Amazon launched autonomous agents inside Quick, its workplace AI assistant, at AWS Summit New York this week. The update lets any employee create a background agent by describing what it should do in plain language. The agent then runs continuously, even overnight, handling tasks like processing purchase orders, monitoring compliance updates, or tracking a sales pipeline.
Quick started life as Amazon Quick Suite in October 2025, combining business intelligence, research, and workflow tools into one product. It replaced an earlier AWS tool called Amazon Q Business, which was more technical and required IT involvement. The current version is designed so that someone in finance, operations, procurement, or sales can set it up themselves.
The product now connects to over 100 third-party tools and services. This week alone, 16 new integrations launched, including Shopify, Snowflake, WhatsApp, Zapier, and ZoomInfo. You can ask one question and Quick pulls answers from multiple connected systems at once, rather than making you log into each one separately.
On pricing, Quick starts free with no AWS account required, just an email address. Paid plans start at $20 per user per month. Microsoft Copilot and Google Gemini both sit at around $30 per user per month. Amazon is using price as a way in, especially with companies that do not run exclusively on Microsoft or Google tools.
That is the structural argument Quick is making: Copilot is best if your company lives inside Microsoft 365, and Gemini is best if your company runs Google Workspace. Quick is positioned for organisations running a mix of both, plus Salesforce, Slack, and other tools from different vendors. Most medium to large businesses are in that mixed position.
Early customer results cited by AWS include 3M saving its sales representatives more than five hours per week, and Amazon's internal book division cutting the time spent on coordination documents by 80 percent. New York Life, one of the largest US insurers, is reportedly using Quick for nightly reconciliation and compliance reporting.
The honest concern sitting behind all of this is governance. A recent security report found that 88 percent of organisations have already experienced confirmed or suspected AI agent security incidents. Only about 14 percent of businesses send agents into production with full IT and security sign-off first. Separately, Amazon's own retail site suffered a serious outage in March 2026 traced back to an engineer acting on incorrect advice from an AI agent that had read an outdated internal document.
The risk with agents is not that they are malicious. It is that they act on whatever data they can see, and that data is not always correct or current. If your internal records are messy, an agent will execute on the mess. Quick includes audit trails and permission controls, but setting those up properly still requires deliberate attention before you hand meaningful tasks to an agent.
The practical starting point for most operators is narrow and low-stakes: pick one repetitive process where errors are recoverable, connect the relevant data sources, and watch the agent for a few weeks before expanding its scope. The tools are genuinely useful. The risk comes from moving too fast and assuming the agent understands the context a colleague would naturally bring.