DEV Community: Nahom Ashagrea

LeetCode Added an AI Assistant. We Built an AI Opponent. Here's Why That Matters.

Nahom Ashagrea — Sat, 08 Aug 2026 00:08:06 +0000

Ask Leet helps you solve coding problems. CodeDuel makes you compete against AI instead. The difference is more important than it sounds.

LeetCode just launched Ask Leet — an AI assistant deeply integrated into their platform that helps you brainstorm solutions, optimize code, generate test cases, and debug errors. It supports GPT-5.2, Gemini, Claude, and other models. It's genuinely impressive and LeetCode's largest product update in years.

I think it's moving in the wrong direction.

Not because AI assistance is bad. It isn't. But because the problem most developers actually need to solve right now isn't "how do I get help solving coding problems faster." It's "how do I make sure my raw problem-solving ability isn't quietly disappearing while I use AI tools every day."

Those are opposite problems. And they require opposite solutions.

What Ask Leet Does

Ask Leet is an AI that sits alongside you while you practice on LeetCode. Stuck on a problem? Ask Leet for a hint. Not sure why your solution is slow? Ask Leet to analyze it. Want to understand a different approach? Ask Leet to explain it.

It's essentially GitHub Copilot for LeetCode. The AI reduces friction, fills in gaps, and helps you get unstuck faster. For someone learning to code for the first time that's genuinely valuable. For someone preparing for interviews with AI assistance that might be acceptable depending on whether their target companies allow AI in interviews.

But for working developers who want to know honestly where their skills stand — Ask Leet doesn't answer that question. It makes the question harder to answer because every session has AI assistance baked in.

The Problem With AI Assistance During Practice

There's an important distinction between AI that helps you learn and AI that does the thinking for you.

A tutor who gives you a hint when you're completely stuck is different from one who just hands you the answer. The first develops your ability to solve problems — the second develops your ability to recognize correct solutions when you see them. Both have value but they produce different skills.

CodeDuel actually offers both modes. In a competitive duel there's no AI assistance — just you and Claude racing to solve the same problem. But in practice modes like Step by Step, Claude acts as a tutor — giving you hints one at a time without revealing the full solution, helping you develop the approach yourself rather than copying it.

The difference between CodeDuel's practice modes and Ask Leet isn't that one uses AI and the other doesn't. It's that CodeDuel's AI assistance is structured around developing your independent problem-solving ability — hints that guide you toward the answer rather than tools that generate the answer for you. And when you're ready to test what you've built you switch to duel mode and find out honestly where you stand.

Ask Leet sits at the other end of that spectrum. It's a full AI assistant integrated directly into your practice session — capable of generating solutions, optimizing your code, and explaining approaches on demand. For learning new concepts that's genuinely useful. But for measuring whether your unassisted skills are where you want them to be it makes the answer harder to find, not easier.

What We Built Instead

CodeDuel starts from a different premise. Instead of asking "how can AI help you solve problems faster?" we asked "what happens when AI is your opponent instead of your assistant?"

Here's how a duel works. You and Claude receive the same coding problem simultaneously. Both start working. You write your solution in a Monaco editor — the same editor that powers VS Code — while Claude generates its solution via the Anthropic API in real time. Both solutions run against the same hidden test cases in a sandboxed execution environment. You're scored on three dimensions: correctness, code quality, and speed. Claude explains its approach and gives you specific feedback on where your solution differed.

No AI assistance during the duel. Just you, the problem, and an extremely capable opponent waiting to see what you come up with.

Your ELO rating adjusts after every duel — up when you win, down when you lose, with harder problems carrying more ELO stakes. After twenty duels you have an honest number that tells you where your raw coding skills actually stand relative to AI. Not where they stand with AI help. Where they stand on their own.

Why the Distinction Matters

I'm not arguing that AI assistance is bad or that Ask Leet is a bad product. It serves a real use case and LeetCode's team built it well.

I'm arguing that there are two distinct developer needs right now and they require different tools:

The first need is learning to code faster and more productively with AI tools. Ask Leet serves this. GitHub Copilot serves this. Claude Code serves this. The entire industry is building for this need and doing it well.

The second need is maintaining and measuring raw coding ability independently of AI assistance. Almost nobody is building for this. CodeDuel is one of the very few products that takes this need seriously.

Why does the second need matter? Because technical interviews still largely test unassisted problem-solving ability. Because there are situations where AI tools aren't available — network outages, company restrictions, security environments. Because the developers who understand what they're doing at a fundamental level are better at directing AI tools than those who don't. And because raw problem-solving ability is a skill that atrophies without practice, regardless of how productive you are with AI assistance.

Two Tools for Two Different Goals

The honest answer is that Ask Leet and CodeDuel aren't really competing. They serve different goals and different moments in a developer's journey.

Use Ask Leet when you're learning something new and want AI to help you understand it. Use CodeDuel when you want to know honestly where your unassisted skills stand and push yourself to improve them.

Both are valuable. But in a world where every tool is adding AI assistance, the tools that measure and develop what you can do without AI are becoming more important, not less.

The best developers of the next decade will be fluent with AI tools AND capable without them. The first skill is being served by a wave of excellent products. The second skill needs more attention.

That's what CodeDuel is for.

Originally published at duelai.dev

Nahom Ashagrea
Founder of CodeDuel · Full Stack Developer · B.S. Computational AI, RIT

Are AI tools making Developers worse?

Nahom Ashagrea — Tue, 28 Jul 2026 12:02:35 +0000

GitHub Copilot and Claude are making developers faster. But are they making developers better? An honest look at what AI tools might be doing to raw coding skills.

I noticed something uncomfortable a few months ago. I was shipping faster than ever. Copilot was autocompleting my functions. Claude was helping me debug in minutes what used to take hours. By any productivity metric I was performing at my best.

Then I sat down to do a LeetCode problem without any AI assistance. Something felt off. The muscle memory was slower. The instinct to reach for the answer before working through the problem had gotten stronger. I wasn't sure if I was imagining it or if something real was happening to my raw coding ability.

I started asking other developers if they felt the same thing. Most of them did. Nobody had a good answer for what to do about it.

The Productivity Trap:

AI coding tools are genuinely remarkable. GitHub Copilot has tens of millions of users. Claude Code launched in 2026 and became one of the most discussed developer tools of the year almost immediately. Every major IDE now has AI integration. The productivity gains are real and well documented.

But productivity and skill are not the same thing. You can be more productive with a calculator without becoming better at mental arithmetic. You can navigate more efficiently with GPS without becoming better at reading maps. The tool does the work so your brain doesn't have to — and over time your brain gets out of practice.

The question is whether the same thing is happening with AI coding tools. Are developers becoming more productive while their underlying skills quietly atrophy?

What the Research Suggests:

A 2023 study from Stanford found that GitHub Copilot users wrote code 55% faster than those without it. That number gets cited constantly. What gets cited less often is the follow-up question nobody studied rigorously: what happens to the skills of those developers over time?

We have some indirect evidence from adjacent fields. Research on GPS navigation has found that heavy GPS users show reduced spatial memory and reduced ability to navigate without assistance. Research on calculator use in education found that students who relied heavily on calculators performed worse on mental arithmetic assessments than those who practiced without them.

None of this proves that AI coding tools are degrading developer skills. But it suggests the mechanism is plausible, when a tool removes the need to exercise a cognitive skill, that skill may weaken over time without deliberate practice to maintain it.

The Interview Problem:

The most concrete place where this tension surfaces is technical interviews. Most technical interviews still require whiteboard coding or live coding without AI assistance. A developer who has spent two years coding primarily with Copilot may find themselves less prepared for that environment than they would have been before AI tools existed, even if their day-to-day productivity has significantly improved.

This creates a strange situation where the tools that make you better at your job might make you worse at getting your next job. Developers who want to stay hireable need to maintain raw coding skills that their daily workflow no longer requires them to exercise.

The Honest Answer:

The honest answer is that we don't know yet. AI coding tools at scale have only existed for a few years. The longitudinal research needed to definitively answer whether they degrade underlying developer skills doesn't exist yet.

What we do know is that the anxiety is real and widespread. Developers who use AI tools daily report feeling less confident in their raw coding ability even as their productivity improves. That gap between feeling productive and feeling skilled is worth paying attention to.

The solution isn't to stop using AI tools, that ship has sailed and the productivity gains are too real to ignore. The solution is deliberate practice that maintains raw skills independently of AI assistance. The same way musicians who use music software still practice scales. The same way athletes who use data analytics still do fundamentals drills.

How to Stay Sharp:

A few approaches that developers are using to maintain raw skills in the age of AI:

Regular no-AI coding sessions. Set aside time each week to write code without any AI assistance. Even 30 minutes of deliberate no-AI practice maintains the muscle memory that daily AI-assisted work doesn't exercise.
Competitive coding practice. Platforms like LeetCode, HackerRank, and Codeforces provide problems that require raw problem-solving ability. Regular practice keeps the underlying skills sharp regardless of what tools you use in your day job.
Competing against AI directly. This is the approach we built CodeDuel around. If you want to know honestly how your raw coding skills compare to the AI you use every day, compete against it directly. Same problem, same time, scored on correctness, speed, and code quality. Your ELO rating gives you an honest picture of where you stand and your dashboard shows you exactly which categories need the most work.
Code review without AI. When reviewing other developers' code, resist the urge to run it through an AI first. Form your own opinion about correctness, performance, and style before using AI as a second opinion. This keeps the analytical skills that AI tends to replace most directly.

The Bottom Line:

AI coding tools are not going away and they shouldn't. The productivity gains are real and valuable. But developers who want to stay sharp — for interviews, for situations where AI isn't available, and for their own confidence — need to deliberately maintain the raw skills that AI tools no longer require them to exercise daily.

The best developers of the next decade will be those who can use AI tools fluently AND code effectively without them. Maintaining that second capability requires intentional practice.

Originally published at duelai.dev

Nahom Ashagrea
Founder of CodeDuel · Full Stack Developer · B.S. Computational AI, RIT