Amazon is launching an AI-powered chatbot for AWS customers called Q.

Introduced during a keynote at Amazon's re:Invent conference in Las Vegas this morning, Q (starting at $20 per user per year, now in public preview) can answer questions like "How do I build a web app using AWS?" " Q, trained with 17 years of AWS knowledge, will offer a list of possible solutions along with reasons why he might consider his proposals.

“You can chat, generate content, and take actions easily [with Q],” AWS CEO Adam Selipsky said on stage. "It's all about understanding your systems, your data warehouses, and your operations."

AWS customers configure Q by connecting and customizing it with organization-specific applications and software, such as Salesforce, Jira, Zendesk, Gmail, and Amazon S3 storage instances. Q indexes all connected data and content, “learning” aspects about a company, including its organizational structures, core concepts and product names.

From a web application, a company can ask Q to analyze, for example, what product features its customers are having problems with and possible ways to improve them, or, ChatGPT style, upload a file (a Word document, PDF, sheet calculation and like) and ask questions about that file. Q draws on your connections, integrations and data, including company-specific data, to generate responses along with citations.

Q goes beyond simply answering questions. The assistant can generate or summarize content such as blog posts, press releases, and emails. And he takes actions on a user's behalf through a set of configurable plugins, such as automatically creating service tickets, notifying particular teams in Slack, and updating dashboards in ServiceNow.

To avoid errors, Q has users inspect the actions he is about to perform before executing them and links the results for validation.

Accessible from the AWS Management Console and the aforementioned web app, as well as existing chat apps like Slack, Q has as deep a knowledge of AWS and the products and services available through it as you can imagine. Amazon says Q can understand the nuances of application workloads on AWS, suggesting AWS solutions for applications that only run for a few seconds rather than minutes or hours or applications that only access storage very infrequently, for example. example.

On stage, Selipsky gave the example of an application that relies on high-performance video encoding and transcoding. When asked which is the best EC2 instance for the application at hand, Q would give a list taking into account performance and cost considerations, Selipsky said.

"I really think this is going to be transformative," he said, referring to Q. "We want many different types of people doing different types of work to benefit from Amazon Q."

Q can also troubleshoot issues such as network connectivity issues, analyzing network configurations to provide solution steps.

And Q is related to CodeWhisperer, Amazon's service that can generate and interpret application code. Within a supported IDE (for example, Amazon's CodeCatalyst), Q can generate tests to compare software based on knowledge of a customer's code. Q can also create a draft plan and documentation to implement new features in software or transform code and update code packages, repositories and frameworks, plans that can then be refined and executed using natural language.

Selipsky says that a small team within Amazon used Q internally to upgrade about 1,000 applications from Java 8 to Java 17 (and test those applications) in just two days.

Q code transformation features only support upgrading Java 8 and Java 11 applications to Java 17 (coming soon, .NET Framework to cross-platform .NET), while all Q code-related features, including code transformation code, require a CodeWhisperer Professional. subscription. It is unknown when or if the subscription requirement will change.

Amazon says it is also developing Q its own products, such as AWS Supply Chain and QuickSight, a business analytics service. Q within QuickSight can provide viewing options for business reports, automatically reformat them, or answer questions about data referenced or contained in a report. In AWS Supply Chain, Q can answer queries like “What is causing my shipments to be delayed?” with up-to-the-minute analysis.

Q is also making its way into Amazon's contact center software, Amazon Connect. Now, with Q technology, customer service agents can get proposed answers to customer questions with suggested actions and links to related support articles without having to type those customer questions into a text bar. Q also generates a post-call summary that supervisors can use to track follow-up steps.

Selipsky stressed several times throughout the keynote that the responses Q gives (and the actions he takes) are completely controllable and filterable. Q will only return information that a user is authorized to see, and administrators can restrict sensitive topics, having Q filter out inappropriate questions and answers when necessary.

To mitigate hallucinations (i.e., cases where Q might make up facts, a common problem with generative AI systems), managers can choose to have Q only extract documents from the company rather than insights from any underlying model. The models that power Q (a combination of models from Bedrock, Amazon's AI development platform, including Amazon's internal Titan family) are not trained on a customer's data, Selipsky said.

Those points were undoubtedly aimed at companies wary of adopting generative AI for liability and security reasons. More than a dozen companies have issued bans or restrictions on ChatGPT, raising concerns about how data entered into the chatbot could be used and the risk of a data breach.

"If your user doesn't have permission to access something without Q, they won't be able to access it with Q either," Selipsky said. "Q understands and respects your existing identities, roles, and permissions...we will never use [enterprise content] to train the underlying models."

The heavy emphasis on privacy aside, in many ways Q feels like Amazon's answer to Microsoft's Copilot for Azure, which in turn was Microsoft's answer to Duet AI on Google Cloud. Both Copilot for Azure and Duet AI take the form of chat-based assistants for cloud customers, suggesting configurations for applications and environments and assisting with troubleshooting by identifying potential problems and solutions.

But Q appears to be a bit more comprehensive, covering a wide range of configuration, programming, and business intelligence use cases. Ray Wang, founder and principal analyst at Constellation Research, told TechCrunch that he believes it's re:Invent's "most important" announcement yet.

“It's about equipping AI developers to be successful,” he said; an important note to note considering that, according to at least one recent survey, many companies piloting generative AI are struggling to find business use cases and overcome poorly conceived implementations. . 0 0 0 0 0 0 0 0

We'll have to see if Q works as well as Amazon says.