Are Your Coding Skills Deteriorating Because of AI?

A CTO who leads a 40-person engineering team at a $91M AI company tried to write code without AI assistance the other day. He froze.

Deepak Bapat, CTO and co-founder of Tabs — the AI-native contract-to-cash platform used by Cursor, Statsig, and 200+ other companies — was experimenting with a new library and decided to write the code himself. No Claude Code, no Cursor, no copilot. “It was surprising how quickly the skills deteriorated,” he says. “There was a minute where I was just staring at a blank page being like, okay, where do I start?”

The skills came back. But the hesitation — the moment of blankness where muscle memory used to live — stuck with him.

The Dilemma No One Talks About

AI coding tools make engineers faster. The data is clear, the productivity gains are real, and Bapat himself runs a team where Claude Code and Cursor are standard tooling (roughly 50-50 across his engineers). He’s not arguing against AI-assisted coding.

The problem is what happens underneath. Engineers who never build foundational skills can’t evaluate what the AI produces. They can prompt, but they can’t debug. They can ship, but they can’t explain why the code works.

“Just because we have AI doesn’t mean people leave earlier,” Bapat observes. “People are still at the office for as late as they were before, but the expectation of what they can ship is much higher.” The productivity gains get absorbed into higher expectations, not shorter hours. And the skills gap underneath gets wider.

Investing in Junior Engineers at a Startup

Most AI startups optimize for speed. Hire senior, ship fast, raise the next round. Bapat takes a different position — one he acknowledges creates short-term friction.

“What I tell some of the more junior engineers is like, write the TRDs yourself, write some of the code yourself. You don’t want to just outsource all of your thinking because you will never be able to get it back. Because you won’t have built it in the first place.”

He gives junior engineers room to be slower. Three months of investment now, he argues, produces an engineer who can actually contribute independently rather than one who can only operate through AI intermediation.

“I was a junior person at one point and people invested in me,” he says. “So like the idea that I’m not going to do the same… my coaching tree and the people that are going to go off and do interesting things that have trained under me is important.”

This isn’t just philosophical. Bapat spends 20 minutes daily doing one-on-ones with individual engineers, auditing their technical review documents, reviewing their GitHub pull requests, and discussing the work in detail. It’s hands-on mentorship at a startup that could reasonably argue it doesn’t have time for it.

Code Review Replaced the Coding Exercise

The shift in skills showed up in Tabs’ interview process. One of their three technical onsite exercises is no longer a coding problem — it’s a code review.

“We stopped doing a coding exercise. One of our coding exercises became a code review exercise because we actually think that that is almost the more important thing that engineers are able to do today,” Bapat explains.

The reasoning is practical. If engineers increasingly write code through AI, the critical skill becomes evaluating code — reading it, finding problems, understanding implications. An engineer who can’t review a PR thoughtfully is more dangerous than one who can’t write a function from scratch.

Tabs still runs a coding exercise and a system design exercise. But the code review addition signals a shift in what “engineering skill” means. Adaptability matters more than memorized algorithms. “If we start to prompt you in the right way, can you adapt? Adaptability has become so important.”

Do More With More, Not More With Less

The common AI pitch is efficiency — do more with less. Bapat flips it. “If we have resource for X, I’m just gonna say we’re gonna go build more. It’s almost less limitation on what we can do as a product team.”

AI didn’t shrink his team. It expanded what the team builds. More engineers, more ambitious projects, fewer communication bottlenecks (product managers now read the codebase directly through AI tools). The productivity gain goes into scope, not headcount reduction.

“I think being able to do more with more is substantially better,” he says, pushing back against the narrative that AI’s primary value is doing the same work with fewer people.

FAQ

Are coding skills deteriorating because of AI tools?

Yes, according to hands-on experience. Tabs’ CTO tried writing code without AI assistance and found himself staring at a blank page — muscle memory had atrophied from consistent AI-assisted coding. The skills returned, but the initial deterioration was notable even for someone with 15 years of engineering experience.

Should junior engineers use AI coding tools?

Use them, but not exclusively. Tabs’ CTO pushes junior engineers to write technical review documents and some code themselves before reaching for AI tools. The concern: engineers who never build foundational skills can’t evaluate AI output, debug at a deep level, or explain why code works. Three months of manual investment pays compound dividends.

Why did Tabs replace a coding interview with code review?

If engineers increasingly write code through AI, the critical skill becomes evaluating code — reading it, spotting problems, understanding implications. Tabs replaced one of three technical onsite exercises with a code review because “that is almost the more important thing that engineers are able to do today.” They kept coding and system design exercises alongside it.

How is AI changing what engineering managers look for in hires?

Adaptability now matters more than memorized solutions. Tabs tests whether candidates can adapt mid-interview when prompted in new directions. The ability to evaluate code, understand system implications, and reason through novel problems outweighs the ability to write a sorting algorithm from memory.

Should startups invest in junior engineers despite AI productivity tools?

Bapat argues yes, despite the short-term friction. Junior engineers given room to build foundational skills become independently capable after a few months. Those who only operate through AI tools plateau as prompt operators. Tabs’ CTO spends 20 minutes daily mentoring individual engineers, reviewing their PRs and technical docs.

What does “do more with more” mean for AI engineering teams?

Instead of using AI to reduce headcount, Tabs uses it to expand what the team builds. AI reduced communication tax — product managers read codebases directly through AI tools. The productivity gain goes into more ambitious projects, not fewer engineers. Bapat argues this creates substantially more value than the “do more with less” efficiency narrative.

How should CTOs manage the AI coding skills gap?

Require engineers to explain their pull requests in detail — if they can’t explain a PR, they need to go deeper. Pair AI-assisted speed with periodic manual coding exercises. Invest in strong engineering management, not just flat structures with maximum autonomy. Audit technical review documents and PRs personally. Accept that short-term speed may decrease for long-term capability.

What engineering interview format works best in the AI era?

Tabs uses a three-part technical onsite: one coding exercise, one system design, and one code review. The code review tests whether candidates can read, understand, and critique code — the skill most needed when AI generates first drafts. System design remains essential because architectural thinking hasn’t been commoditized by AI tools yet.

Is AI making engineering teams more productive?

Yes, but the gains show up as expanded output, not reduced headcount. Tabs’ 40-person engineering team ships more than a team that size could have produced pre-AI. Engineers stay just as late — the bar for what constitutes a day’s work has risen. Token costs are a new line item CTOs need to manage across engineering budgets.