Comprehensive software reviews to make better IT decisions

Amazon Q Turns Citizen Developers Into James Bond

When James Bond’s quartermaster, Q, demonstrated a high-tech car in the 2015 film Spectre, it came loaded with advanced AI features like navigation and automatic deployment of weaponized defense mechanisms.

When Amazon Web Services (AWS) launched Q to general availability on April 30, 2024, it gave customers similarly advanced AI features that are more focused on productivity than spycraft. Sorry, no ejector seat is included with this chatbot, but it does come with a fully managed service with lots of customization available for individual users. Crafted around generative AI, Amazon Q presents a front end that seeks to compete with market leader Microsoft Copilot to be your enterprise AI assistant. It also promises to be your coding assistant and provide a new path for “citizen developer” employees that want to craft their own apps to automate tasks in their workflow.

To that end, Amazon Q is offered in two main flavors (Business and Developer), has been integrated into Amazon QuickSight (analytics) and Amazon Connect (contact center), and will be integrated with Amazon Supply Chain (inventory and demand planning).

As a fully managed service, we don’t know what large language model(s) (LLM) are behind the scenes with Q. Nor are we sure how data is being converted into embeddings in a vector database to stage the retrieval-augmented generation that Q orchestrates for us. But just like James Bond, who didn’t ask too many questions about the AI behind the car before he stole it out of the garage for an action-packed joy ride, Amazon Q users may not care. The real question is: does it help you get your work done?

The name is Q… Amazon Q

Q will connect to enterprise data stores to aid with search, summarization, and analysis. AWS is reaching out beyond the IT admin desk here and targeting business users with Q, providing an always-available assistant that sits at the ready throughout the workday. Q connects to a long list of AWS resources from Aurora (MySQL) relational database management system to Amazon WorkDocs, but it also connects to many other third-party data sources found in a typical enterprise workflow. That includes Microsoft, Google, IBM, Oracle, Salesforce and Adobe sources, with the complete list totaling more than 40 options. If that doesn’t do it for you, upload your custom data to an Amazon S3 cloud bucket and connect it.

AWS promises not to use any of your data as training data for any business subscriptions to Q – a key security and privacy concern for many organizations. You’ll still need to ensure your users have the right permissions to access only the data they’re supposed to. After all, you don’t want to allow everyone in the organization access to everyone else’s payroll information or chatbot messages.

Accessing the combined repository of an enterprise’s internal knowledge base promises to be powerful on its own. To mitigate the risk of LLMs providing erroneous outputs or taking users on irrelevant tangents, Amazon provides the option to limit the chatbot to draw only from the sources it is provided, rather than delve into the mysterious depths of its unknown training foundation. Output is provided complete with citations, so users can follow information back to its source and verify context.

Further management of the experience is aided by Amazon Guardrails, which allows users to input no-go zones for the chatbot. (For example, you may not want employees using Q to receive gambling advice or asking it to act as dungeon master for an epic Dungeons & Dragons campaign.) Identity-based permissions can also be applied to limit employee access to sensitive data.

It’s like using ChatGPT with all your enterprise data connected. That’s what most users have wanted since they discovered OpenAI’s viral web experiment back in 2022. But Q doesn’t stop there.

A no-code app maker for the citizen developer

Our 2021 Tech Trends report featured a Citizen Development 2.0 trend. We envisioned a bright future for non-technical workers that wanted to design their own workflow automations. Previously these sorts of requests had to be funneled through IT and wouldn’t always be delivered upon as IT might have viewed them as added customizations that would require maintenance. But no-code/low-code tools that put some development capabilities in the hands of non-technical users allow a self-service approach. AI was the key enabler that would allow more users to harness tools to create meaningful productivity enhancers.

From the What’s Next section of the same report: “Soon all that’s needed to build an app could be to explain it in plain language. Research lab OpenAI, cofounded by Elon Musk, offers the natural language generation tool GPT-3, which has already been used to create games, write job descriptions, and answer fitness questions.”

Today, OpenAI is the first among many entries in an extraordinarily competitive generative AI marketplace. OpenAI did not author the algorithm behind the Amazon Q chatbot, but it was likely another LLM so close in capability that, for most common business productivity tasks, it doesn’t matter. The important point here is that the natural language app creation capability we forecast in 2021 is now fully realized.

Users will be able to create Amazon Q Apps, which are chat-based workflows that can connect to all the data sources discussed above. The feature is turned on by default so long as the application environment was created using the IAM Identity Center in the Amazon Q Business Console, according to AWS documentation.

So what sort of work can you get done with a Q-based app? Here are some examples:

• Write documentation for software including a readme file.
• AWS imagines a marketing team member could find a useful response to their question and then generate marketing content that adheres to company branding guidelines.
• Create a research assistant.
• Apps can also be created programmatically by using the Q API.

It will be interesting to see what else users dream up given more time to experiment with Q.

While it’s a boon for non-technical users, developers will also appreciate Q for its coding assistant features. It’s designed to sit alongside you as you work, suggesting code snippets for common features, debugging script, and generally acting like the other half of a programming pair.

We could go on here about all the different integrations that Q has with other Amazon apps. Remember that before it graduated to full front-end interface, Q was an AI assistant in Amazon QuickSight that produced analytics visuals and supported natural language queries for your business intelligence. But you can probably ask Q to explain that to you.

Cool gadgets for every budget

The standard versions of Q Business and Developer each cost a very Copilot-comparable $20 per user per month. Unlike Copilot, AWS offers a “Lite” service tier at just $3 a month. Q Lite offers a Q&A-capable chatbot that has access to the same documents the user can access. Responses from the Lite version are shorter, and this tier doesn’t provide app-builder access. It’s also possible there will be additional latency or less detailed responses. Where budgets are tight, this could be an entry point to try out generative AI in your enterprise without breaking the bank.

You’ll also need to purchase units of indexing power – it’s charged hourly, but it seems likely you’ll keep the index running all the time. Indexing is priced in units, and each unit provides access to 20k documents, 100 hours of connector time, and 200 MB of text. If you exceed any of those thresholds, you’ll need to buy another unit of indexing. One unit of Starter indexing capability costs about 14 cents an hour or $100 per month. A unit of the more powerful and redundant Enterprise level indexing capability costs around 26 cents an hour or $190 per month.

The new interface for AWS shops

Organizations that already have a deployed AWS environment are going to be the first adopters of Q. While it does connect to multiple business data sources, organizations already invested in AWS will get the most value out of it.

So long as organizations are comfortable without control over what’s happening under the hood (regarding what LLM is providing outputs or how embeddings are created), then this managed service will be a no-brainer for AWS customers to turn on. Its no-code features for citizen developers represent a competitive differentiator from Copilot and may draw in more customers yet.

Organizations planning a Q deployment will need to think about how identity-based permissions extend to the new interface and whether further guardrails are required around the experience. But it won’t be much effort to get up and running, providing employees with new tools to automate their own workflows or coding efforts and driving increased productivity.