Dev

Hi, I run a small dev agency. 6 developers. Over the past year I've hired 4 of them. Two were great. Two were complete disasters that I'm still recovering from. Both of the bad hires absolutely nailed the technical interview. LeetCode mediums solved in 15 minutes. Clean code. Good explanations. And then they joined the team and I was shocked to see that they had no clue what they're doing. I'm not exaggerating. One of them solved a dynamic programming problem on the whiteboard and then spent 2 days trying to figure out why his POST request wasn't working. It was a typo in the URL. The other one aced a system design question but didn't know what an environment variable was. The signs were there in hindsight. The little pause before they started coding. Eyes clearly tracking something off-screen. Solutions that were weirdly optimal on the first attempt. When I asked follow-up questions they got vague. "I just thought about it logically." When I showed one of them his own interview code 2 months later he didn't recognize it. I'm not against AI. Actually the opposite. I want my team to leverage AI heavily. Copilot, Claude, ChatGPT, whatever makes us faster and better. That's the whole point. But there's a difference between someone who uses AI as a power tool and someone who used it to fake their way into a job they can't do. The cheaters can't even prompt properly because they don't understand the fundamentals. They don't know what to ask for. That's actually the second pain point and just as bad: so many candidates, if they know how to code then refuse to use AI tools to code. Not as a replacement for thinking, but as a productivity multiplier. In an agency environment, speed and quality matter. The devs on my team who combine their experience with AI produce the best work. But plenty of candidates act like using AI is cheating, or they paste AI output blindly without reviewing it, which is worse. Some of them take three hours to do something that someone using AI responsibly finishes in thirty minutes with better quality. Running an agency means client deadlines. Reputation. Real money on the line when someone delivers garbage. I can't afford to spend 6 months "coaching" someone who lied about their skill level. And I definitely can't keep explaining to clients why things are taking twice as long. We’ve already tried different things. We replaced some algorithm questions with small real-world tasks. We added a short take-home assignment.(The good Devs don't want to do that!). We do live pair coding during onboarding. We extended probation periods. Some people improve. Some don’t. When the baseline skill isn’t there, no amount of coaching closes the gap fast enough for client deadlines. As a small agency, we don’t have the luxury of letting someone take six months to learn fundamentals they should already know. I've thought about ditching coding interviews entirely. Just talk to people and check their GitHub. But people fake that too. Take-homes? Good candidates refuse them. Pair programming sessions? Better, but still gameable. I'm genuinely asking: how are you all handling this? What's actually working? Are there technical interview tools or platforms that make cheating harder while still being respectful to candidates? I’m tired of hiring developers who look great on paper but can’t ship reliable work for clients. I’m tired of reviewing PRs that show no understanding. And I’m tired of trying to push people to use tools that could make everyone’s life easier. I would really appreciate advice from other agency owners or team leads. How do you filter out LeetCode-only candidates? How do you assess real-world ability quickly? And how do you handle the AI adoption problem without turning the team into code janitors for people who won’t adapt?