Coding interviews in the AI-assisted era: why I built a practice loop inside my editor

A few things make coding interviews unusually awkward for cloud engineers.

In day-to-day work, I am rarely handed a clean function signature and asked to fill in the middle. I am reading Terraform plans, following a request across services, working out why a Kubernetes workload cannot reach a dependency, or deciding whether a design needs an event, a queue, a cache, or a different data model. The work is collaborative, messy, and full of documentation, logs, diagrams, and trade-offs.

Then an interview arrives. The constraint changes: explain your thinking clearly, turn an idea into working code, test it, and do all of that while the clock is visible.

That gap is real. Knowing AWS, GCP, Kubernetes, networking, or distributed systems does not automatically make the coding portion of an interview comfortable. And in an AI-assisted editor, there is another wrinkle: most of us now have excellent help one keystroke away. That is useful in production. It is less useful when the interview is testing whether you can structure the problem and defend the implementation.

I built Karaniph Coding Interview Coach to make that practice more deliberate without turning it into another browser tab, account, or subscription.

It is a VS Code extension, also available through Open VSX for compatible editors such as Cursor, Windsurf, Kiro, and VSCodium. The premise is simple: practice the interview loop in the editor where you already write code.

The problem is not only algorithms

There is plenty of good algorithm material available already. The difficult part, at least for me, is building a repeatable habit around the parts candidates often rush through:

• confirming the actual problem before choosing a data structure;
• naming assumptions and edge cases out loud;
• tracing an approach before committing to code;
• writing an implementation without treating the first draft as the final answer; and
• verifying that it works rather than relying on confidence.

In a real interview, the route to a solution matters. A candidate who says, “I will use a hash map because lookup needs to be constant time; here are the duplicate and empty-input cases I need to handle,” gives the interviewer useful signal. A candidate who jumps straight into code and discovers the cases late has made the conversation harder than it needs to be.

That is the behavior I wanted to rehearse.

A small workflow with a deliberate order

The extension is built around four stages:

Clarify → Trace → Code → Verify

1. Clarify

Before opening the editor to code, write down your understanding of the problem, constraints, examples, and likely edge cases. This is the part that maps most directly to a strong interview conversation.

For cloud engineers, it is familiar territory. We already clarify requirements before proposing architecture: expected traffic, failure modes, consistency requirements, budget, operational ownership. The same discipline works on a smaller coding problem.

2. Trace

Walk through the intended algorithm against an example. This is where an off-by-one error, a missing base case, or an invalid assumption is cheaper to find than it will be after ten minutes of implementation.

The extension creates a local solution.js in a trusted workspace at this point. The practice stays in a normal project folder, not in a hosted exercise environment.

3. Code

Implement the solution in your editor. You still have a proper file, syntax highlighting, and the normal rhythm of writing code. The goal is not to simulate a blank whiteboard; it is to make the important reasoning steps explicit before the code starts to take over.

4. Verify

Run the example, inspect the output, and correct the implementation. That last step sounds obvious, but it is exactly where interview pressure can make people skip a simple test or fail to explain what they would validate next.

The extension can run the local solution only in a trusted workspace. It does not require a remote execution service for the core workflow.

Why this matters more now

AI coding tools are part of engineering work now. I use them for exploring unfamiliar APIs, producing scaffolding, reviewing a first pass, and reducing repetitive work. That is not a problem to solve; it is a capability to use well.

The problem is confusing assisted output with retained skill.

A coding interview is one of the few places where you may need to demonstrate the intermediate steps: what you noticed, why you selected an approach, how you reasoned about complexity, and how you checked the result. Keeping those muscles exercised is worthwhile even beyond interviews. It makes design reviews clearer, incident response calmer, and pair-programming more effective.

The point of Coding Interview Coach is not to ban AI. It is to make the practice session yours first.

Optional AI feedback, on your terms

The extension has an optional AI Coach. It can review your approach and provide concise feedback, risks, and next-step questions rather than immediately dumping a complete answer into the exercise.

It uses a bring-your-own-key model. You can select OpenAI, Anthropic, Google AI Studio, OpenRouter, Vercel AI Gateway, or Cloudflare AI Gateway, then enter a model ID available to your account. The core workflow works without configuring any provider.

There are two practical reasons I prefer that model:

1. Choice matters. Multi-cloud engineers often already have an approved provider, gateway, or budget path. A fixed model list becomes obsolete quickly.
2. Practice data should not be mysterious. AI feedback is opt-in. Credentials are stored using VS Code Secret Storage, and your chosen provider or gateway handles the request under its own terms and billing.

This is especially useful if you want a nudge such as, “What happens for an empty input?” rather than a solution you can read once and forget five minutes later.

Who I had in mind when building it

This will be useful for more than one kind of developer, but I built it with these people in mind:

• AWS, GCP, Azure, platform, and SRE engineers preparing for software or infrastructure interviews that include coding rounds.
• Senior engineers and tech leads who spend more time in architecture, reviews, and operations than in standalone algorithm exercises, and want a consistent return-to-basics routine.
• Developers using AI-native editors who want to keep their independent problem-solving sharp while continuing to use AI responsibly at work.
• Candidates who prefer local tools over another website, dashboard, or managed interview-prep platform.

It is not positioned as a replacement for system-design preparation, language-specific depth, or realistic mock interviews. It is the coding-practice loop that sits alongside them.

What a short practice session looks like

A session can be small. Open a trusted local folder, select Coding Interview from the Activity Bar, and choose Start Session. Work through the four stages with one problem.

The useful unit is not “finish twenty questions.” It is “practice one complete explanation and verification cycle.” Do that consistently and you begin to notice where your process breaks down: perhaps you skip edge cases, perhaps you code before you have an approach, or perhaps you do not test the case you were most worried about.

That feedback is more useful than a streak counter alone.

Install it

Karaniph Coding Interview Coach is available now from both extension registries:

• VS Code Marketplace: Install Karaniph Coding Interview Coach
• Open VSX: Install from Open VSX

It is designed for VS Code and editors that support VS Code extensions or Open VSX, including compatible versions of Cursor, Windsurf, Kiro, and VSCodium.

The source code, documentation, and issue tracker are on GitHub: karaniph/interview-coach-rep.

If you are preparing for a coding interview while working in the age of Copilot-style tools, give it a try. Keep the assistance available when it is helpful—but keep a place in your workflow where the reasoning has to come from you.