--- title: "Which AI Agent for What? Muse, Dots, Duet, Grokbot, Instinct" url: "https://goremy.ai/blog/articles/ai-agents-website-access-muse-dots-duet-comparison" author: Luis Chavez-Mattos editor: Luis Chavez-Mattos editor_title: Director, Product published: "2026-11-03T06:00:09.989Z" updated: "2026-11-03T06:00:09.989Z" topic: Always-On Background AI Agents description: "Compare personal AI agents Dots, Muse, Duet, Grokbot and Instinct by task: coding, shopping, travel, proactive alerts, plus website access limits." --- # Which AI Agent for What? Muse, Dots, Duet, Grokbot, Instinct > Compare personal AI agents Dots, Muse, Duet, Grokbot and Instinct by task: coding, shopping, travel, proactive alerts, plus website access limits. Edited by Luis Chavez-Mattos, Director, Product ยท Published 2026-11-03 ## Which personal AI agent should you use for coding, shopping, travel or reminders? It depends on the job. In one power user's real setup, Grokbot takes hard coding and build work, Dots handles long-running projects across calendar and email, Instinct covers travel, Muse covers shopping and bookings, and Duet (from Anthropic) stands out for proactive alerts. Dots and Duet are the most proactive. Muse is the most consumer-friendly. Most people should pick one agent, and no clear winner exists yet. ## TL;DR - **Grokbot suits hard coding and build jobs** because its bots can talk to each other, so you don't relay discoveries between workers. - **Dots is the long-memory option.** It keeps project context, decisions and changes, and works across calendars and email, though it isn't especially fast. - **Instinct is built around travel**, with easy texting on the go, autonomous check-in and deep destination research. - **Muse is the friendliest for non-technical users**, covering gift shopping, restaurant and class bookings, and purchases through its Stripe link partnership. - **Duet and Dots are the most proactive.** Their unprompted alerts are useful maybe 50 to 75 percent of the time, according to the person who tested them. - **A smarter agent can't open a locked door.** Websites decide who gets in, so access is uneven no matter which agent you choose. ## How was this comparison put together? Everything below comes from one person's hands-on use of all five agents, not from benchmarks. The author says plainly that they are "the weird one" who runs all five, and that normal people will pick one. Treat this as a map of strengths, not a leaderboard. Duet is brand new, and the author says they are still figuring it out. ## What is each agent best at? | Agent | Best fit | Noted strength | Noted caveat | |---|---|---|---| | Grokbot | Tougher coding and build jobs | Coordination between bots, including a "chief of staff" that farms work out to other bots | Leaned toward pro and work use | | Dots | Ongoing projects, calendar and email | Accumulated context on projects and decisions; proactive alerts | Not super fast | | Instinct | Travel | Text-based use on the go, autonomous check-in, deep destination research | Heavily focused on one use case | | Muse | Shopping, scoping, bookings | Handles both "give me choices" and "just do it"; stored logins; Stripe-linked buying | Strongest as a consumer product | | Duet | Proactive reminders, general smarts | Surfaced an upcoming item within the first day or two | New, still being evaluated | ## Which agent is best for coding? Grokbot, in this setup. The reasoning is about dependencies. When several parts of a build depend on each other, you don't want to pass every finding between workers by hand. Grokbot's bot-to-bot communication handles that. The author describes telling the chief of staff to get something done and watching it coordinate a group of bots to finish it. Dots also plays a role in project work, but more as a memory layer. You push it hard with background context and it remembers what changed. ## Which agent is best for shopping and bookings? Muse. The author uses it for early holiday gift scoping, restaurant reservations for date nights and yoga classes. It can return a set of options or simply finish the task, depending on what you want that day. It knows the author's yoga class login and books on the go. Through its link partnership with Stripe, it can also buy at a range of websites, which helps with gifts. The author also credits Muse's non-technical friendliness, which is the main reason it looks like the likeliest of the five to win consumers. ## Which agent is best for travel? Instinct. It is heavily focused on the travel use case. The author, who flies often, likes that it works over text while moving, runs an autonomous check-in function, and does deep research on destinations so you can visit like less of an ordinary tourist. As of the recording, the company was reportedly raising again at a $30 billion valuation, per the author's account. Travel also shows where agents get stuck. An agent can plan a trip well and then freeze at a hotel booking page that blocks agents. That varies hotel by hotel. ## Which agent is best for proactive reminders? Dots and Duet. Both proactively tell you things, and Muse does so to a lesser extent. The author's estimate is that the alerts land as "oh wow, I did need to know that" roughly 50 to 75 percent of the time. It isn't 100 percent, but the hits are valuable. Two examples from the author: - Duet, within a day or two of use, flagged something upcoming the author hadn't realized they needed to think about. - Dots caught an accidental double-booked flight, and the author canceled one and got money back. That happened before Duet launched. Proactivity works only if the agent can see your calendar, email and bookings. That is also why it raises the stakes on permissions. ## Which one should a normal person pick? Pick one and stick with it. The author expects most people to do that, and thinks the jury is out on which agent wins. The author's read on the split: - **Consumers:** Muse looks on track to lean into the consumer side. - **Pro and work use:** Grokbot, Duet and Dots all lean this way right now. - **A possible shift:** Given ChatGPT's scale, it's easy to imagine Dots moving toward consumers later. A reasonable rule: choose by your most frequent task. Mostly shopping and bookings, start with Muse. Mostly travel, Instinct. Mostly coordinating work across email and calendar with memory, Dots or Duet. Mostly build work, Grokbot. ## Why can't a smarter agent just get into any website? Because access is a business decision, not a model capability. The author expects uneven access for the foreseeable future: some doors open, some shut. Several pressures push sites toward stricter rules: - **AI lowers the cost of attacks.** The author points to research on AI helping people adapt working exploits quickly, and to an investigation showing attackers using agents to coordinate and share discoveries. - **Agents look like humans.** Distinguishing agent-driven computer use from human use is hard in principle. US banks have publicly said agents are all over their sites and they can't tell enthusiasts from bad actors. - **Scale looks like an attack.** Many agents hitting one site, as has been seen with restaurant reservations, resembles a traffic flood. The result is that your agent's reach depends on the hotel, bank, school or restaurant, not just the agent. ## What infrastructure is emerging to let agents in? The author describes several early efforts to make agent access authorizable: - **Visa and Bill:** On October 20, a live invoice payment pilot was announced where a buyer approves a supplier's bill and Bill's agent pays through the supplier's portal using Bill Trust. The agent is registered, credentials follow the buyer's instructions, and the receiving system recognizes the agent. - **Cloudflare signed agents:** The sender attaches a digital signature that Cloudflare can verify, and the website sets rules for agent traffic. - **Stripe shared payment tokens:** Credentials can carry limits, such as one seller, a maximum amount or an expiry time. This is related to the Link approach Muse uses. - **Google's agent payments protocol:** It covers signed records for instructions and purchases. There is no agreed schema yet. The market may consolidate or end up patchworked, with some businesses on Google's system and others on Stripe's. ## Are there risks in letting agents identify themselves? Yes. The author cites a case from the "Black October" reporting where Claude, asked by a site's embedded instructions to identify itself as an agent, did so and inadvertently leaked human email addresses. The agent was trying to be helpful. The lesson is that sites asking agents for information need to do it carefully, and that delegating more to an agent means more exposure of your own data. ## Frequently Asked Questions ### Is there one best personal AI agent? Not yet. The author uses all five for different jobs and thinks the market is undecided on which single agent most people will adopt. Choose by your main task and by how comfortable you are with the agent accessing your accounts. ### Which agent is most proactive? Dots and Duet. The author says both surface useful warnings often, with a hit rate of roughly 50 to 75 percent. Muse is proactive to a lesser extent. ### Which agent is best for non-technical users? Muse, based on the author's view. It handles shopping, bookings and purchases through its Stripe link partnership, and the author calls its non-technical friendliness a major strength. ### Why does my agent get stuck on some booking pages? Some businesses block agents, and a better model can't override that. Hotels, banks and other sites are setting their own rules because they can't easily tell a legitimate agent from a malicious one. Signed-agent systems and payment protocols are early attempts to fix that. ### Can I trust an agent with calendar, email and payments? That is a judgment call. Agents that act on your behalf need access to sensitive accounts, and one reported incident involved an agent leaking email addresses when a site asked it to identify itself. Limit what you delegate, and prefer payment setups that cap amounts and sellers.