For the past 30 years I’ve had a front-row seat as technology disrupted industries. Headlines predicted that speech recognition would empty call centers, CAD systems would eliminate architects, and enterprise software would automate away entire professional functions. Yet it never quite worked out that way. As automation eliminated the routine, it elevated the complex. New roles that nobody envisioned emerged at the moment of disruption.

This time is certainly different. AI capability is advancing exponentially faster than anything we’ve witnessed before. And this time, the technology is coming for the technologists. 

Employment for 22- to 25-year-olds in the most AI-exposed occupations, software development among them, now sits 19% below where it would have been had it kept pace with less-exposed jobs, according to a recent report from Stanford’s Digital Economy Lab. Recent computer science graduates have a 7% unemployment rate, among the highest of any major. Tech internship postings on Handshake fell 30% in two years. But at the same time, postings seeking agentic AI skills grew more than 280% in a single year, and IBM says it will triple U.S. entry-level hiring in 2026. 

We may be in a temporary holding pattern rather than a permanent shift, as many CIOs and even one of Stanford’s own economists suggest.

Regardless, one thing is clear. The entry-level technical skills now in demand are fundamentally different from the ones the traditional on-ramp from curriculum to career was designed to develop. 

To better understand what is happening, we need to unbundle what we call “technical work” and understand that for the first time, AI is separating three distinct capabilities: 

  • The first is writing code. AI coding assistants can now generate functional software at extraordinary speed and low cost.
  • The second is evaluating code: determining whether software is secure, scalable, reliable, and appropriate for the business environment where it will operate.
  • The third is deploying systems safely at scale: architecture, governance, failover planning, operational risk, and accountability when systems fail.

Even as other capabilities become more valuable, AI is commoditizing the writing of code—an inconvenient truth since the traditional tech talent pipeline (academia, bootcamps, etc.) was designed specifically to meet this need. This is also a challenge because the current hiring trends with respect to entry-level tech talent run the risk of eliminating the very work experience through which these other skills have always been acquired. For decades, junior engineers learned through incremental exposure to real systems and operational pressure. They learned from outages, deployments, and broken integrations, with experienced engineers sitting beside them during difficult moments.

As a CIO, I spent countless nights on production outage calls affecting millions of customers. Everyone on the call could see the same dashboards and logs. But only a handful of veteran engineers could reliably diagnose the deeper issue, assess the downstream risks of a fix, or determine whether a rollback would cause even greater damage. In one case a senior technical leader who understood both the business process flow and the system design was able to troubleshoot the issue while more than 100 other operations and engineering staff looked on and waited. 

The most important decisions were never about typing code faster. They were about judgment amid uncertainty. Now those skills are required for even entry-level jobs—and our traditional model for tech education is structurally insufficient for the task.

For decades, as higher-order programming languages like C++ or Java replaced COBOL or other lower-level languages, the curriculum was updated but the skills being taught—learning a language and writing code—remained the same. So the teaching and learning model kept pace.

But this shift requires different skills. Even at the entry level, the skills that matter most now are ones developed through contact with real systems, real enterprise architectures, real failures, and real consequences. No syllabus can replicate that. 

While it may make near-term financial sense for individual companies to reduce entry-level hiring as AI allows senior engineers to produce more with fewer junior staff—collectively it means that the system is dismantling the on-ramp through which future expertise has always been built without planning for how to replace it.

Instead of continuing to question whether we still need junior engineers, industry leaders, academics, and workforce organizations need to focus on developing new models for supervised, applied technical learning in an AI-first world.

Some organizations are beginning to experiment with alternatives, including enterprise simulations, microinternships, service learning, and supervised project environments embedded directly into degree programs. 

These are a great start, but isolated pilots are not a system. We need a serious, structured effort to design, test, and validate new applied learning models and to share what works broadly enough that it can be adopted at scale. The stakes are high. What we do now will determine who builds what comes next and who’s in the room where the most important technology decisions of our time are made.

Every technological disruption in my career ultimately created more sophisticated and more valuable roles, not fewer ones. The challenge has always been whether we could build pathways into tech careers for everyone with the talent to contribute.

We now have a brief window to intentionally build those pathways. If we fail, a new system will emerge by default—one accessible primarily to the privileged few. We’re already seeing this gatekept system realized, as shown in a recent piece in The Wall Street Journal. The cost will be measured in lost economic mobility and in something harder to quantify: which problems AI gets pointed at and which ones get overlooked.  

We built that kind of system before. We shouldn’t do it again. 

Judith Spitz is the founder and CEO of Break Through Tech, a nonprofit organization building workforce development infrastructure for the AI economy. She previously served as a chief information officer at Verizon.