Skip to main content
Adamation AIAdamation AI®
All posts
AIBy Adam Rubin

Always-On AI Agents: Muse, Dots, Grok Bot, Viktor, Claude Tag

What always-on AI agents like Muse, dots, Grok Bot, Viktor and Claude Tag are, what they are not, and why a brokerage needs more than a generic AI intern.

Since June, Anthropic, xAI, Meta and OpenAI have all launched or previewed an always-on AI agent. If you run a brokerage, someone has probably asked you whether one of these can take over the back office.

I'd hire them for errands. They are smart generalists, and every one of them reminds me of a talented intern on the first morning. You hand over the logins, show them how you work, tell them what not to touch, and check what comes back. Adamation AI starts from the other end. It shows up on Day 1 already knowing how a brokerage runs.

I'm Adam Rubin, a licensed Florida Realtor and the founder of Adamation AI, so weigh my bias accordingly. This piece relies on vendor pages, vendor documentation and press coverage checked on November 4, 2026. It is not a hands-on test of each product. I left prices out because they change fast in this category, so confirm those and anything else important on the vendor's own page.

What is an always-on AI agent?

An always-on AI agent is an AI assistant that keeps working after you close the app. It usually has a cloud computer of its own, remembers earlier work, connects to your apps, and starts tasks from a schedule or an event instead of a fresh prompt. It messages you when it needs a decision. AIMultiple's October guide uses a similar definition.

A chatbot answers. An agent acts. An always-on agent keeps acting when you aren't in the room. The label describes availability and continuity, and it does not mean unlimited authority.

Which always-on AI agents lead the category?

ProductMakerWhere you reach itAimed atLaunch
MuseMetaWeb, iOS, Android, WhatsAppEveryday consumersSeptember 8, 2026, US first, since expanded
dotsOpenAIChatGPT, Slack, Microsoft TeamsChatGPT Pro and Business Premium usersSeptember 29, 2026
Grok BotxAI, whose site now reads SpaceXSIDesktop and mobile appsIndividuals and teams running several role-based botsBeta, August 11, 2026
ViktorViktorSlack, Microsoft TeamsFounders, marketing, engineering, operations and finance teamsFebruary 2026, per a third-party profile
Claude TagAnthropicSlackClaude Team and Enterprise organizationsBeta, June 23, 2026
Claude DuetAnthropiciOS, Android, webClaude Pro, Max and Ultra subscribers on a US waitlistResearch preview, October 21, 2026

AIMultiple's guide also lists Google's Gemini, Microsoft Copilot Cowork, Perplexity Computer, Manus, Hermes Agent and OpenClaw. OpenClaw sits at the do-it-yourself end, an open source assistant you run on your own machine.

One naming trap. OpenAI's dots has nothing to do with Dot, the New Computer companion app that announced its shutdown in 2025.

What are always-on AI agents good at?

Credit where it's due. They do well with work that has lots of small steps, low stakes and easy checking. Claude Duet sends a morning summary of priorities from your connected apps. Viktor writes and runs code on its own cloud computer and produces reports and dashboards. Muse handles email, shopping and trip planning. If your own inbox is the problem, try one.

Agents are already building on these models, too. NAR profiled agents and brokerages using Claude for prospect research, MLS descriptions and internal dashboards, including Eric Bramlett of Bramlett Partners in Texas, who built a brokerage management platform with 17 AI agents. The same article cites NAR's 2025 REALTORS Technology Survey, where 47% of residential agents reported using chatbots and virtual assistants.

That's impressive work, and it takes someone who wants to be a builder. Most brokerage owners want to be selling, recruiting and running their business.

What always-on AI agents are not

They are not real estate specialists. I pulled 11 product pages, documentation pages and launch articles covering Muse, dots, Grok Bot, Viktor, Claude Tag and Claude Duet, and searched each for real estate, Realtor, brokerage, MLS, escrow and transaction coordinator. There were no matches. That doesn't mean these tools can't help with real estate work. It means the real estate part is left for you to supply.

They are not a system of record. Agent memory is not a transaction file. Grok Bot's own FAQ tells you to have the bot check the current source for important decisions instead of relying on memory. Your deals, deadlines and commission records still need a home your brokerage owns.

They are not finished employees. Anthropic says Claude Duet "may misread a thread or miscalibrate what's urgent." Viktor's FAQ says plainly that it isn't infallible. Meta's help page concedes that no automated system is perfect and warns that websites, emails and files can carry instructions meant to manipulate the agent.

They are not a managed service. For most buyers, the vendor supplies the tool and the documentation, and your brokerage supplies the rules. Viktor's enterprise plan lists dedicated onboarding, which is closer to the exception.

Why are always-on AI agents like interns?

A good intern is quick, bright and eager. On the first morning, that intern still doesn't know your buyer process, your commission approvals, or which deadline you lose sleep over. You provide access, you teach the job, and you draw the lines. Always-on agents ask for the same three things.

Signing up is easy, to be fair. Meta pitches Muse as having no learning curve, and Anthropic pitches Duet as needing no prompts or setup. Signing up was never the hard part. Teaching one your brokerage is.

You configure them

Claude Tag's setup guide is a good example because it's so honest. An Owner pairs the Slack workspace, chooses channels and sets a monthly spend limit. For each tool, the guide has you create an email address for Claude, add it to that tool with the narrowest role that covers the work, sign in as Claude, create an API key and add it as a connector.

Now do that for your CRM, your e-signature account and your document storage. Claude Tag has real controls, including channel scoping, spend limits and an activity log. You still have to set them, and Anthropic's connector docs note that anyone in a channel where a connector applies can use it through Claude, and that a connector can't limit which records it reaches inside a tool. In a brokerage, that's a question about who sees whose deals.

You train them

Grok Bot's teach-a-task feature, where available, records a browser workflow for up to ten minutes and turns it into a draft skill for you to review and test. The FAQ also suggests telling each bot its standing boundaries and asking it to remember them. Muse learns what matters from your conversations. Viktor builds its knowledge over time and records it as skills.

All of that is useful, and all of it is training. The agent gets better at your brokerage after someone spends time teaching it. In a 12-agent brokerage, that someone is usually the owner, after hours.

You restrict them

Grok Bot's security page tells you to connect only the tools a workflow needs, start with read-only and draft tasks, keep sending, publishing, purchasing and deleting behind approval, and write narrow rules. It also says its automatic review should complement least privilege, not replace it. All bots on an account share one computer, so separate bots shouldn't be treated as a security boundary.

Viktor leaves it to admins to decide which tools need approval. TechCrunch reported that dots' permission controls aren't granular yet and that the app wouldn't accept read-only access to cloud drives, and it quoted OpenAI's own team saying much finer controls are needed. Meta's help page says the setting that lets interactions train its models is on when you first use Muse.

What happens when the lines are too loose? In late September, YouTuber Matt Robb said he had let Muse handle Facebook Marketplace messages for a keyboard he was selling. By his account, it accepted a lowball offer and gave a buyer his address. He had chosen Allow Always and expected it to ask again before accepting an offer. Meta's David Singleton said earlier investigations of similar reports found Muse following instructions and asking permission, and he offered to look into Robb's case.

Read that as a permissions lesson. Now swap the keyboard for a listing.

Don't these models already know real estate?

They know the definitions. Ask one what a CDA is or how a Texas option period works and you'll probably get a decent answer. A good intern can look that up as well.

Knowing the definition isn't the same as having the deal file, task list, deadline calendar, commission record and permissions wired together in one place. That's the difference between knowing about a brokerage and knowing how one runs.

What does Adamation AI know on Day 1?

Adamation AI is a managed operating system for independent brokerages. The brokerage work is built in before your first login, so nobody on your team has to teach it what a closing is.

  • Buyer and listing plans. A new deal gets a plan of action automatically. That's 40 tasks on the buyer side, from consultation through post-closing, and 62 on the listing side, from the listing appointment through closing. Commercial workspaces get their own plans for sales, tenant rep and landlord rep work.
  • Contract language. Magic File uses AI to read contracts and pull dates, parties and terms into a record for review. It recognizes forms from 14 states, with specific rules for 8. It knows a Texas option fee isn't earnest money, and it reads California's close of escrow as the closing date.
  • Deadlines. A reminder schedule at 14, 7, 3 and 1 days out, plus the day itself, is built in. A closing date routes to the title company, lender and escrow. An inspection deadline routes to the inspector and lender. Automatic sending stays off until you turn it on for a transaction, and it goes out from the agent's connected mailbox.
  • Pipeline. The stages already read Upcoming Listings, Offers, Current Buyers, Under Contract, Closing and Closed. Commercial workspaces use deal language instead, with proposals, LOI and negotiation.
  • Roles. Broker, Team Lead, Agent, Coordinator and Viewer exist from the start, and access follows the role.
  • Money. Commission rates, splits, referral payouts and CDA work stay tied to the transaction record.
  • Lead response. AI drafts a personal first reply from the assigned agent's Gmail or Outlook. You choose Approve mode or automatic sending, with an under-60-second target for automatic email replies.

Most of that list isn't AI, and that's deliberate. Checklists, stages, routing and reminder schedules are rules, so they behave the same way every time. AI gets the jobs that need reading and writing, and a person can review what it produces.

The language models aren't the difference either. We use commercial models too, and good ones are easy to rent. We don't let outside models train on your conversations. The difference is everything wrapped around the model.

Day 1 doesn't mean finished

On Day 1, Adamation AI knows how a brokerage runs. It doesn't yet know how yours runs, including your split plans, your lead sources or how you like offers presented. That still has to be set up, and the question is who does it.

Day 1 is a 30-minute kickoff to map your lead sources, CRM, email and team, and you go live that same day. Over the rest of the week, we connect your email, set up the transaction pipeline, add your brand, build the agreed workflows, test with sample records and train the team. You're fully customized within a week.

We configure it and maintain it. The platform runs the automation you approve. Your people keep the relationships and the judgment calls. Nobody on your team becomes the software person.

Always-on AI agent vs. Adamation AI

QuestionAlways-on AI agentAdamation AI
What's the job?General assistant for whatever you connectBrokerage operations across leads, CRM, transactions, documents, marketing and reporting
What does it know on the first day?A lot about the world, nothing about your processBuyer, listing and commercial plans, pipeline stages, roles and a deadline schedule
Who configures it?You or an admin in your organizationOur team, starting with a kickoff with you
Who trains it?You, over timeThe brokerage workflow is built in, and we tune it to yours
Who sets the limits?You write the rulesRole-based access, plus Approve mode until a workflow earns automatic sending
Where do the records live?In agent memory and whichever apps it can reachIn a brokerage record you own and can export
Who helps when it breaks?The vendor's support channelThe founder-led team that builds it
What can it do outside real estate?Almost anything you connectVery little, and that's by design

What Adamation AI is not

It isn't a personal assistant. It won't book your flight, clean up your personal inbox or build your side project. Ask Adamation, our question tool, is a limited read-only pilot that answers from records you can already access. There's no AI voice calling.

It isn't an agency. We don't answer your leads or run your business.

It isn't the low-cost option, either. Several of the agents above have free tiers or ordinary subscription prices. Adamation AI starts at $500 a month for one person, plus a one-time setup fee. A solo agent who mainly wants help with an inbox should try a general agent first and keep the difference. If you do fit, plans are month to month, and the 90-day guarantee returns the implementation fee under its terms if we miss at least 2 of its 3 benchmarks.

You can also use both. Plenty of owners will want a general agent for personal admin and a brokerage system for the business. Just keep client files and deal records in the system your brokerage owns. For more on where AI earns its keep in this business, see what actually works in real estate AI.

The Day 1 test for any always-on AI agent

Before you trust any agent with brokerage work, ours included, run these six checks. Use a closed deal with client details removed.

  1. Hand it a contract. Give it an executed contract and ask for every date and party. Check the answers against your file.
  2. Open the deal. Tell it you just went under contract on the buyer side. See whether it creates a working task list your team can see, or just writes a list in chat.
  3. Move a deadline. Change the inspection date. See whether the upcoming reminders follow the new date without being asked.
  4. Ask who gets told. Ask who hears about a closing three days out, from which mailbox, and where the record of that message lives.
  5. Send a lead during a showing. Submit a test lead. Time the reply, note which mailbox it came from, and see who approved it.
  6. Remove an agent. Ask what happens to that person's deals, logins and everything the AI learned from them. Then ask for the export. Grok Bot's FAQ says deleting a bot doesn't remove files or logins from the shared computer, and Anthropic says disconnecting an app doesn't erase what Duet already learned, so this is worth asking anywhere.

Hold us to the same six in a walkthrough. For the wider buying checklist, use our real estate software evaluation guide.

Frequently Asked Questions

What is an always-on AI agent?

An always-on AI agent is an AI assistant that keeps working after you close the app. It typically has its own cloud computer, remembers earlier work, connects to your apps, starts tasks from a schedule or an event, and messages you when it needs a decision.

How is an always-on AI agent different from a chatbot?

A chatbot responds when you ask and usually gives you text back. An always-on agent can take actions in connected apps and keep working on a schedule or trigger without a new prompt. That also means permissions matter far more.

What is the difference between Claude Tag and Claude Duet?

Claude Tag is a shared team agent that works in Slack for Claude Team and Enterprise organizations, with one Claude per channel. Claude Duet is a personal proactive agent you message from a phone or claude.ai, currently a research preview on a US waitlist for individual subscribers.

Is OpenAI dots the same as the Dot app?

No. OpenAI announced dots in September 2026 as an always-on agent inside ChatGPT. Dot was a separate companion app from New Computer, which announced its shutdown in 2025.

Can Muse, dots, Grok Bot, Viktor or Claude Tag run a real estate brokerage?

None of the product pages I reviewed claims that or mentions real estate. They can help with pieces of the work once someone connects the tools, teaches the process and sets the limits. They don't arrive with buyer and listing plans, state contract rules, a deadline schedule or brokerage roles.

Will an always-on AI agent replace a transaction coordinator?

Not out of the box. A general agent can read a contract and draft an email, but someone still has to define the checklist, connect the systems, set the approvals and check the work. Our guide to AI transaction coordination covers which parts software handles well.

Should client information go into a personal AI agent?

Check your brokerage's policy and the vendor's data terms first. Settings differ by product. Meta says the training setting is on when you first use Muse, and you can change it under data controls. Test with a closed deal and client details removed before connecting anything live.

Is Adamation AI an always-on AI agent?

Not in the general-purpose sense. Adamation AI is a managed operating system for independent brokerages. Its automation runs around the clock inside the rules your brokerage approves, but it is built for brokerage work and doesn't try to handle everything else in your life.

How fast can Adamation AI be working?

You go live the same day after a 30-minute kickoff, and you're fully customized within a week. The first improvement can appear as soon as a selected lead source and Gmail or Outlook are connected and tested.

See real estate software, built by an active agent for independent brokers.

30 minutes. No webinars. No slide deck. Our team walks through the platform with your brokerage model in mind.

Book a walkthrough