# OpenAI’s DevDay Puts Dots at the Center of Work

- Date: 30 Sept 2026 (2026-09-30T16:06:09.000Z)
- Summary: OpenAI’s DevDay centered on dots, persistent assistants meant to carry tasks across workplace tools, shared Spaces and code. The launch makes delegation the product story, while live-demo friction leaves reliability and permission boundaries to be tested in practice.
- Tags: `digest`, `ai-discourse`, `openai`, `devday`, `dots`, `agents`, `coding-agents`

## Sources

1. [Simon Willison - OpenAI DevDay 2026 live blog](https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/) (website)
2. [OpenAI - Introducing GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/) (website)
3. [OpenAI - Introducing dots](https://openai.com/index/introducing-dots/) (website)
4. [OpenAI - DevDay 2026 Recap](https://openai.com/index/devday-2026-recap/) (website)

## Executive Summary

OpenAI’s clearest DevDay bet was not a single model upgrade but a persistent assistant that can take work across conversations, workplace tools and code. Its new **dots** product, paired with shared ChatGPT Spaces, makes the agent the interface and the model one component behind it. The promise is substantial; the live demonstration also showed why reliability, access control and verification remain the questions that matter. This is a delayed-discovery item: the event happened on September 29, but the substantive account was reviewed for this September 30 report rather than established from the initial listing notice.

## What Happened

At the keynote, OpenAI presented dots as named assistants able to work with ChatGPT, Slack and Teams, and to use Codex on a laptop to build an app. In [Simon Willison’s on-site live blog](https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/), the demo moves from delegating a task to a dot into a shared ChatGPT Space; he notes that OpenAI staff have been delegating to dots in Slack under their own identities. Pro and Enterprise customers were told they could access dots that day. OpenAI also described specialist dots for legal and finance and a Microsoft 365 collaboration. These are launch claims and demonstrations, not evidence that a dot will perform unattended work reliably in an arbitrary organization. One live handoff stalled with a “still checking” message.

The enabling stack was broad. OpenAI announced [GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/), pitched during the keynote as near-Astra intelligence at one-fifth of its price, and an Ultrafast mode described as up to 300 tokens per second at six times standard pricing. Willison also recorded an open-source Codex harness, Codex Cloud, computer use in an Agents API, and a security offering with scheduled scans and patch generation. The prices, speed and reliability figures are OpenAI’s claims as reported from the stage, not independent comparisons. OpenAI’s [dots announcement](https://openai.com/index/introducing-dots/) and [DevDay recap](https://openai.com/index/devday-2026-recap/) are the product reference points; the first-hand observations here come from Willison’s account.

## Why It Matters

The architecture points toward a different unit of adoption: not a person prompting one model, but a delegated task that moves through tools, shared documents, code and other people. ChatGPT Spaces supplies a common work surface; dots supplies the actor; Codex and computer use supply ways to execute. Sign in with ChatGPT, plugins and the announced marketplace extend that strategy into other developers’ applications. OpenAI appears to be competing for the place where work is assigned and checked, not merely for the best answer in a chat box.

That changes the practical test. A faster or cheaper model can improve the economics of an agent without proving that it has the right permissions, knows when to stop, or produces reviewable results. Willison reports that dots do not automatically inherit everything shared with ChatGPT, according to the closing Q&A, and that users can choose the responsibility they grant. Those boundaries are important design choices; how they hold up across Slack, shared Spaces and local computers will take actual use to assess. His same-day attempt to create a dot from the announcement link asked him to switch to desktop, a small reminder that the polished cross-surface vision and the available experience may differ.

## The Bigger Story

This reinforces a developing view of agents as workflow products rather than isolated model feats. The DevDay package links model cost and speed to delegation, collaboration, distribution and security. The strongest evidence today is that OpenAI is assembling those pieces and showing how it wants them to fit; it is not yet evidence of an organization-wide productivity gain. The live-blogged stalls and uneven demos make that distinction more useful than another capability ranking.

## Further Reading

- [Simon Willison’s OpenAI DevDay 2026 live blog](https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/) — detailed, timestamped observations of the keynote and subsequent sessions, including what failed live.