Should a solopreneur let an always-on AI assistant work unsupervised?

Only inside limits you set and can check. OpenAI's new dots are described as assistants that keep working across projects while you stay in control, with boundaries on what they may do and when they must ask. Start with one low-stakes recurring task, review what the assistant did for a few weeks, and widen its permissions only as the record earns it.

Does cheaper AI mean I should move everything to the newest model?

No. OpenAI says GPT-6.1 Sol delivers near-Astra intelligence at one-fifth of Astra's standard prices. That makes routine work cheaper, but a lower price does not tell you whether the output is right for your business. Test it on a few tasks you can judge yourself before you switch anything.

This Week in AI: Assistants That Keep Working, and What You Hand Them

This week AI assistants became always-on, and the price of near-frontier models fell again. The cheaper and more autonomous the tool, the more the decisions about access and review fall to you. For a one-person business, the useful question is no longer which model is best. It is what you let an assistant do while you are not watching, what it can see, and how you would know it got something wrong.

OpenAI’s dots: assistants that keep working

OpenAI introduced dots, proactive assistants that keep working across complex projects and everyday tasks while you stay in control of what they do. Latent Space’s DevDay roundup adds detail: dots run on dedicated cloud computers, connect to more than 4,000 applications, let users set boundaries for autonomous actions and approvals, and are available on Pro, Business Premium and Enterprise plans.

An assistant that works while you sleep is the first version of the leverage solopreneurs have wanted. It is also the first version where a mistake can happen without you present. That is The Reliance Calibration Dial: trust should move up only as far as the assistant’s track record supports. Start dots on one recurring, low-stakes task. Read what it did. Widen the limits when the record is clean.

A frontier model at one-fifth of the price

OpenAI released GPT-6.1 Sol, which it describes as near-Astra intelligence for coding, computer use and professional work at one-fifth of Astra’s standard API prices. Latent Space puts the price at $2 and $10 per million input and output tokens. These are the vendor’s figures.

For you, this means routine work is now very cheap, and that is good. It does not mean the cheaper model is right for your work. Pick three tasks whose output you can judge, run them on the new model, and keep what holds up. The cheapest model that does the job well is the one to use, and only you can say what “well” means in your business. The earlier post on this fall’s price cuts makes the same point about who decides when something is done.

What you hand over: the data boundary

Two items this week are about what an AI system can see. Matthew Green’s argument, quoted by Simon Willison, is that AI agents kept in separate sandboxes can still pass something harmful to one another through anything they share, such as email or a chat channel. Anthropic’s Frontier Red Team reports that advanced cyber capability is now appearing in more than one lab’s models.

You do not need to follow the technical detail. The business rule is The Data Boundary: decide which accounts, files and customer information an assistant may touch, write it down, and give it nothing more. An always-on assistant connected to 4,000 apps is only as safe as the short list you actually connect. Revisit that list monthly.

Small teams are the target market

OpenAI announced a partnership with America’s SBDC to expand hands-on AI training and local support for small businesses, along with a new report on how small teams are using AI. Small business is now an explicit customer segment for the largest vendor, which means more training, more templates and more pressure to adopt. Free local training is worth a look. Adopt on your own timeline, with a single use case you can measure.

Google’s payments to publishers are small

Ars Technica reports that Google’s early attempt to pay websites for AI answers is struggling, with about 100 websites paid and the amounts described as tiny. If your business depends on search traffic to content you publish, the lesson is that AI answers may reduce visits and the compensation is not yet material. Own the relationship with your audience through an email list and a site you control, not a platform you rent.

The throughline

Tools that run on their own and cost less are good news only if you own the controls. The limits on an assistant, the list of data it can see, and the way you check its work are the parts of the system that compound. They stay yours when the model underneath changes next quarter. That is the argument behind marketing-grade decays, engineering-grade compounds.

More on building AI into a one-person business you actually own: curiochat.ai/solopreneur/.