OpenAI casts Dots as a digital coworker at $100 a month

OpenAI introduced Dots as proactive assistants that can continue complex projects and everyday tasks while keeping the user in control. The primary OpenAI page was blocked during verification, so the details here are cautiously grounded in its public metadata, the DevDay presentation and reporting by The Verge.

Dots are designed to stay on course through long assignments

OpenAI says Dots run on GPT-6 Astra, with each individual agent called a Dot. The company demonstrated delegating QA for an unreleased product, website design, a team offsite, portfolio rebalancing and an email campaign. Interaction is intended to include text, voice and phone calls over time.

The service is starting as a premium product on the $100-a-month ChatGPT Pro plan and above. OpenAI attributes the price to heavy compute use. Specialist Dots for accounting and legal analysis, along with integration into Microsoft Agent 365, are still planned rather than generally available.

The price turns an assistant into a purchase of labor

OpenAI is selling Dots as a coworker for extended knowledge work, rather than another window for quick answers. For product and engineering teams, the central question is whether a Dot can retain context, request approval at the right time and deliver an output that can be inspected.

That puts Dots against Meta's Muse, which The Verge reports offers free access and a dedicated virtual machine. OpenAI is betting on a more expensive service for higher-value work and stricter handling of identity, payments and company data. That is a sensible distinction only if it survives routine use outside a keynote.

The voice demo failed and several channels are still pending

Voice demonstrations did not work at DevDay, and messaging the agent was still described as coming soon. Claims about staying on track over long periods also come primarily from the vendor. A staged demonstration does not reveal how much human intervention follows a bad instruction, a permissions change or a conflict between data sources.

Privacy will be as important as capability. OpenAI presented its Private Intelligence framework, stronger business controls and zero data retention options. For an agent touching email, calendars, documents and payments, the practical standard is an audit trail for every step and a chance to stop an action before execution.

Completed jobs without human rescue will settle the price argument

The first useful signal will come from long assignments outside OpenAI's scripts: how many a Dot completes correctly, how often it requests approval and how often a person must take over. The pace of access for cheaper plans and the actual arrival of voice, messaging and business connectors will matter too.

If a team gets reliably completed work with traceable steps for $100 a month, the price can be defensible. If it gets a cute character sitting on top of a long chat, free rivals have the easier pitch.

Lilith's verdict

A Dot is applying for the coworker seat that finishes an assignment overnight. At $100 a month, it needs to show up in the morning with the result and a readable trail of its steps, not another list of excuses.