OpenAI Dots: what founders need to know
OpenAI's always-on agents landed, the Pro plan got smaller and a US$500 tier arrived. Here's what it means for your product and what to do.
OpenAI shipped three things on 29 September. One new product and two price changes. Here's what matters if you build products.
The short version
- Dots are always-on agents. A Dot works in the background, across your apps, without you asking each time.
- The US$200 Pro plan got smaller. It now gets 10x Plus usage in Codex and ChatGPT Work. It used to get 20x.
- There's a new US$500 tier. It gets 25x usage and a faster mode in Codex.
If your team builds with Codex or ChatGPT every day, check your plan this week.
What a Dot is
A Dot is an agent that stays on. It has its own cloud computer and its own browser. It runs on GPT-6 Astra, OpenAI's top model.
It can plug into over 4,000 apps. You talk to it in ChatGPT, Slack or Teams. It sends you updates and asks questions as it goes. It learns from your feedback over time.
It also looks for work on its own. OpenAI calls this proactive research. But when it does that, it can only read your apps. It can't send a message or change a thing without you.
Who gets one
It's rolling out to "eligible markets" first. Check it's live where you are before you plan around it.
What changed on the plans
Codex and ChatGPT Work share one pool. Chatting with your Dot doesn't use it. Work your Dot does in Codex or ChatGPT Work does.
Ultrafast is only on Pro 500. It runs GPT-6 Astra up to 8x faster in Codex. It also burns through your allowance much faster, so don't leave it on by default.
What it means if you build products
Your AI budget just moved
Same price, half the room on Pro 200. If your devs live in Codex, they'll hit the limit sooner from 29 October. Plan for it now, not the day the work stops.
Agents are becoming users
Here's the thing. A Dot gets work done through app plugins. If your product has no plugin or API, a Dot can't use it.
New to agents? Start with what agentic AI means for a business.
Read first, act later
Dots can look around freely. They need a person before they act. Copy that if you build agents. Let them read. Make them ask before they change anything.
Not sure agents suit you yet? This guide helps you decide.
What to do this week
Start with the first box today. It takes two minutes.
Keep going
- More breakdowns like this one: AI resources.
- No API for an agent to use yet? Our dev studio, Devwiz, builds them.
Sources