Always-on agents powered by Artificial Intelligence (AI) remain active and continue working on tasks instead of stopping after a single response. New products from Perplexity, Anthropic, Amazon, Airbnb, and Meta show how this model could affect work, travel, shopping, personal finance, and small custom apps. What matters to you is not merely whether an agent runs around the clock, but whether it acts reliably, stays affordable, and receives only the data it genuinely needs.
What persistent AI agents are designed to do
An AI agent is an assistant that pursues a goal and carries out several steps with some autonomy. Perplexity calls its new workplace agents “Teammates”: they can take on different roles inside a workspace, operate in parallel, and draw on existing projects, skills, and stored information from other Perplexity tools. Unlike Perplexity Computer, which completes individual assignments, Teammates are intended for ongoing work at a larger scale, according to the company.
Each Teammate runs on its own virtual computer, meaning a software-based environment with a desktop and browser, according to a report on Perplexity’s persistent agents. You can monitor progress in separate threads inside the Perplexity app and communicate with the agents through Slack. Several agents can handle different assignments simultaneously, such as maintaining an ongoing project while processing other workplace tasks.
This is more than a chatbot waiting for another question, but it is not a fully independent digital employee. You still need to define goals, provide access, and review results. “Always-on” therefore describes persistent availability and memory for ongoing tasks, not a guarantee of error-free work.
Pricing creates a substantial barrier: Perplexity Teammates requires its Max subscription, which costs $200 per month. Comparable workplace agents are listed at lower starting prices in the reports: OpenAI’s Dots at $100 and Anthropic’s Duet at $20. Cheaper plans may reach usage limits sooner, however; high consumption by Grok Bots recently led to tighter limits for that service.
Which everyday tasks agents can take over
For shopping, the agent becomes an interface between a short message and an order. Amazon’s Alexa for Shopping can receive product requests by text, search matching Amazon listings, return recommendations, and order an item on the customer’s behalf. Amazon’s examples include weekly groceries, gift recommendations, and finding a similar product from a photo.
The assistant is also supposed to remember preferences and notify you about tracked prices or deals. If you regularly request vegan products, Alexa can take that preference into account in future recommendations, according to Amazon. The texting feature is initially available only to a small group of Amazon customers in the United States and is expected to expand to millions more accounts; the report does not specify availability outside the US.
Airbnb is taking a different approach to travel. The company plans to launch its own AI travel agent in the first half of 2027 and does not want to rely on a text-heavy chat alone. Instead, the system is intended to assemble an interface using maps, photos, comparisons, calendars, and itineraries based on the traveler’s request.
Airbnb argues that planning a trip is not simply an unwanted chore. CEO Brian Chesky said 80 percent of travelers enjoy spending time researching trips; this figure comes from the provider and is not independently verified in the supplied reports. The planned Airbnb travel agent is therefore supposed to support comparison and group decisions without removing the inspiration and anticipation from planning.
Meta’s Muse “spaces” move toward user-created mini apps. You can describe a shopping hub, school dashboard, music tracker, or trivia game, and Muse will turn the request into an interactive view. If you connect services such as Gmail, Google Calendar, OpenTable, or Notion, the agent can continuously update prices, dates, and other details in the background; you can also check off items, add notes, and make edits.
These spaces can be displayed according to your schedule, such as showing a weekly planner during workdays and weekend plans on Saturday. Meta is also adding “Active Memory” and “Active Focus” so Muse can recognize when past information becomes relevant and work toward goals over time. Meta says spaces are rolling out to all Muse users, while sharing and collaborative features are still being explored.
How AI handles your finances
Anthropic’s Claude can connect to bank and card accounts through Plaid, a service that acts as a bridge between financial accounts and other applications. Plaid manages the connection without showing Claude your banking password. Access is read-only, meaning the system can analyze information but cannot move, manage, or invest your money.
One practical example is asking where your money went last month. Claude can organize spending into categories such as rent and utilities, shopping, or food, check which bills remain before your next paycheck, and assess how a large purchase such as a vacation fits your current finances. It can also explain unfamiliar financial terms and answer questions about balances or transactions, but Anthropic says it is not designed to provide investment advice.
The amount of visible information is considerable. According to the report on Claude’s bank connections, the system can see bank and account names, account types, the final four digits associated with an account, balances, credit limits, and up to two years of transaction history. Read-only access reduces the risk of an unwanted transfer, but it does not remove the consequences of an incorrect analysis or broad disclosure of personal financial data.
Anthropic additionally says that conversations within Personal Finance will not be used for model training, connected accounts can be removed at any time, and the feature will not display advertising or sponsored recommendations. It is launching as a beta, meaning a test version that is not yet final, for Pro, Max, and Ultra subscribers in the US. OpenAI offers a comparable Plaid-based product called Finances there, while Swiss availability is not specified for either service.
Pros and Cons of persistent AI agents
Pros:
- Continuity – An agent can keep track of ongoing assignments and goals without making you explain the full context in every conversation.
- Less routine work – Grocery orders, spending categories, schedule overviews, and project tasks can be partially combined or automated.
- Purpose-built interfaces – Travel views with maps or Muse spaces may be easier to use than lengthy responses inside a chat window.
- Multiple roles – Workplace agents can reportedly process different assignments in separate environments at the same time.
Cons:
- High prices – At $200 per month, Perplexity Teammates is aimed more at intensive professional use than occasional personal tasks.
- Broad data access – Persistent agents become more useful when they know your accounts, calendars, messages, or transaction history, but that also concentrates sensitive information.
- Limited availability – Several features are launching only in the US, remain in beta, or have merely been announced for 2027.
- Errors remain possible – A plausible spending analysis, product recommendation, or travel overview may still be incomplete or wrong and requires review.
What this means for you and Switzerland
If you are a beginner, the most sensible first step is a narrowly defined use case. You might let an agent structure a shopping list, maintain a weekly plan, or summarize spending you have already chosen to share, rather than immediately granting ordering rights, full email access, and several years of financial records. This makes it easier to judge whether the agent is reliable and whether the time saved outweighs the additional review work.
If you are an advanced user, you can gain more by separating tasks, data sources, and approval points. One workplace agent could monitor an ongoing project while a dedicated Muse space combines appointments and notes from connected services. For purchases and financial matters in particular, an agent can prepare recommendations while you retain control over costly or consequential actions.
Immediate availability in Switzerland remains unclear. Amazon’s shopping texts and Claude’s financial feature are explicitly starting in the US, Airbnb has not identified launch countries for its planned service, and the Perplexity reports do not provide a regional list. The sources also do not say whether Plaid supports the Swiss bank connections needed for this specific Claude feature.
Language, pricing, and privacy create further uncertainties. The subscription prices are quoted in US dollars, with no information about Swiss pricing, taxes, or billing. The reports also do not explain German-language support or Swiss data protection terms, so a broad product announcement does not yet amount to dependable local availability.
Always-on AI agents shift digital assistance from isolated answers to tasks that remain under continuous supervision. They can simplify planning and routine work when their responsibilities are clearly bounded, while specialized interfaces for travel, shopping, or finance may prove more useful than a single universal chat. The unresolved risk is the combination of extensive data access, possible bad decisions, and recurring costs that continue even when your digital coworker is producing little of value.
Sources
- Perplexity launches its own take on AI agents for work – The Verge, 2026-11-05
- Nach OpenAI und Anthropic bringt auch Perplexity dauerhafte KI-Agenten für komplexe Aufgaben – The Decoder, 2026-11-06
- Airbnb says it’s launching an AI travel agent in the first half of 2027 – TechCrunch, 2026-11-05
- Amazon will let you text Alexa your shopping list – The Verge, 2026-11-06
- Claude AI can plug directly into your bank account – The Verge, 2026-11-05
- Anthropic baut Claude zum persönlichen Finanzassistenten aus – The Decoder, 2026-11-06
- Meta’s Muse AI can create custom ‘spaces’ for the version of the app you need right now – The Verge, 2026-11-06
Image: Mustafa ezz via Pexels