Amazon Q Developer: Everything you need to know

When I reviewed Amazon Q Developer in 2024, I noted that it was able to generate whole functions in common programming languages with only a few fixes. It was useful for completing lines of code, doc strings, and if/for/while/try code blocks as you type. It could also scan for vulnerabilities and help you fix code problems. However, it could not generate full functions for some use cases, but instead reverted to line-by-line suggestions.

At the time, Amazon Q Developer was “powered by Amazon Bedrock” and trained on “high-quality AWS content.” I never knew what the first really meant, but that’s now moot: Amazon Q Developer now gives you a choice of Claude Sonnet versions, including 4.5, which is competitive with GPT-5 Codex.

At this point, completing lines of code qualifies as a “gimme,” the equivalent of a two-foot putt in golf. Generating whole functions is easy, generating complex applications is moderately difficult, and fixing reported bugs in large repositories ranges from moderately difficult to difficult.

Some of the current differentiators for coding agents are their ability to call tools (e.g. read files, run applications, show edit diffs, and understand Git and GitHub) and their ability to define and use Model Context Protocol (MCP) servers. MCP servers are tools that allow AI models to interact with external services, such as databases, APIs, and enterprise systems, using a standard, unified protocol. Another differentiator is the ability to run tools and projects in an isolated environment.

At risk of giving away the punch line, Amazon Q Developer supports tools, supports MCP servers but makes it harder than necessary to configure them, and doesn’t have the best implementation of isolated cloud environments. It’s still useful, however, and we can only hope that AWS will fill in the obvious gaps.

Amazon Q Developer competes with Google Jules, OpenAI Codex Cloud and OpenAI Codex CLI, Claude Code, Windsurf, Augment Code, Amp, Cline, Zencoder, and Gemini CLI. The current Amazon Q Developer versions as of this writing are Plug-in 1.104.0 and CLI 1.19.6.

Amazon Q Developer in editors

Amazon Q Developer runs in JetBrains IDEs, Visual Studio Code, Visual Studio, Eclipse, and the command line (CLI). By the way, the Q CLI has been rebranded as Kiro CLI with a version bump to 1.20; one change from the Q CLI / Kiro CLI I’ve been testing is the addition of an “auto” model option.

Amazon Q Developer generates real-time code suggestions ranging from snippets to full functions based on your comments and existing code. It also supports inline chat directly in the code editor, and CLI completions and natural language to Bash translation in the command line. Q Developer can write unit tests, optimize code, and scan for vulnerabilities. If you enable agentic coding, Q Developer can perform tasks on your behalf by reading and writing files, generating code diffs, and running shell commands, while incorporating your feedback and providing real-time updates along the way.

You can use chat and inline chat for any programming language. Inline code suggestions work best with C, C++, C#, Dart, Go, Java, JavaScript, Kotlin, Lua, PHP, PowerShell, Python, R, Ruby, Rust, Scala, Shell, SQL, Swift, SystemVerilog, TypeScript, CDK (TypeScript, Python), HCL (Terraform), JSON, and YAML. The chat works in at least 13 natural languages and automatically responds in the language you use for your query. This is a major improvement over last year’s language capabilities.

Amazon Q information in Visual Studio Code. At the right is the plug-in description. At the left you can see the chat help text above the chat input box. Note the blue switch at the lower left, which lets you choose whether to use agentic coding for the current request. Also note the choice of model, Claude Sonnet 4.5.
Foundry

Testing Amazon Q Developer Chat

To compare Amazon Q Developer Chat with Codex Chat I tried essentially the same prompt to create a command-line program to calculate pi.

Generate a C++ program to calculate the value of pi to the number of decimal digits specified in the command line. Make sure the program will be accurate to at least 1000 digits. Use the algorithm that is most efficient at run-time, regardless of the amount of code required. Save the program in the AmazonQ folder.

I used the Claude Sonnet 4.5 model, which I expected to do well on this task. Amazon Q Developer responded with the following.

I'll create a C++ program using the Chudnovsky algorithm, which is one of the fastest known algorithms for calculating π. It uses the GMP library for arbitrary precision arithmetic.
Creating the AmazonQ directory if it doesn't exist
Creating the C++ program that uses the Chudnovsky algorithm for calculating pi to arbitrary precision
Creating a Makefile to simplify compilation of the pi calculator
Creating a README with build and usage instructions
I've created a C++ program that calculates π using the Chudnovsky algorithm, which is one of the most efficient algorithms for computing π (converges at ~14 digits per iteration).
Files created in
/Users/martinheller/Code/AmazonQ/:
• pi_calculator.cpp
- Main program using GMP for arbitrary precision
• Makefile
- For easy compilation
• README.md
- Build and usage instructions
To use:
1. Install GMP: brew install gmp
2. Build: make
3. Run:./pi_calculator 1000 (for 1000 digits)
The program accurately handles 1000+ digits and uses the fastest known algorithm for runtime efficiency.

That sounded more promising than Codex’s first try, but it didn’t compile. For background, you should know that GMP has both C and C++ libraries; I only installed the C library, but Amazon Q Developer didn’t know that. I asked it to fix the build bug after highlighting that output, but it didn’t take the context. No matter, it still fixed the problem:

Let me try to build it first to see the specific error:
Attempting to build the program to identify the specific build error
Removing the unnecessary gmpxx.h include and fixing the C constant calculation
Removing -lgmpxx from the Makefile since we don't need the C++ wrapper
Building the program again to verify the fix works
Testing the program with 100 digits to verify it works correctly
Testing the program with the correct path
Running the pi calculator to verify it works
Fixed. The build issue was the missing
gmpxx.h header. I removed the unnecessary C++ wrapper include and simplified the constant calculation. The program now builds and runs correctly, calculating π to the specified precision.

I asked Amazon Q Developer Chat to time the program for 100 and 10,000 digits so that I could compare the efficiency with the MPFR/GMP implementation that Codex generated on its second try. Unsurprisingly, the calculation was a lot slower for 10,000 digits. I asked Q Developer how it might make this calculation more efficient for very high numbers of digits.

Yes, there are several optimizations for very high digit counts:
1. Binary Splitting - The most significant optimization. Instead of computing terms sequentially, recursively split the sum into halves and combine. This reduces complexity from O(n²) to O(n log n) for n terms.
2. Use MPFR instead of GMP - MPFR is optimized for floating-point operations and has better performance for transcendental functions.
3. Avoid Redundant Factorial Calculations - Your current code recalculates factorials from scratch each iteration. Binary splitting naturally avoids this.
4. Memory Management - Reuse mpz_t variables instead of allocating/deallocating in loops.
5. Parallel Computation - For extremely large calculations (millions of digits), parallelize the binary split recursion.
I can see you already have a more optimized version in
Code/Codex/pi_mpfr.cpp that uses binary splitting with MPFR. That implementation will be significantly faster for large digit counts (10,000+).

Oops. I didn’t mean to leave that file open where Amazon Q Developer could see it. Nevertheless, Q Developer/Claude Sonnet 4.5 answered correctly for all five points.

Pi calculator in C++ generated by Amazon Q Developer Chat using the Claude Sonnet 4.5 model. This implementation uses the GMP library for high-precision integer arithmetic.Foundry

Testing the Amazon Q Developer CLI

Before testing the Amazon Q Developer CLI (again) I checked its menu bar icon drop-down for updates. It was already up-to-date. Then I ran q in iTerm2 and asked it for help information.

Amazon Q Developer CLI help information display. This is as of Q 1.19.4. With version 1.20, the Q CLI is rebranded as the Kiro CLI. Foundry

For a first test I changed to the Cline rep directory, started the Q CLI, switched Q to the claude-sonnet-4.5 model (from the default 4.0), and asked it to explain the current repository.

Amazon Q Developer CLI with claude-sonnet-4.5 explaining the Cline repository. This is a reasonable high-level summary. It’s not as detailed as the explanation given by Codex last month, but it is easier to understand.
Foundry

Then I repeated the experiment to have Q explain the amazon-q-developer-cli repository.

Amazon Q Developer CLI explaining its own repository. Again, it gave a high-level summary that is easy to understand.Foundry

Next, I cloned the websocket repository from my own fork of the project, switched to that directory, ran the Q CLI, switched to the claude-sonnet-4.5 model, and asked Q to pick an open issue in the websocket repo and fix it in a new branch. Q struggled a bit with some of the issues, but found a problem with the chat example that it was able to fix.

Amazon Q Developer CLI explaining a fix it made to websocket. Issue #986 is about the chat example not working due to a missing CheckOrigin implementation. It was a simple three-line addition to the example. Q created the fix and checked it in to a new branch on my laptop.Foundry

I had to push the branch to my GitHub repository myself:

martinheller@Mac websocket % git push
fatal: The current branch fix-chat-example-checkorigin has no upstream branch.
To push the current branch and set the remote as upstream, use

git push --set-upstream origin fix-chat-example-checkorigin

To have this happen automatically for branches without a tracking
upstream, see 'push.autoSetupRemote' in 'git help config'.

martinheller@Mac websocket % git push --set-upstream origin fix-chat-example-checkorigin
Enumerating objects: 9, done.
Counting objects: 100% (9/9), done.
Delta compression using up to 12 threads
Compressing objects: 100% (5/5), done.
Writing objects: 100% (5/5), 686 bytes | 137.00 KiB/s, done.
Total 5 (delta 4), reused 0 (delta 0), pack-reused 0 (from 0)
remote: Resolving deltas: 100% (4/4), completed with 4 local objects.
remote:
remote: Create a pull request for 'fix-chat-example-checkorigin' on GitHub by visiting:
remote: https://github.com/meheller/websocket/pull/new/fix-chat-example-checkorigin
remote:
To https://github.com/meheller/websocket.git
* [new branch] fix-chat-example-checkorigin -> fix-chat-example-checkorigin
branch 'fix-chat-example-checkorigin' set up to track 'origin/fix-chat-example-checkorigin'.
martinheller@Mac websocket %

I didn’t create a pull request for the upstream gorilla/websocket repo that I had forked from, since that repo hadn’t changed in eight months, which tells me that the repo is inactive.

Amazon Q Developer CLI with claude-sonnet-4.5 didn’t do as good a job on this task as Codex Cloud. Codex Cloud operated directly on the GitHub repository and was able to fix a bug that Sonnet misunderstood and skipped.

I looked at the GitHub blame view for the new branch of my fork of websocket. The fix is in lines 38 through 40.

GitHub blame view for the fixed client.go in the chat example of my fork of gorilla/websocket.
Foundry

Amazon Q Developer agentic and MCP capabilities

The Amazon Q Developer CLI (newly renamed Kiro CLI) currently has the following built-in tools available to agents:

fs_read – Read files, directories, and images

fs_write – Create and edit files

execute_bash – Execute shell commands

use_aws – Make AWS CLI API calls

knowledge – Store and retrieve information across sessions

introspect – Provide information about Q CLI capabilities

In the built-in default agent, only fs_read can run without asking permission, but all tools are available, as well as legacy MCP servers. You can define your own agents, and set a default agent.

In addition, you can define and use MCP servers and tools, both in the CLI and the IDE. Oddly, the Amazon Q Developer plug-in for VS Code does not use any of the MCP or extensive tool capabilities of VS Code. Instead, it and the Q CLI / Kiro CLI use their own JSON MCP configuration files, which look about the same as Claude’s and the MCP specification’s config files. (If you’re interested, you can find the MCP specification here.)

Like Claude Code, the Q CLI / Kiro CLI has text commands to manage MCP server configurations; the Amazon Q Developer editor plug-ins have forms for that. Both require you to know the command that invokes the MCP server, but you can look that up for public MCP servers. The MCP standards organization maintains a registry of public MCP servers; so do GitHub (also here) and Anthropic (Claude).

Amazon Q Developer productivity effects

According to AWS, their internal developers have improved productivity using Q both quantitatively and qualitatively. One huge win was to ingest their internal Amazon knowledge repository (millions of documents) into Amazon Q Business so that developers could get answers based on information spread across those repositories. They reported “we reduced the time Amazon developers spent waiting for technical answers by over 450k hours and reduced the interruptions to ‘flow state’ of existing team members.”

AWS also unlocked “possibilities for large-scale technical modernization that previously seemed impractical,” which “fundamentally changed how [they] think about technical debt and system modernization.” Another effect was reducing the time to learn new languages and codebases. “One developer reported cutting their typical three-week ramp-up time for learning a new programming language down to just one week using Q Developer.”

With the Amazon Q Developer CLI agent, another internal developer was able to work with an unfamiliar codebase to build and implement a non-trivial feature within two days using Rust, a programming language they didn’t know, stating, “If I’d done this ‘the old fashioned way,’ I would estimate it would have taken me five to six weeks due to language and codebase ramp up time. More realistically, I wouldn’t have done it at all, because I don’t have that kind of time to devote.”

Amazon Q Developer on AWS

AWS notes that Amazon Q Developer is an expert on AWS. Q is available in the AWS Management Console (and in Microsoft Teams and Slack) to help optimize users’ cloud costs and resources, provide guidance on architectural best practices, investigate operational incidents, and diagnose and resolve networking issues. Of course, it’s easy for AWS to train its own model on its own documentation, procedures, best practices, and APIs. It might be a bit harder for customers.

Amazon Q Developer.NET porting and Java upgrades

Amazon Q Developer has agents that can help port.NET code from Windows to Linux, and help upgrade code from Java 8 to Java 17. These agents have upgraded more than 1,000 production Java applications, according to AWS.

Amazon Q Developer, data, and AI

In addition to straight code, Amazon Q Developer knows about data integration workflows. It can generate ETL scripts, troubleshoot errors, translate natural language to SQL queries, and work with data across 20+ data sources. It probably doesn’t hurt that AWS offers 20+ data sources as paid services. Q Developer can help you build machine learning models and also reduce the time to build, train, evaluate, and deploy AI models in SageMaker Studio.

Amazon Q Developer pricing

The Amazon Q Developer perpetual Free Tier gives you 50 agentic chat interactions per month. You can also transform up to 1,000 lines of code per month. The Pro Tier has expanded limits for $19 per month per user. I was able to perform this review using the free tier.

Conclusion

Amazon Q Developer has improved significantly over the last year. I can certainly recommend it to AWS customers. The free tier was good enough for me, but if I was using it all day, every day I’d most likely have to upgrade to the pro tier.