Q&A: Understanding durable skills in the age of AI

Vishnu Shankar, Chief Data Officer at Draup explains to Digital Journal about how HR leaders can better identify which technical skills are likely to remain durable going forward as AI adoption accelerates, and which are becoming more exposed to automation.

Shankar draws on a recent Draup analysis of ~3 million active job descriptions across engineering, data, and AI-related roles to explain why companies should stop asking only which jobs AI might replace and start looking more closely at which parts of those jobs will continue to create value.

Digital Journal: How is AI changing the way companies think about technical skills?

Vishnu Shankar: AI is changing the shelf life of skills. A skill that looks scarce today may be much easier to automate or assist a year from now. That creates a real problem for HR leaders because hiring and reskilling decisions are long-term bets. You’re not just filling a role for today. You’re deciding which capabilities the organization should keep building over the next several years.

That’s why we recently analyzed ~2.85 million active job descriptions across nine engineering, data, and AI-related roles from June 2025 through June 2026. We wanted to look inside the roles and understand which skills were still holding value as AI adoption accelerates.

That analysis became the basis for a new framework we created to help leaders shift the question from “Will this job survive AI?” to “Which parts of this job will still matter?” That helps them decide where to hire, where to reskill, and where to rethink the work altogether.

DJ: What makes a skill durable according to this framework, and which skills are most exposed to AI?

Shankar: I’d define a durable skill as one where the person still has to make the call. Debugging is a good example. That’s why, in our analysis of the Software Engineer role, debugging showed up in more than 166k job descriptions and systems design showed up in more than 100k.

The same pattern shows up in areas like model evaluation, data governance, and reliability engineering across the other roles we looked at. These skills hold value because the work is hard to fully hand off, not just because they’re trendy.

Routine coding, standard SQL, manual testing, and traditional ETL are different. AI can already help with much of that work, so those skills still matter, but they’re becoming easier to replicate.

That’s the split we saw in the data: durable skills require judgment, system-level reasoning, accountability, and human context. Exposed skills are more routine, repeatable, and easy for AI to assist.

DJ: How should organizations use this framework in hiring and workforce planning?

Shankar: I think the biggest mistake is treating today’s task list like a long-term workforce plan. Data work is a good way to see this.

If someone’s main value is writing routine SQL queries or building standard dashboards, AI is going to help with more of that work over time. But if someone can frame the right question, spot bad assumptions, understand where the data came from, and explain what it actually means for the business, that person becomes much more valuable.

That’s how HR leaders should use the framework. It helps them look past job titles and separate the work that’s likely to be assisted by AI from the capabilities they should keep building. In practice, that means asking: Which skills should we hire for, which should we train for, and which are becoming table stakes?

The point isn’t to stop hiring for technical skills. It’s to be much clearer about which technical skills are worth long-term investment. Organizations that make those decisions well will be in a much stronger position as AI continues to reshape technical work.

DJ: What surprised you most when analyzing nearly 3 million job descriptions?

Shankar: One thing that stood out was how consistent the pattern was across very different technical roles.

We looked at software engineering, DevOps, data engineering, machine learning, QA, data science, and several others. The specific skills changed from role to role, but the pattern did not. Employers consistently placed the highest value on skills that require judgment, design, oversight, and decision-making.

AI fluency was another big signal. GitHub Copilot, Cursor, and Claude appeared in more than 60k job descriptions across the nine roles we analyzed. That tells us AI fluency is no longer a differentiator. It’s quickly becoming a baseline expectation.

We also saw senior, staff, and principal titles outpacing more generic title variants across every role we analyzed. That matters because many of the routine tasks that once helped junior employees build experience are also the tasks most exposed to AI assistance.

That creates a real challenge for employers. You can’t simply automate the bottom rung of the career ladder and expect future senior talent to appear later. Companies will have to be a lot more intentional about giving early-career workers exposure to design, review, judgment, and problem-solving earlier in their careers.

DJ:  If there’s one thing HR leaders should take away from this research, what is it?

Shankar: I’d encourage leaders to stop asking which jobs AI will replace and start asking which skills will continue creating value. That’s really what this research is about.

Concretely, that means going role by role and separating what AI can already draft from what still needs a person to own the judgment call. For data roles, that’s the difference between someone who pulls a number and someone who can explain whether the number is right. For engineering roles, it’s the difference between writing code and being accountable for what ships.

Once you’ve made that split, the workforce plan writes itself. You hire and pay for the judgment work, you let AI absorb the routine work, and you make sure your junior talent is getting exposure to the judgment side early, not stuck doing only the tasks AI is about to take over.

AI isn’t eliminating technical work. It’s redistributing value within technical work. The organizations that map that shift now, rather than reacting to it later, are the ones that will be hiring from a position of strength in two years.