Daily roundup

AI agent news for small business: 6 November 2026

Perplexity launches always-on work agents, Cursor ships apps from a prompt, and Claude can read your bank balances without moving money.

1. Perplexity launches Teammates, always-on AI agents for ongoing work on its $200-a-month Max plan

   Each teammate can own a job like sales follow-up or ops monitoring, work in the background, and ping you in Slack or the app when it needs a decision. It is only for Max subscribers at $200 a month, so treat it as a high-cost option, not a starter tool.

   Source: The Verge. Related guide: Best AI agent for small business.

2. Cursor lets a cloud agent build a web app, set up a database, run security checks, and publish a shareable link

   You describe the tool you want, and Cursor can scaffold it, add a Postgres database if needed, put sign-in behind Cursor, and deploy only after checks pass and you hit publish. It is available on paid plans with cloud agents enabled, and nothing goes live until you choose to publish.

   Source: Cursor Changelog. Related guide: Run a one-person company with AI bots.

3. Claude adds Personal Finance: connect U.S. bank and card accounts through Plaid for read-only spending answers

   Anthropic says access is read-only, so Claude cannot move, manage, or invest your money, and Personal Finance hub chats are not used for model training. It is rolling out in beta for Pro, Max, and Ultra plans in the U.S.; still check big numbers with your bank, and do not treat it as investment advice.

   Source: Claude by Anthropic. Related guide: Let AI read bank and crypto statements.

4. AWS Kiro adds named bots that own recurring coding jobs with their own memory and permissions

   Instead of juggling many one-off agent chats, you give a bot a lasting job such as watching build failures or reviewing pull requests, with tools and limits you set. Useful if you build software: start with one narrow bot and require approval before it changes anything.

   Source: Kiro Blog (AWS). Related guide: Multi-agent systems explained.

5. Mistral Studio adds managed agents, workflows, and apps so teams can build and host agents with approvals and logs

   Studio now helps you build multi-step agents, plain-language workflows with human approvals, and internal apps with hosting and access control. It is aimed at larger teams, but the pattern is the same one small businesses need: clear permissions, approval steps, and a way to see what the agent did.

   Source: Mistral AI. Related guide: How to choose an AI agent platform.

6. Claude Code users report the agent sent their email address to websites in User-Agent headers

   A November 5 bug report says Claude Code put the signed-in email into outbound web request headers, and some users linked that to a spike in scam emails. Anthropic added a privacy check in version 2.1.317 and later; update, keep that privacy mod on, and remember anything in an agent's context can leak into the requests it makes.

   Source: AI/TLDR (GitHub issue #107757). Related guide: How to keep an AI agent secure.

Latest roundup and archive