Follow Datanami:
May 3, 2024

The Semi-Autonomous Agents of Amazon Q


Amazon Web Services (AWS) says business productivity will soar as a result of Amazon Q, its new generative AI service that became generally available this week. In particular, the capability to deploy semi-autonomous AI agents that can perform various tasks will be a boon for software developers and office workers, says Doug Seven, the company’s general manager and director of AI developer experience.

“I feel fairly confident saying that, for customers who are leveraging Q across a number of different use cases, they can see upwards of 80% more productivity in various roles in different jobs, whether that’s in software development or other sort of lines of business,” Seven told Datanami this week.

AWS unveiled a preview Amazon Q at its re:Invent conference in November, and it has been put through the paces by various customers, including as Blackberry, BT Group, and Toyota. Wednesday’s GA announcement brought some small changes, including the delivery of two main versions of the product, Amazon Q Developer, which is focused on working with computer code, and Amazon Q Business, which is focused on working with documents and data.

Amazon Q Developer leverages large language models (LLMs) and other generative AI technology to create an AI model that understands computer code. Q Developer can be used to generate, understand, troubleshoot, refactor, and debug code in a variety of languages, including SQL and Java, according to AWS. The software has also been trained to detect security vulnerabilities and apply fixes, and also has been trained on a knowledgebase of the AWS environment.

As part of this week’s launch, AWS is also releasing a new capability called Amazon Q Developer Agents. According to Seven, these agents go beyond functioning as a typical co-pilot because software developers can dispatch the agents to perform specific tasks on a semi-autonomous basis, thereby allowing the human developer to focus on other things.


Seven explained what the agents can do:

“Once you and Q agree on the work plan, Q’s going to go off on its own and it’s going to spin up a development environment of its own. It’s going branch that code. It’s going to compile the code. It’s going work on the code and it’s going make the changes and do all the things it needs to do,” Seven said. “And then it’s going come back to you at some point to say, okay, I’m done, here’s the code changes I’m suggesting. You as the developer can then review them just like you would do a review of a peer’s code and if it’s what you want, you accept it. And you move on to the next thing.”

Amazon Q also lets developers ask questions about their codebases, using natural language. For instances, a developer can ask Amazon Q Developer to “explain this code to me” or “walk me through how it works at a high level,” and Q will do it. It can be used to automatically remediate code vulnerabilities or even to upgrade Java code (.Net is next on the docket). Users can ask Q Developer about their AWS environments, including how much they spent last month on EC2 instances. It can even optimize SQL queries and ETL pipelines.

“I’m really excited about the agents and this idea of this autonomous or semi-autonomous capability that that AI can have,” Seven said. “We still believe there has to be human in the loop. So it’s going to do the work and it’s going to present it as a change and you’re can review it just like you would another developer. So it puts you in total control of that, but it’s really exciting.”

Customers are reporting good results. For instance, the UK-based telecom company BT Group used Q to generate about 100,000 lines of code in the first four months of use, and they received acceptance rates a little bit above the industry average around 37%, he said. Another early tester, National Australia Bank, had code acceptance of 50%. “We’re seeing great results from customers who are using it,” Seven said.

Amazon Q Business can be used to create apps, such as an employee onboarding plan generator (Image courtesy AWS)

AWS said that Q scored quite well on SWE-Bench Leaderboard and SWE-Bench Leaderboard (Lite), which are industry benchmarks for GenAI models. According to Seven, Amazon Q notched the top scores out of any GenAI products.

Amazon Q Developer is available now. Customers can access it via AWS Console, in Slack, or in IDEs, such as Visual Studio Code and JetBrains. The pro version of Q Developer costs $19 per developer per month.

Amazon Q Business, meanwhile, has been trained to function as GenAI assistant that’s knowledgeable about one’s business documents and data. The product sports more than 40 pre-built data connectors to pull data from S3, Gmail, Salesforce, ServiceNow, Slack, Sharepoint, Box, OneDrive, and other systems. It also has built-in analytics capabilities to be able to build reports and dashboards based on the data it finds.

“One of the things that’s really remarkable is how digital assistants like this work are increasingly capable,” Seven said. “They’re capable at taking in massive amounts of information and then summarizing that information. So to be able to ask things like having your Q instance connected to your Salesforce data and asking ‘What are the top five customer opportunities available to me right now.’ Or having it connected to your ticketing system where you can ask, ‘What is the overall customer sentiment in the last 30 days?’”

There is also a Q connection (a quonnection?) to Amazon QuickSight, AWS’s flagship business intelligence and analytics product. QuickSight is powerful, Seven said, but sometimes it can be difficult for the uninitiated to get it to do exactly what they want. With Amazon Q Business functioning as the natural language layer, newbie users can now tell QuickSight (via Q) exactly the type of data they want to see sliced and diced in the dashboard.

“So having Q in QuickSight means that I can express in natural language what it is I want,” Seven said. You can tell it to “‘Create a visualization of customer adoption rates over the last 60 days by region,’ and it can create those visualizations. Or ‘Give me a dashboard that I can check sales by team or sales force. That’s really remarkable in terms of making it easy to create those visualizations.”

One earlier adopter has even set up a Q Business endpoint in Slack that allows employees to get answers to questions directly within a Slack channel. By simply using the appropriate name with the @ symbol, it will call Q Business, reason over the data, and generate the response, Seven said.

Q Business respects data access policies. So if a user don’t have access to a specific piece of content, they won’t be able to access it.

AWS is also launching Q Apps, which allows users to turn their Q Business queries into reusable shrink-wrapped applications downloadable from a gallery. This new offering, which was inspired from AWS’s PartyRock experiment, will expand the potential impact that Q Business can have, Seven said.

A potential use of Q Apps could be generating onboarding plans for new employees, based on who the new employe works for and what division, Seven said. Or it could be a somebody in sales creating a sales script using Q Apps.

“I think it’s really cool,” he said. “I think we’ve tried for so long to make tools for non-developers, for subject-matter experts, to build applications for line of business. And I personally have built tools for this. But it’s always been a struggle. And this is one of the most remarkable advancements I’ve seen and one of the best generative AI in terms of supporting non-developers in self-servicing.”

Both Q Developer and Q Business are based on a variety of underlying technologies, including AWS products like the Trainium and Inferentia chips, first- and third-party LLMs on Amazon Bedrock, and the Sagemaker AI development environment, Seven said. “It’s a little bit of everything,” he said. “We’re not biased one way or the other.”

The pro version of Amazon Q Business is available for $20 per user per month. For more information, see

Related Items:


AWS Unveils Major Bedrock Upgrade: More AI Models and Enhanced User Flexibility

Amazon Launches AI Assistant, Amazon Q

AWS Adds Vector Capabilities to More Databases