Blog
/Google DeepMind CEO Demis Hassabis advices grad students on how to "Leapfrog" the Intern Ladder and Future-Proof Your Career post AGI
"If I was to talk to a class of undergrads right now I would be telling them to get really unbelievably proficient with these tools."
That was the direct advice given by Google DeepMind CEO Demis Hassabis during a recent, high-profile debate with Anthropic’s Dario Amodei. As an educator training hundreds of students here in Australia, this resonated deeply with what I see on the ground. Hassabis suggests that mastering these tools "can be maybe better than a traditional internship," allowing students to "leapfrog" the standard hierarchy and become immediately "useful in a profession".
What Hassabis describes as "leapfrogging" is actually the emergence of a powerful equalizer in the workforce. In the short term, "unbelievable proficiency" means you can effectively compete with professionals who have 10 years of experience but have failed to update their workflows.
We are entering a phase where the mechanics of building are changing fundamentally. Anthropic CEO Dario Amodei noted that he already has engineers who "don't write any code anymore"—they let the model write it and they simply edit. An experienced senior developer might underestimate the complexity required to pull together modern AI frameworks, coding agents, and tools. If you can navigate this complex stack fluently, you bridge the experience gap. You become the builder who leverages AI to do the work of a team, while others are still writing every line by hand.
The reason you can gain this edge is what Hassabis calls the "capability overhang." He revealed that even the teams building these systems are "so busy building it it's hard to have also time to really explore" the full depth of even today's models.
This is your competitive advantage. While the creators are focused on the engineering, you have the time to explore the edges of what the models can actually do. Proficiency isn't just generating text; it is discovering the workflows and solutions that the engineers haven't even documented yet.
This isn't just about landing a junior role in a tightening market; it is preparation for the immediate future. Both Amodei and Hassabis discussed timelines for AGI that are startlingly short—potentially within the next few years,. Amodei predicts that models could be capable of doing most of what a software engineer does within "six to 12 months".
Hassabis warns that once we reach that point, we will be in "uncharted territory". By becoming fluent in these tools now, you are not just learning a software package; you are learning the language of the AGI era. The next five years will be defined by rapid compounding of intelligence, and the students who treat these tools as an extension of their own minds today will be the ones navigating that shift tomorrow.
Hassabis predicts the labor market shifts for entry-level jobs will play out prominently over the "next five years". In our collaborations with education partners across Australia, we are already seeing this reality. Employers are looking for the "unbelievably proficient"—those who can sit down on Day 1 and perform at a level that used to require a decade of experience. My advice echoes the head of DeepMind: Don't wait for a formal internship to hand you experience. Use these tools to build your own.