Agentic Q Developer First Impressions

Last week, I shared that AWS had announced an agentic CLI for Amazon Q Developer and why AWS has a tactical advantage in the category. 

It was too early for me to draw any conclusions about Q Developer CLI as compared to Anthropic’s Claude Code, but now I’ve had a week of testing under my belt, and have formed some first impressions of Q Developer’s agentic features. 

Let’s break them down into three categories: user experience, capabilities, and value. To make life a little easier, let’s agree to call this category of tool “agentic code assistants” or “ACAs” for short.

User Experience

ACAs primarily operate within a shell / terminal environment, which means that the user experience is confined to textual interaction with the assistant. This is not too different from, say, a ChatGPT session that happens in a browser, but it has the benefit of access to the filesystem, commands, and other assets in the environment. ACAs have some parallels to “reasoning” models like o1 and DeepSeek-R1, which insert themselves into the prompt/response cadence behind the scenes to enable complex problem-solving. By executing commands and reading content on disk, ACAs enrich their context to drive executable responses and code generation.

When it comes to Q vs. Claude, the user experiences are remarkably similar. Prompts are typed into the terminal session, and the ACA then responds, often requesting the user allow them to run a local command or to edit or create files. Each request is detailed, showing precisely what command will be executed or what content will be created or changed. While this “human in the loop” approach is shared between both ACAs, Claude allows the user to go full YOLO and whitelist commands during a session. Claude is a bit more conversational than Q, which tends to just get down to business.

One critical UX difference between Claude and Q is how they handle their context windows. With Claude, as the volume of information pulled into context increases, a warning is displayed informing the user how close to the limit the session is. The user is given the ability to send a “/compact” command that uses the model to deeply summarize the session and start a new one fresh. If the user runs out of context, Claude will automatically compact the session.

Q, on the other hand, handles all of the compaction behind the scenes in an attempt to create a more seamless experience for the user. Q’s approach is aspirational at this stage, as occasionally the fact that a session has been rebooted leaks through in responses, but I do think that the aspiration is the correct one. ACAs will be maximally useful when their technical implementation details don’t have to be surfaced to the end user. Bring on bigger and bigger context windows!

Capabilities

Both ACAs have effectively the same goals and scope, and I’ve spent the last week alternating between each ACA in the same project. I’ve been quite impressed with both ACAs, which are deeply capable with version control, build systems, and shell commands. When it comes to problem solving, each ACA has strengths and weaknesses. 

Claude excels with a tight iteration cycle, running test suites and build scripts, examining output, and taking action. I experimented with YOLO mode, letting the ACA run wild without too much intervention, with varying degrees of success. With smaller sets of warnings and build errors, Claude was consistently able to make progress, even without much human supervision. Claude was a bit stubborn at times, however. While adding a new feature to my project, Claude had generated string templates within a JavaScript file. It struggled mightily to fix an issue that required it to understand that there were multiple syntaxes at play: JavaScript and a template language. With very clear coaching, I was able to break the cycle and get Claude to make the right changes.

In what should come as a surprise to no one, Q was particularly strong with AWS APIs and artifacts like CloudFormation templates. While Claude was generally able to match Q in this regard, it often took several rounds of coaching. Q’s biggest challenge was amnesia. From time to time Q would make changes in a file that were internally consistent, but fail to make necessary changes in other parts of the code. In one such instance, it renamed a class to be much more descriptive, but failed to make the required changes to other parts of the code, resulting in build errors. Again, with coaching, I was able to get Q to stick the landing.

Value

Both ACAs are relatively early in their lifecycle, and I imagine that the monetization strategy isn’t fully worked out, yet. Currently, Claude Code cost is entirely based upon API calls, tightly coupling consumption and cost. This is typically a very Amazonian way of pricing services, and the challenge here is the same as ever when it comes to cloud services – it's really easy to get a large bill. While Claude Code is excellent, using it at scale would be prohibitively expensive. In its current state, I would hesitate to take full advantage of it for fear of cost overruns.

Q Developer is priced like an end-user product, not a consumption service. You pay a fixed rate per user per month. There are some reasonable service limits in place, but there’s no question Q Developer’s pricing is much more attractive at the moment. I am betting that the monetization strategies will evolve over time, hopefully in the direction of fixed user-based licensing.

Conclusions

We’ve explored and compared two compelling products across three dimensions, so what can we conclude? Which ACA should you use, if any? What do I advise based upon my experience?

In short, I think you’ll find a great deal of value in either ACA, but your use case will strongly dictate how effective the ACA is at accelerating your velocity. If you are working on IaC or directly with AWS APIs, Q is likely to be very effective. If you’re doing general purpose application development in Python, I found Claude Code to be a bit more accomplished. If money is no object, try both! If you’re on a budget? Q Developer is your pick!