OpenAI Dots and Grok Bot Are Here. Seepient Made a Different Bet.

Two of the loudest AI labs shipped the same idea within a week of each other: an AI that keeps working after you close the laptop. OpenAI announced dots at its Dev Day on September 29. xAI launched Grok Bot a few days before that. Both run in the cloud, both learn how you work, both promise to hand you finished jobs instead of chat replies.

I've spent this year building Seepient, which shares the same starting point but makes a different bet about what this thing should be. With both launches in the wild, it felt like the right moment to put the three side by side: what they can do, where they agree, and where Seepient deliberately walks the other way.

What OpenAI dots actually is

Dots are always-on agents that live inside ChatGPT, powered by OpenAI's GPT-6 Astra model. Each dot gets its own cloud computer and browser. It keeps working between your conversations, connects to more than 4,000 apps through OpenAI's plugin ecosystem, and reaches you wherever you are: ChatGPT on desktop and mobile, Slack, Teams, with texting on the way.

The pitch is "remarkably capable, always-on agents built to handle everything." You start with one primary dot, give it a name, and make it your own. OpenAI's examples read like job descriptions: a dot that watches customer feedback and hands you finished code fixes to review, a dot that notices you forgot to invoice a client and prepares the invoice, a dot that reruns your analysis when new data lands.

The safety story is worth noting because it's more detailed than most. When a dot works in the background on its own, it can only look at your connected apps, not touch them. Actions that could affect your accounts get checked against your instructions before they run, and some things, like changing a password, always stay with you. Dots are rolling out to ChatGPT Pro and Business Premium subscribers in eligible markets first, with the first dot included at no extra cost.

OpenAI also previewed "specialist dots" for organizations: agents with their own work identity and credentials, managed through existing IT tools, built with Microsoft's Agent 365 governance controls.

What Grok Bot actually is

Grok Bot is xAI's version, and it leans harder into the teammate framing. Bots get their own computer in the cloud, sign into the tools you already use, and work across apps, inboxes, and websites, including ones with no clean programming interface. They finish jobs end to end and come back only when something needs your approval.

Three things stand out to me about how xAI positioned it.

• You teach by showing. The recommended way to train a Bot is to let it follow along the next time you do a job. It watches the steps, saves the workflow, and runs it solo next time. That's a genuinely different onboarding model from writing instructions.
• Bots talk to each other. You can run several in parallel with a "chief of staff" Bot managing specialists, or drop them in a group chat where they assign work themselves and only pull you in for judgment calls.
• It ships as its own app. Grok Bot is in beta for SuperGrok and Cursor subscribers, on desktop and iOS, with usage billed separately from your existing plans.

The xAI launch page quotes employees describing it as "bringing on a coworker" and handing work off "the way you would text someone on your team."

The pattern all three share

Put the launches next to each other and the industry consensus is obvious. Everyone converged on the same four ideas.

Every one of these products has its own machine to work on. Every one claims to learn your preferences and get better over time. Every one works in the background on goals rather than answering one prompt at a time. And every one makes approval gates the trust story: the agent proposes, you approve.

That last point matters most if you run a business. The interesting question about these launches isn't whether the AI is smart. It's whether you can hand it your invoicing, your inbox, or your customer follow-ups and sleep at night. All three companies clearly understand that, which is why the approval language is everywhere in their announcements.

So on capability, Seepient and I agree with the direction. Here's where I went a different way.

Where Seepient differs: you raise it, you don't configure it

Seepient is a personal AI companion you name, teach, and grow with. It ships as a terminal app, an SDK, and a server, but those are the mechanics. The working line is "Bring your Seepient alive. Let it grow with you."

The differences from dots and Grok Bot are philosophical, not feature gaps.

• One companion, not a team. Grok Bot's model is a fleet: a chief of staff over specialists, a Bot per lane. OpenAI is heading the same way with teams of dots and specialist dots. Seepient bets the opposite way. The same Seepient can learn code, marketing, accounts, personal assistance, or something nobody has tried yet. Capability lives in the teaching, not in hiring a new agent per job. You never need more than one, though a family is there for people who want it.
• Identity comes from the relationship, not the roster. This is the part I feel most strongly about. Two Seepients raised by two people end up unlike each other, the way two apprentices of the same trade do. You can name a dot, as OpenAI's launch video does with one called Alfred, but the personality behind the name is the same model with the same defaults. A Seepient differs by everything you've taught it. Its appearance starts as an unformed glow and grows into its own look as it learns, so your companion eventually looks like your companion.
• It runs where you run. A dot lives in OpenAI's cloud and a Bot lives in xAI's. A Seepient runs on your machine, with your model of choice behind it. That's the multi-model design I wrote about in Seepient's provider control room: when a provider fails or changes pricing, you swap the engine and your companion keeps its memory, its teaching, and its personality. Your companion isn't someone else's hosted service.
• Trust is built into the species definition. Seepient proposes, you guide, approve, revise, or stop. Sensitive areas like money and home automation only ever appear with their exact tools and limits documented. I wrote a full breakdown of how the permission system works under the hood, and the short version matches what dots and Grok Bot are now saying publicly: an agent that acts without a human approval gate isn't a product, it's a liability.

One more honesty note, because the comparisons invite it. Seepient's "species" language is a metaphor for personalization and relationship, not a claim that the software is conscious. The site says so plainly. I'd rather keep that line clear than let the marketing drift somewhere the product can't follow.

The comparison at a glance

What I'd tell a business owner this month

If you're on a ChatGPT Pro plan and drowning in follow-up work, try dots. If you want several workers running in parallel today and you're comfortable in a beta, look at Grok Bot. Both are real products from teams that ship fast, and the approval controls are more thought through than the early coverage gives them credit for.

But notice what you're trading. Your teaching, your preferences, and the accumulated context that makes these agents useful live on the provider's servers, priced on the provider's terms, shaped by the provider's roadmap. That's the same vendor lock-in problem I keep meeting in client work, just wearing a friendlier face. If that concerns you, read why I gave Seepient a provider control room before you commit your workflow to any single cloud.

The always-on agent era is here, and it arrived with two big logos on it. I think that validates the two years of quiet building behind Seepient, and I also think the companion bet, one agent that truly becomes yours, is the version of this future worth owning. If you want help thinking through where an agent like this fits in your own operation, book a call and we'll map your workflows before you hand any of them over.