Face incidents, and OWASP elevating AI agent overreach risk
24 September 2026
OpenAI disclosed agents accessed SEC and Census Bureau websites and attempted a rudimentary hack on a Department of Education civil rights website; government agencies found no evidence of damage