OpenAI dots could be the OpenClaw moment for non-geeks
OpenAI launched its personal AI assistant, dots, on 29 September at its DevDay event. It runs on GPT-6 Astra. Each dot has its own computer and continues to work between chats.
I think this could be the OpenClaw moment for people who have no interest in setting up agents.
OpenClaw is an open-source AI assistant that can manage your inbox, organise meetings, answer messages on WhatsApp and Telegram and can be installed and run on your own device with your chosen model. For a tinkerer, the setup is the fun part.
Someone trying to finish a proposal before lunch might be less keen on acquiring a small, unpaid position in IT.
The main difference with dots is that OpenAI provides the computer and browser, and there’s nothing to install. You still have to explain the job, connect apps and set permissions. But for people who don’t want to run an agent themselves, it removes a big barrier.
OpenAI didn’t buy OpenClaw, but it did hire the person who built it. Peter Steinberger joined OpenAI in February and now works on personal agents there. OpenClaw has since become an open-source project under a foundation that OpenAI supports.
Dots are part of a wave of new AI agents, from Grok Bot (11 August) to Meta’s US-launched Muse (8 September) and the invite-only Instinct, which is reportedly valued at $2.5 billion after just a few weeks. Instinct, which you text or call, was described by investor Sheel Mohnot as “OpenClaw for normal people”. That’s annoyingly close to my headline. I’m keeping it anyway.
They got there first. But OpenAI says ChatGPT is used by 1.2 billion people a week, and I’d expect more people to meet their first always-on agent there than anywhere else.
What dots do
A dot is designed to learn your preferences, remember useful information and manage many different projects. It can use your existing ChatGPT memories and carry context across ChatGPT, Slack and Microsoft Teams. It can be extended with a catalogue of plugins that OpenAI says connects to more than 4,000 apps (with your permission). You can watch its computer and redirect tasks if needed.
Apps in OpenAI’s plugin catalogue. The sphere has one dot for each of the first 4,000, and your dot only reaches the ones you connect.
Connected: 9. Click the sphere to connect another.
OpenAI’s launch post has an example from an early tester. His dot spotted that an invoice hadn’t been sent, created one for review and sent it with his approval. It’s one example, reported by OpenAI, but it’s the most exciting thing I’ve read about invoicing in years. That’s a sad sentence to have written.
For a non-technical business owner, the question is whether dots actually deliver. I’d create a dot, connect email and give it a simple job each morning, such as working through questions it hasn’t answered, compiling information on those customers and drafting answers for me to review, with rules on what should be answered first.
That’s a better test than asking it to “sort out the business”, an instruction that has defeated several generations of consultants.
Availability and cost
Pro users in the UK, EEA and Switzerland can’t yet access dots, with no timeframe given. Pro or Business Premium subscribers in eligible regions can create one at no additional cost, and Business Premium is available in all supported ChatGPT regions. Enterprise, Edu and Healthcare admins can enable dots as a beta. Dots will be rolled out over the next few days.
| ChatGPT plan | In the UK at launch |
|---|---|
| Pro | Not yetExcluded in the UK, EEA and Switzerland |
| Business Premium | YesAll supported ChatGPT regions |
| Enterprise, Edu and Healthcare | BetaOff until an admin turns it on |
More complex tasks come out of an allowance, with more included for the first month. Chatting with your dot won’t count towards ChatGPT limits, but actions kicked off in Codex and ChatGPT Work will. So always on doesn’t mean unlimited.
Permissions
When a dot works in the background without being asked, OpenAI calls it proactive research. It only uses read-only tools then, so it can’t message other people, change content in connected apps or control a browser.
You can add Custom Rules, such as never sending emails. Some actions, such as removing data permanently, can require confirmation. Others, such as changing passwords and moving money, are always handed back to you to finish. Auto-review checks that actions involving data sharing and account changes line up with your instructions and safety requirements.
OpenAI says dots may make mistakes. It’s nice to have something in common.
What happens when your dot tries to…
Proactive research uses read-only tools. It can’t send messages to other people, change content through plugins or control a browser or computer.
Based on OpenAI’s dots privacy, security and safety FAQs.
Memory
One gap is that you can’t yet view or delete the memories a dot creates. What it gathers from a connected app stays when you disconnect the app, a bit like telling someone a secret and then leaving the group chat. The only way to remove those memories is to delete the dot.
Five apps connected. Your dot forms memories from what it finds in them.
Whatever job you pick, check it regularly, note the fixes you have to make and see whether, after a week, you’ve saved or lost time. Only then expand its remit. If you’ve lost time, you’ve lost a week. I’ve had worse.
Getting a non-technical business owner through that first useful week would do more for the OpenClaw moment argument than another impressive demo.