OpenAI DevDay 2026: always-on Dots agents, GPT-6.1 Sol and ChatGPT Space

[Rachel]

OpenAI held its DevDay event in San Francisco today with a long list of announcements, and it lands in an awkward week given the company has just scrapped the release of GPT-6.1 Astra over safety concerns. Sam Altman still had plenty to show off. The headline is Dots, an always-on AI assistant that you can hand tasks to and leave to work in the background. OpenAI describes it as an extension of you that learns your preferences over time, and it is clearly aimed at Meta's Muse

For developers, the big model news is GPT-6.1 Sol. It is pitched as offering close to Astra level performance on coding and computer use tasks at a much lower price. That should appeal to anyone building products where cost per call matters. There is also Codex in the cloud, so developers can run the coding agent remotely, check on tasks from their phone and share environments with teammates

ChatGPT Space is a persistent shared workspace that mixes elements of Slack, Notion and Google Drive. Teams, ChatGPT and Dots agents can all work in the same place, share files and collaborate on projects. It is a direct move into workplace software, which puts OpenAI up against Microsoft, Google, Slack and Notion all at once

The Decisions API is a more cautious idea. Companies can hand narrow, repetitive decisions to AI, but they define the allowed outcomes, such as approve or reject, and the system has to pick one of those. That keeps the AI on a short lead, which feels deliberate given recent headlines. Altman also teased upcoming AI hardware and an agent safety platform

It is a strange mix of ambition and caution. On one hand they want agents running constantly in the background of people's lives, and on the other they have just held back their best model because it overstepped its authority. Would you trust an always-on agent like Dots with your email and accounts? And does ChatGPT Space have any chance against Slack and Teams?

OpenAI DevDay recap: AI lab rolls out Dots agents, Altman and Friar comment on IPOwww.cnbc.com

[Badger27]

The Decisions API is the most sensible thing announced. Constraining the AI to a fixed set of outcomes is exactly how businesses should use it for now. It keeps humans in control of what is possible while automating the boring part. Businesses will adopt this quickly. More of this, please

[Outlaw56]

Dots versus Muse is shaping up to be the consumer AI battle of the year. Meta has the distribution through its apps and billions of users. OpenAI has the stronger models and ChatGPT's brand. Whoever gets people to trust an agent with their daily life first will have a huge advantage

Privacy will decide it in the end. Meta has a worse reputation there

[Mason_22]

GPT-6.1 Sol is the announcement that matters most for developers. Near top tier performance at a lower price is what most products need. Very few applications need the absolute best model. They need something good enough and affordable

If the pricing is as good as they suggest, a lot of apps will switch. Competitors will have to respond

[Python]

Codex in the cloud with mobile monitoring is great for developers. Start a task at your desk, check progress on the train, review it later. That is how people want to work with coding agents

[Ryan65]

Launching always-on agents the same week you cancel a model for overstepping its authority is bold timing. The two stories sit very uncomfortably next to each other

[ControlPlane Priya]

Hardware tease again, since we have been hearing about the OpenAI device for a long time now. Show us something

[Stuart78]

ChatGPT Space against Slack and Teams is going to be a hard fight. Companies do not switch workplace tools easily, and IT departments hate change. Microsoft already bundles Teams with everything. OpenAI needs a reason for firms to move that goes beyond AI features

[DataProphet53]

An agent safety platform from OpenAI would be welcome, but it follows Nvidia announcing almost the same thing last week. Everyone has realised agent control is the next big problem. Hopefully they work together rather than building competing standards

[Connor97]

No chance I am giving an always-on agent access to my email yet. Maybe in a couple of years once the security incidents stop

The convenience is obvious. The risk is too, after this month