Soft Skills Are the Tiebreaker Now. Here Is What That Means for Your Prep.

A candidate came into a LastRoundAI mock session last quarter with a genuinely sharp system design answer. Clean, well-reasoned, right trade-offs. The practice feedback flagged one thing: every time the AI pushed back with a follow-up question, the candidate’s response closed off rather than opened up. Short answers. No acknowledgment of ambiguity. It read, on paper, like someone who didn’t want to think out loud with another person.

That’s the gap that gets people passed over now. Not the technical part.

What the data actually says

In LinkedIn’s 2025 Workplace Learning Report, 91% of L&D professionals said human skills are more valuable than ever, up from previous years. Seven of the top 10 skills LinkedIn identified as rising for 2025 are interpersonal or cognitive, not technical. That’s a meaningful tilt in a dataset that has historically skewed toward hard skills like cloud certifications and specific programming languages.

Google’s Project Aristotle research, which analyzed roughly 180 teams across the company over two years, found that what actually separated high performers from average ones wasn’t individual technical ability. It was how the team worked together, specifically whether people felt safe enough to take interpersonal risks: admit they didn’t know something, challenge a bad idea, ask a dumb question out loud. They called it psychological safety. Sales teams that scored high on it exceeded their revenue targets by an average of 17%. Teams that scored low fell short by up to 19%.

These findings hold up because the underlying problem is the same across industries: technical skill has a ceiling. Collaboration doesn’t.

Why AI tools made this worse, not better

Here’s an unpopular opinion: AI coding assistants have probably accelerated the soft-skills gap rather than closing it. When GitHub Copilot handles your boilerplate and ChatGPT drafts your first implementation, the floor on “can write decent code” rises fast. Everyone clears it eventually. The differentiation moves up the stack, to the person who can frame the right problem, push back on the wrong requirement, or keep a cross-functional conversation from derailing.

The 2025 Stack Overflow Developer Survey found that only 17% of developers agreed that AI agents had improved collaboration within their teams, making it the lowest-rated impact category by a wide margin. Developers are getting faster at individual tasks and slower, or at least no faster, at the team coordination layer. That’s the layer hiring managers are now trying to hire for.

I’m not certain this trend is permanent. It’s possible that as AI gets better at facilitating async communication, some of the collaboration premium will shrink. But right now, in 2026, the interview evidence points the other way.

The four skills that actually come up in interviews

Not all soft skills are evaluated equally in hiring. These four show up most often in the rounds that actually cut people:

Written communication under pressure

Design docs, async Slack threads, incident postmortems. The ability to write clearly when you’re tired, when the situation is ambiguous, and when the audience includes both engineers and product managers is genuinely rare. Interviews are starting to test this directly through take-home writing exercises, not just asking “are you a good communicator?”

Thinking out loud in productive ways

This is what distinguishes candidates in live coding and system design rounds. Not the answer, but the narration. Can you flag your own assumptions? Can you say “I’m not sure which direction to go here, so let me reason through the trade-offs”? Interviewers consistently rate candidates who articulate uncertainty better than candidates who project false confidence and get stuck.

Challenging without dismissing

Behavioral rounds now include questions specifically designed to surface how you handle disagreement. “Tell me about a time you pushed back on a product requirement.” The interviewers aren’t looking for compliance. They’re also not looking for someone who fights every decision. The candidates who move forward are the ones who can frame a technical concern as a trade-off conversation, not a judgment call about someone else’s competence.

Asking clarifying questions before coding

This one is simple and most candidates still skip it. Before writing a single line on a whiteboard or shared editor, good candidates ask about edge cases, scale assumptions, and constraints. This isn’t just a demonstration of thoroughness. It’s a signal that the person can work with incomplete requirements, which is most of what the actual job involves.

How to actually practice this before interviews

Reading about soft skills doesn’t do much. The practice has to be behavioral, which means you need reps in situations that feel like real stakes.

• Record yourself in mock interviews. Not to watch for filler words, but to notice where you stop narrating your thinking and go silent. Those silences are what interviewers experience as “hard to read.”
• Ask for written feedback from peers on design docs or async messages. The gap between what you intended to communicate and what they understood is the gap you need to close.
• Practice behavioral questions with someone who will push back. A question like “tell me about a conflict you had with a teammate” needs a follow-up questioner to actually be useful prep, not just a monologue rehearsal.
• Do system design practice out loud. Alone if necessary. The habit of narrating your reasoning is a muscle that atrophies if you only think silently.

LastRoundAI’s mock interview practice surfaces this specific pattern during AI-led sessions: candidates who narrate their reasoning get longer, more productive follow-up conversations from the AI interviewer, which is a decent proxy for what happens in real rounds. The behavioral interview prep guide also covers question frameworks that help structure answers without making them sound rehearsed.

One thing worth arguing about

There’s a version of this conversation that tips into “soft skills are more important than technical skills,” and I think that’s wrong. A candidate who can’t solve a reasonable medium-difficulty problem and who can’t discuss trade-offs in a system design round doesn’t get saved by being warm and collaborative. The floor matters.

What’s changed is what happens above the floor. When two candidates both clear the technical bar, which is increasingly common as prep resources and AI tools spread, the interpersonal layer is the tiebreaker. More often than not, it’s not even close. One person made the interviewer feel like a collaborator; the other made them feel like an evaluator. Companies hire for the first feeling and avoid the second one.

If you’re spending 90% of your prep time on LeetCode and 10% on everything else, that split probably needs to change.