Some AI assistants that text you first are free to try
A Wired test of six always-on AI agents found they work more reliably than last year, though privacy and cost questions remain.
What happened
Wired reports that one of its software reporters spent a day testing six always-on AI agents. An AI agent is software that can take steps on its own, like scanning your email, building a lunch order or researching a topic, and these ones nudge you first when they spot a chance to help. The six were Meta's Muse, OpenAI's Dots, Instinct, Anthropic's Claude Duet, Google's Gemini Spark and xAI's Grok Bot.
According to Wired, agents still make mistakes, but they follow through on tasks and control a web browser better than they did last year. Muse is free, and Instinct is free with an invite code. Gemini Spark and Claude Duet cost $20 a month, Grok Bot is $30, and Dots is $100 a month. The outlet adds that Claude Duet is a wait-listed research preview.
Why it matters
The Wired reporter says he spent last year complaining that AI agents were bad at everything and that none of the so-called agents actually worked. Now he writes that they work more reliably, though not perfectly.
The same piece cautions that the software is still nascent and "often fraught with privacy and security concerns." It also relays a claim from an X user that an agent connected to his bank account and work Slack posted how much money he had for colleagues to see.
Cost is the other thing to watch. One of the AI-written story pitches in Wired's test claimed that Claude Duet bills its unprompted check-ins and overnight work to the same usage limit you pay for to chat. Treat that as a claim to check, not a proven fact.
Hanah-Marie Darley, co-founder and Chief AI Officer at Geordie, writes in Forbes that a higher bill could mean an agent did valuable work. It could also mean a task repeated itself or used a pricier model than needed. She suggests starting with a few workflows and naming the person or team accountable for judging whether the result was worth the spend.
TechCrunch found in its own testing of Dots that the agent flags calendar conflicts and deadlines, though not always in a timely way. It also reports that the privacy controls aren't very granular, and the app refuses read-only access to cloud drives.
What it means for your business
- Law and accounting firms: Keep client files and email out of any trial. Start with a calendar-only briefing and a public-news summary for your field.
- Clinics: Patient information doesn't belong in a test. Try an agent on scheduling reminders for staff meetings or supply research instead.
- Restaurants and local retail: Let an agent compare supplier prices or scan reviews, and check what it actually produces before you rely on it.
- Agencies: A morning briefing on client news and competitor moves is a good first job, and the output is easy to judge.
One thing to try this week
Pick one free option and give it one job: a morning briefing. Connect only your calendar, not your email or any accounts with money in them.
For five days, jot down three things each evening: what it sent, what was wrong, and whether it used anything you didn't ask for. If you're on a paid plan, check your usage limit too. If it earned its keep and stayed in its lane, widen its access a little. If not, you've lost nothing.