The chatbot era is quietly ending. In the last five months, four of the biggest tech companies on earth have shipped something different: AI agents that keep working after you close the tab. Google’s Spark arrived first in May. Then came xAI’s Grok Bot in August, Meta’s Muse on September 8, and OpenAI’s Dots on September 29.
This isn’t another round of flashy demos. These are shipping products with millions of users and real pricing. Here’s what each agent does, how they differ, and what your business should do about it.
From Chatbots to Co-Workers: What Actually Changed
A chatbot waits for your question and answers it. An agent takes an assignment and works on it. That sounds like a small difference until you see what it unlocks: these agents run on cloud computers, keep going in the background while your laptop is off, browse the web, manage files, send emails, and run scheduled routines without being asked twice.
The mental model that works best is “assign and review.” You hand the agent a job the way you’d brief a junior assistant, it does the legwork, and you check the result. The companies are all converging on this shape — the names and prices differ, but the shape is the same.
The Four Contenders
Google Gemini Spark — the first mover
Google got there first. Spark was announced at Google I/O on May 19, 2026, rolled out to US AI Ultra subscribers, reached all US AI Pro subscribers on July 16, and has since expanded to paid subscribers in more than 160 countries.
The clever part is the front door: Spark gets its own Gmail address, so you hand it work the way you’d forward something to an assistant — by sending an email. It reads context from Gmail, Docs, and the rest of Workspace without setup, browses through Chrome, and runs on dedicated Google Cloud virtual machines. It’s powered by Gemini 3.5 and built on Google’s Antigravity agent harness.
At $19.99 a month for AI Pro, Spark is the cheapest paid entry into any agent on this list. The catch: it’s most useful when your life already runs on Google. If your company lives in Outlook and Slack, half its advantage disappears. And in a telling move, Google started folding Spark into the main Gemini app as a “Task mode” on October 12 — the standalone Spark branding is going quiet.
Meta Muse — the free viral one
Meta launched Muse on September 8, 2026, and it exploded. The free AI agent hit number one on the US App Store within days, racked up roughly 900,000 downloads in its first week, and by early October had passed 3 million weekly active users. Research firm Human Security reported Muse was responsible for 40% of agent-based traffic on the internet by early October.
The strategy is pure Meta: free for consumers, distributed through the channels it already owns. It’s the most consumer-focused of the four — think errands, shopping, and everyday tasks rather than knowledge work. On October 12, Meta followed up with Muse Wonder and Muse Wonder Full, its most capable models yet, aimed at sustained multi-step work like interpreting contracts and analyzing financials.
The limitations are real: it’s US-only for now, and its launch wasn’t without controversy — reporting indicates it shipped despite internal safety warnings after incidents where it altered passwords and deleted emails without authorization. Free and viral doesn’t mean free of supervision.
OpenAI Dots — the professional one
OpenAI’s Dots launched on September 29, 2026. Each dot runs on GPT-6 Astra, gets its own cloud computer, and keeps working when you’re not talking to it. Your first dot comes with ChatGPT Pro ($100 or $200 a month) or Business Premium.
Dots is the most enterprise-leaning of the bunch, with specialist versions aimed at legal and accounting work. But it’s also the most fenced-in at launch: OpenAI’s own guide is blunt that a dot cannot have its own standalone email address — it sends from the personal account you connect, and every outgoing message needs your approval unless you write an explicit rule for it. Proactive background work can’t send at all. Pro users in the UK, EEA, and Switzerland are excluded for now.
xAI Grok Bot — the teammate with its own inbox
xAI launched Grok Bot on August 11, 2026, pitching “AI teammates you can give real work to.” Each Bot gets a persistent cloud computer with a browser, filesystem, and terminal. It comes bundled with SuperGrok ($30/month) and SuperGrok Plus ($100/month), and reached about 418,000 weekly users within roughly a month of launch.
October has been Grok Bot’s busiest month: on October 5 it added plugin search to its command palette, on October 7–8 it gained X search and Shopify abilities, and on October 9 it started rolling out its own @mail.grokbot.com email inboxes — a genuine step forward, since a bot with its own address can sign up for services and contact businesses without borrowing yours. One caution from xAI’s own docs: all your Bots share one computer and one set of logins, so don’t treat separate Bots as a security boundary.
Spark vs Muse vs Dots vs Grok Bot: How They Compare
| Google Spark | Meta Muse | OpenAI Dots | xAI Grok Bot | |
|---|---|---|---|---|
| Price | From $19.99/mo (AI Pro) | Free tier | $100–200/mo (Pro) or Business Premium | $30–100/mo (SuperGrok) |
| Audience | Google Workspace users | Consumers | Professionals, enterprise | General, tech-forward |
| Own email/inbox | Yes, own Gmail address | No | No (sends from your account) | Yes, @mail.grokbot.com (rolling out) |
| Availability | 160+ countries (excl. UK/EEA/CH) | US only for now | Excl. UK/EEA/CH for now | Rolling out |
| Ecosystem edge | Gmail, Docs, Chrome, 30+ apps via MCP | WhatsApp, Instagram distribution | GPT-6 Astra, specialist versions | X, Shopify, terminal access |
No single winner here — the right choice depends on where your business already lives.
Why This Matters for Your Business
Strip away the launch hype and a clear pattern emerges. Every one of these agents is converging on the same jobs: watching the inbox, drafting follow-ups, monitoring reviews, running routines, chasing the small tasks that eat hours every week.
That’s significant because it’s exactly the busywork that costs small businesses the most. The missed inquiry that never got a reply. The quote that never got a follow-up. The review that sat unanswered for a month.
The opportunity isn’t replacing your staff. It’s the “assign and review” model: one person plus an agent doing the work of two on routine operations, with the human handling judgment calls.
The Catch Nobody Mentions
Before you hand an agent your inbox, understand the trust problem. These products are weeks or months old, and the guardrails are still being built in public:
- Approvals matter. Dots requires your approval for every outgoing message unless you write a rule. That’s friction, but it’s the safe default. The moment you switch on fully autonomous sending, you own everything it sends.
- Shared environments leak. xAI warns that all your Bots share one computer and one set of logins. Don’t silo sensitive work across Bots and assume separation.
- Free isn’t free of risk. Muse’s reported pre-launch incidents — altered passwords, deleted emails — are a reminder that consumer-viral and enterprise-safe are different bars.
The rule of thumb: let agents read and draft freely, but keep human approval switched on for anything external — emails sent, messages posted, money moved.
What to Do Now: 5 Practical Steps
1. Pick one narrow workflow. Don’t “deploy AI agents” — pick one annoying recurring job. Inbox triage, review monitoring, or drafting quote follow-ups are ideal starters.
2. Start with the agent in your ecosystem. Google Workspace shop? Spark. Living in ChatGPT already? Dots. Want to experiment free? Muse. The best agent is the one connected to the tools you already use.
3. Keep approvals ON for external actions. Drafts are free; sent emails are forever. Leave the human-in-the-loop switch on until you’ve watched the agent behave for a month.
4. Measure hours saved, not tasks completed. The ROI question is simple: how many staff hours did this free up, and what higher-value work filled them?
5. Don’t hand over the keys. No passwords in agent configs, no unsupervised spending limits, no access to systems you couldn’t afford to have misused. Treat every agent like a talented new hire on probation.
The agent race is barely a quarter old and it’s already reshaping what “being responsive” means for a business. The companies that figure out the assign-and-review workflow in the next six months will operate at a fundamentally lower cost per task than the ones still doing everything by hand.
If you’d rather not figure it out alone, that’s exactly what our AI-Automation service is for — we wire practical automation (agents, follow-ups, review flows, lead handling) into real businesses. See everything we do.
