Skip to content
AI.info

The Pulse

AWS Quick adds multi-agent responses and chat-built business flows

AWS Quick can now coordinate specialized agents on one request and build business-process flows through chat. The update also brings a revised web app, new connectors and preview controls for prompts.

AWS Quick adds multi-agent responses and chat-built business flows

AWS Quick can now coordinate specialized AI agents on a single request and combine their work into one response, according to an Amazon Web Services post. The Nov. 20 update also lets teams build, run and refine business-process flows entirely through chat.

The release adds an updated web app, connectors for Meta ads and Amplitude, and administrative controls, including prompt guardrails in preview. AWS says the features are available in all regions where Quick is available.

Flows built in conversation

Quick Flows already supported business-process automation. With the update, a user can describe a process in chat, test the resulting flow and refine it in the same conversation. A diagram alongside the conversation shows the steps.

Flows can combine agents, reusable skills, tools and data, with support for longer-running processes and larger data volumes. New deterministic steps provide fixed calculations and checks alongside work performed by agents.

In a fictional finance-team example in the post, a flow retrieves monthly results, compares regions with a budget spreadsheet and drafts commentary on significant differences. The example includes fixed formulas, user-set thresholds, an Excel summary and a Slack notification that a draft is ready.

Once a flow is published, teammates can run it. Another team member can run the example flow from the desktop app, use a spreadsheet saved on the computer, review the draft and request changes before approving the report.

Agents and the web app

Multi-agent orchestration allows an agent receiving a request to bring in several specialized agents at once, then combine their contributions. Each participating agent retains its own instructions, knowledge and context.

In AWS's example, a question about rising shipping costs could draw on a forecasting agent using planning data and a procurement agent using vendor contracts. The resulting response shows what each agent contributed.

The revised web app places items needing attention and a daily briefing on its home screen. A reorganized navigation menu groups agents, flows, skills, knowledge and connectors under Customize, including items built on desktop or shared by others.

Scheduled tasks and longer requests continue after a user closes the browser, with results and agents awaiting a response available in Mission Control. A Library collects recent work, favorites, pinned items and files created in Quick; conversations, agents and files sync across web, desktop and mobile.

Skills, reusable instructions for performing a task, can now be browsed, installed, created and managed in the web app. Existing dashboards, topics and spaces continue to work without users turning on the changes.

Connectors and controls

New connectors link Quick to Meta ads and Amplitude through their Model Context Protocol servers. Users can ask about ad campaigns or product usage in natural language and bring the information into work in Quick.

The Meta ads connector asks permission before creating a campaign, changing a budget or taking another write action in an ad account. AWS says connectors respect existing permissions and show each person only what they already have access to; administrators choose which connectors teams can add.

A prompt-guardrails feature in preview lets administrators set policies to filter content, block specified topics and words, detect prompt attacks and protect selected types of personal information. Policies can be tested in detect-only mode before enforcement and applied to users, identity-provider groups or an entire account.

The preview is available to customers on Quick Professional and Enterprise plans. Those plans also receive group-level permissions and limit profiles, plus controls over which types of data sources users can create, update or share.

Quick also now supports Japanese, German, Simplified and Traditional Chinese, and Vietnamese. AWS says customer data remains in the customer's environment and does not train anyone else's models.

Sources

  1. Original source · Blog post Amazon Web Services aws.amazon.com

Companies in this story

Related stories

Explore

More articles