On October 27, 2026, WIRED senior writer Reece Rogers published a hands‑on review recounting a day spent testing the performance of the current mainstream personal AI agents. The article, titled "I Tested Every Major AI Agent for a Day — Here's the Best One," evaluated Meta's Muse, OpenAI's Dots, Instinct, Anthropic's Claude Duet, Google's Gemini Spark and xAI's Grok Bot.
This is a relatively rare side‑by‑side comparison based on real everyday tasks. Approaching the test from a non‑technical user's perspective, the author designed five basic tasks and rated each agent from 1 to 5 on proactivity, adherence to instructions, and browser operations, then produced an overall ranking.
Testing background and methodology
The author notes that a year ago most AI agents "couldn’t do much well," but by 2026 the situation has clearly improved. These tools are no longer just chatbots: they can proactively scan email, track flight deals, assist with everyday shopping, and execute tasks through a virtual browser as "always‑on" assistants. They typically exist as single‑threaded conversations, more akin to a group‑chat experience.
The six tested products vary widely in price and form factor:
- Meta Muse: Free, standalone app, high download numbers since its September release.
- OpenAI Dots: Built into ChatGPT, $100/month, the most expensive.
- Instinct: Free but invite‑only, primarily interacts via SMS.
- Anthropic Claude Duet: $20/month, launched as a research preview the night before the test.
- Google Gemini Spark: A testing feature inside the Gemini app, $20/month.
- xAI Grok Bot: Standalone app, $30/month.
The author only connected a personal Google Calendar and Gmail account, deliberately limiting data exposure. He emphasizes that these tools are still early‑stage and that privacy and security risks are real — some users who connected bank accounts or Slack have inadvertently exposed sensitive information.
Main tasks and notable results
Morning briefing and story‑idea suggestions
The agent was asked to conduct in‑depth research within the author's WIRED beats and generate tech story ideas. Only Muse and Dots actually spent close to 30 minutes researching; most others produced quick outputs. In the end, Gemini Spark's angle was judged the most newsworthy, while Grok Bot and Instinct offered ideas that felt dated or bland.
Lunch ordering
The task required ordering for the author (prefers stronger flavors) and a vegetarian boss (prefers Vietnamese), with restaurants near the office. Most agents recommended reasonable places; Dots ultimately completed the actual order and performed best. Gemini Spark only ordered a $5 spring roll, which was clearly insufficient.
Handling schedule conflicts and other daily tasks
Agents performed unevenly when resolving calendar conflicts, setting night‑time reminders and similar scenarios. Some proactively suggested sensible next steps; others returned irrelevant or irritating suggestions.
After aggregating results across the five tasks, the author's final assessments were:
- Best overall: OpenAI Dots. Despite its high price and middling performance on the author's specialist story task, Dots achieved the highest overall task completion and reliability and appears relatively mature.
- Easiest to use / Best for typical users: Meta Muse. Free and simple to set up, Muse scored just behind Dots.
- Best interface: Anthropic Claude Duet. Although it had only been live for a day and placed mid‑pack overall, its dedicated app excelled at task tracking and smooth interaction flows.
- Most frustrating: Instinct. The SMS‑based interaction felt disjointed to the author, and its error responses were not user‑friendly.
A measured view: clear progress but still early
The value of this test lies in moving beyond vendor benchmarks to real scenarios. The results reveal several clear signals:
- Capabilities have improved. Compared with a year ago, mainstream agents have made substantive progress in browser control, multi‑step tasks and proactivity, and are no longer just demos.
- Price and availability gaps are significant. The free Muse can approach the performance of paid high‑end products, which shows that "useful" doesn't necessarily mean "expensive." Dots' $100/month remains a barrier for many users.
- Privacy and trust remain core obstacles. Almost every agent requires access to personal data to deliver maximum value, and user concerns about data security persist. Meta's privacy commitments are mentioned, but its past record still makes some users hesitant.
- Product form factors are not settled. Some are standalone apps, some are embedded in existing chat interfaces, and some use SMS. Which approach will ultimately prevail remains for the market to decide.
The author makes clear this is not an endorsement of any product. AI agents still make mistakes and have limited reliability in complex or high‑risk situations. The tests covered only a limited set of tasks; long‑term use and accumulated user data could change performance over time.
Industry significance
WIRED's cross‑comparison provides a rare "user‑perspective" reference amid the AI agent boom. It affirms technological progress while reminding readers that product value is determined not by marketing claims of "revolution," but by whether an agent can consistently and safely help with mundane daily tasks.
For now, OpenAI Dots leads on overall capability, Meta Muse has won early favor among ordinary users with its free model and ease of use, and Claude Duet shows potential in interaction experience. With further product iterations and price adjustments ahead, the competition has only just begun. For typical users, a pragmatic approach may be to start with free or low‑barrier trials, strictly control data permissions, and maintain human oversight of agent outputs.
▶️ Enjoy this piece? Creating it was hard work!
Follow our YouTube channel:
AI Rosetta
Your likes, subscriptions and shares are our greatest support! 🙏
More AI news at: https://www.airosetta.com