Introduction

Open your inbox on a Monday morning and you will probably find a list of small jobs waiting. A client wants last week’s numbers. A lead needs a follow-up. A bug alert lands in Slack and someone has to decide whether it matters. None of this work is hard. It is just constant, and it usually falls on whoever has a spare minute.

That is the problem OpenAI, xAI and Meta are all trying to solve with a similar product. OpenAI has launched Dots inside ChatGPT. xAI has built Grok Bot. Meta has released Muse. Each one is a persistent AI Agent that keeps working after you close the chat window, uses its own computer, connects to your tools, and reports back for approval.

Why now? Chatbots are good at answering questions. Companies want software that finishes tasks. Once models can reliably use a browser, read files and call apps, the obvious next product is an assistant that owns a task for days or weeks instead of producing a single reply.

The Shared Pattern

Although each company has its own angle, the products share four parts:

  • A dedicated environment. Each agent gets a cloud computer or virtual machine, so it can browse, open files and run software without touching your own machine.
  • Persistent memory. It keeps track of preferences, projects and decisions across conversations.
  • Connections to apps. Email, calendars, messaging, documents and other tools can be linked so the agent can act on them.
  • Permission controls. The agent asks before sending messages, spending money or changing records.

OpenAI Dots

Dots were announced at OpenAI DevDay and run on the GPT-6 Astra model. Each Dot gets its own cloud computer and browser, and it can be reached from ChatGPT on mobile and web, and from Slack. OpenAI says it will expand to other messaging apps and to phone calls.

Advantages

  • It sits inside ChatGPT, which many people already use every day.
  • Custom Rules let you allow, require approval for, or block specific actions.
  • Work can be handed off in Slack, which cuts down on routine pings.

What it can do

  • Keep several projects moving at once and accept new work without forcing you into separate threads.
  • Watch for a bug alert in Slack and start investigating it.
  • Set recurring tasks, reminders and automations.
  • In one early test, notice a publication invoice that had not been sent, prepare it, and submit it after the tester approved.

What it cannot do (yet)

  • It is limited to eligible Pro, Business Premium and Enterprise accounts. Enterprise needs an administrator to turn it on, and Pro access is rolling out by region.
  • Its memory relies on saved notes about preferences and decisions, which are not a full transcript. Important details can be missed.
  • It is tied to the ChatGPT ecosystem, so it is not a tool you can set up independently of that platform.

xAI Grok Bot

Grok Bot is designed around teams of specialised Bots. Each one has its own role, memory and routines, and several Bots can share a group chat, pass work to one another and take ownership of tasks.

Advantages

  • It supports multi-Bot teamwork. Bots can message each other in group chats and assign work.
  • It has computer use. A Bot can operate a browser and terminal the way a person would.
  • It connects to structured tools through plugins and MCP servers, so it is not limited to xAI’s built-in features.
  • Routines let a Bot run on a schedule. Tools and skills are shared across the account, while memory and routines stay with each Bot.

What it can do

  • Browse the web, work with documents and send email.
  • Run recurring routines and carry work forward over time.
  • Coordinate with other Bots on a shared task.

What it cannot do (yet)

  • Usage runs on a weekly allowance that xAI has not published in size. Once it is exceeded, on-demand usage is billed by model and token cost, and there is no Grok Bot-specific spend cap yet.
  • xAI has warned that an hourly routine can use up a week’s allowance in a single day.
  • Model choice is tied to the subscription tier. Users cannot pick a specific model version for an individual task.
  • Long conversations are summarised automatically near the context limit. There is no manual fresh session or visible context meter.
  • Per-Bot memory is not fully documented, and reviewers have noted that users cannot easily audit or edit what a Bot remembers.
  • There are practical limits of about 50 Bots per account and six per group chat.

Meta Muse

Meta describes Muse as a personal AI agent that does the work rather than only answering questions. It runs on a dedicated Muse Secure VM that holds both the agent and the user’s data. People can talk to it in the Muse app or directly in WhatsApp.

Advantages

  • Muse is designed to feel like messaging another person, which lowers the learning curve.
  • It asks before important actions, such as sending an email or making a purchase.
  • Sentinel permissions distinguish read access from write access, so users can see what the agent is allowed to touch.
  • Users can set the agent’s name, avatar and communication style. If a connection is missing, Muse can build a tool for the task, according to Meta.

What it can do

  • Take on tasks and projects, and turn long-term goals into action plans.
  • Set reminders and suggest ideas based on what it knows about the user.
  • Work through everyday chores. Reports have described it ordering groceries, buying movie tickets, booking appointments and calling customer service on a user’s behalf.

What it cannot do (yet)

  • Meta built Muse for personal goals, so it is less suited to shared team workflows.
  • It launched in the United States on iOS, Android and the web, so availability is limited for now.
  • It needs broad access to personal email, calendar and purchasing. Press coverage has focused on how much trust that requires.

What to Weigh Before Choosing

The three products point in the same direction, but they differ in who they serve and how predictable they are. Before adopting any of them, it is worth checking:

  • Availability. Who can use it today, and in which regions?
  • Cost. Are usage limits published, and is there a spending cap?
  • Control. Can you approve actions, block them and audit what the agent did?
  • Memory and data. What does the agent remember, and can you edit it?
  • Team fit. Is it built for one person or for shared work?

How WorkLLM Helps

WorkLLM takes a different approach. It is built for repeatable team workflows, not a single personal assistant tied to one vendor’s model. You describe a workflow in plain English, and WorkLLM builds an AI Agent that runs it.

  • No setup required. A marketing lead can describe a weekly campaign report and receive a four-step AI Agent without writing code or building workflow logic.
  • Every step visible. Each step is shown and can be edited. Nothing runs until you approve it, and schedules run on the cadence you choose.
  • Shared organization memory. Company documents, decisions and past outputs are available to every AI Agent, so teammates do not need to re-explain the context.
  • 200+ models in one place. Use GPT, Claude, Gemini and other models in the same workspace, and compare them side by side.
  • Team collaboration. Share threads and AI Agents so the whole team can run and build on the same workflows.

The big tech companies are right that the next step is AI that finishes work. The useful question for your team is which AI Agents should run, what they can access, and who approves their actions. To see how that works in practice, visit workllm.io and describe the first workflow you would like off your plate.

Sign up to receive awesome content in your inbox.

We don’t spam! Read our privacy policy for more info.

Author Details

WorkLLM - Dhimant Bhundia
Co-founder & CEO at 

Product-focused founder with deep experience in AI, enterprise software, and data platforms. Passionate about turning complex workplace problems into simple, scalable products.

Would you like to share your thoughts?

Your email address will not be published. Required fields are marked *