There’s a decent chance you installed a coding assistant, watched it suggest a for-loop you already knew how to write, and quietly turned it off a week later. AWS Q Developer is built for people who got bored of that. The pitch isn’t faster typing. It’s handing off the work that eats whole sprints: dragging a Java 8 service up to Java 17, writing the unit tests nobody volunteers for, and combing a dependency tree for problems before someone else finds them.
From autocomplete to agent
Q Developer began life as Amazon CodeWhisperer, a fairly conventional rival to GitHub Copilot. When the product reached general availability in April 2024, Amazon folded it into the broader Amazon Q family and gave it the Developer name. The rebrand wasn’t cosmetic. CodeWhisperer suggested lines. Q Developer runs tasks.
That difference changes how you work. You stop hovering over a keystroke waiting for ghost text and start describing an outcome, letting an agent read the repository, then reviewing a diff. It’s a smaller shift than the marketing implies, but it’s real, and it’s the reason the tool deserves a second look if you dismissed it in its earlier form.
Where you’ll actually meet it
Inside the editor
Q Developer plugs into VS Code, JetBrains IDEs, Visual Studio, and Eclipse. You get inline suggestions, a chat panel that can read your open files, an explain-this-error button, test generation, and a transformation command for upgrade work. None of that is unusual in 2026. The differentiator is that the chat side can act on AWS services, not just on the file in front of you.
In the terminal
The Q Developer CLI is where the tool gets genuinely useful for infrastructure people. It runs on macOS and Linux, offers command completion for hundreds of CLIs, and includes an agent that can read files, run commands, and iterate. It speaks the Model Context Protocol, so you can wire it into internal systems without building a bespoke integration for each one.
In the AWS console
This is the part with no direct Copilot equivalent. Ask why a Lambda function is timing out, why a subnet can’t reach the internet, or what’s driving this month’s bill, and Q Developer will dig through your account and answer in plain language. It also drafts CLI commands from a description, which quietly saves a lot of documentation tab-hopping.
The agentic features carry the real weight
Two capabilities justify the price of entry. The first is the developer agent. Point it at a repository, describe a feature in a sentence or two, and it plans the change, edits files across the codebase, runs tests, and keeps going until the work is either finished or stuck. You review the result the way you’d review a pull request from a contractor you’ve never met: carefully.
The second is code transformation, and it’s the most concrete claim Amazon makes. The Java upgrade path takes a codebase from Java 8 or 11 to 17, updating deprecated APIs, dependencies, and build files along the way. Amazon has said it used the tool internally to move more than a thousand production applications to Java 17, a project it credits with saving thousands of developer-years and roughly $260 million a year in maintenance. Your mileage will vary enormously, but even a partial win on a legacy service pays for the subscription. There’s also a .NET Framework to cross-platform .NET path, plus a mainframe path that converts COBOL to Java, still in preview.
Security scanning and code review
Every suggestion gets scanned for the usual suspects: injection flaws, hardcoded credentials, weak cryptography, log injection. The detector library comes from Amazon CodeGuru, so it’s a mature ruleset rather than something bolted on late. Findings arrive with an explanation and a suggested fix, which beats a red squiggle and a CVE number.
A review agent also comments on pull requests in GitHub and GitLab. It won’t replace a senior reviewer, but it catches the tedious things: missing null checks, inconsistent error handling, a test that asserts nothing. Given how routine credential leaks and dependency compromises have become, with the worst hacks of 2026 so far reading like a list of avoidable mistakes, an automated second pair of eyes is cheap insurance.
What it costs
There’s a free tier with roughly 50 agentic requests a month and capped suggestions, enough to decide whether the workflow suits you. Pro runs $19 per user per month with far higher limits, IP indemnification for suggestions, and a reference tracker that flags code matching public repositories and shows the licence. For organisations, it hooks into IAM Identity Center for provisioning and supports VPC endpoints so traffic stays inside your network.
Under the hood, the models are served through Amazon Bedrock, which is also the foundation most teams end up standing on once they outgrow off-the-shelf assistants. If you want to understand how that layer fits together, our guide to the building blocks for foundation model training and inference on AWS covers the pieces worth knowing.
How to get value instead of noise
Teams that get the most out of Q Developer tend to follow a similar playbook:
- Start with tests. Agents move faster and safer when there’s a suite telling them, and you, whether a change broke something.
- Scope requests tightly. “Add pagination to this endpoint” beats “improve the API” every time.
- Point it at boring work first. Upgrade paths, test scaffolding, boilerplate handlers, and documentation are where the hours actually disappear.
- Keep the diff small. A hundred-line agent change is reviewable. A three-thousand-line one isn’t, however good it looks.
- Watch the request budget. Agentic calls burn through allowances quickly when you let the agent wander.
Less obvious is the trap of measuring success by output volume. More generated code isn’t the same as more shipped features, and chasing token counts makes developers less productive than they think. The metric that matters is how long it takes to get a reviewed, deployed change into production.
Where it still gets in the way
Large monorepos are the weak spot. The agent works best when it can hold the relevant files in context, and sprawling codebases with tangled dependencies will slow it down or send it down a wrong path. Legacy languages outside Java and .NET get much thinner support. And an agent that edits files and runs shell commands needs an audit trail, which is why the industry’s growing interest in formal frameworks for disclosing bad AI behaviour is worth following. The same expectations will land on coding agents soon enough.
None of that makes Q Developer a poor buy. It makes it a tool with a clearly defined sweet spot: teams running AWS workloads, carrying legacy Java or .NET code, and willing to build a little process around an agent rather than expecting magic. That’s a narrower audience than the marketing suggests, and a considerably more useful one.