Entry-level tech positions are down 67% since 2022 and junior roles down 40–50% — not because of something unprecedented, but because AI agents now fill the role juniors used to occupy. If you replace the word "junior" with "agent," the job description doesn't change. Darren and Victor break down what skills actually survive this shift and why adaptability has always been the only durable career asset.
Victor opens with a striking split: on one end, companies like Anthropic paying data scientists $750,000+; on the other, Amazon laying off 16,000, Block laying off thousands. Darren frames it bluntly: "Be better than the others." But the conversation quickly moves past that into the structural forces at play.
Both hosts contextualize the current contraction as part of a recurring pattern: dot-com boom → dot-bomb collapse, 2020 COVID hiring surge → 2022 mass layoffs, 2026 correction. What's different this time is what's filling the vacated junior roles. Previous cycles saw cheaper offshore labor absorb the bottom of the ladder. This cycle, it's AI agents.
"You're junior, let's say software engineer. You enter the company, you're not making decisions. You're not changing the architecture. You're doing what you're told. And what I just described, if you replace the word junior with agent, we get the same result."
The thorny second-order problem: the traditional path to senior was through grunt work — boilerplate, bug fixes, the work seniors didn't want to do. That process was also how juniors learned the craft. If agents absorb the grunt work, the on-ramp disappears. Neither host has a clean answer. Victor's uncomfortable conclusion: junior may be becoming the current senior. The role is transforming, but nobody yet knows what the new on-ramp looks like.
Victor's framework for durable skills:
"If every developer on the internet can do what you do, the model trained on the internet can too."
Victor describes his current workflow with multiple AI agents as the career he always wanted — managing without people problems. The progression: one-shot attempter → pair programmer → manager of a team of agents. He notes the key danger: models almost never say no. That's where experience becomes critical — being able to distinguish good output from plausible-but-wrong output is now a senior skill.
The pattern they identify: developers who've cycled through multiple technology eras understand that Kubernetes will eventually be replaced by something else, just as Mesos was replaced by Kubernetes, and they adapt. Developers whose entire career is one stack can't let go because they have no reference for what comes next.
"There are people who embraced cloud, embraced containers, embraced Kubernetes, and there are still people fighting it. People who made those changes multiple times before — I'm not sure why this could be any different."
Victor's story: he gave junior candidates a task on a company laptop with Eclipse and a browser open. All were young, none knew the answer from memory. The one who switched to the browser and Googled got the job.
"I'm not looking for a person who has the solution to my problem because my problems are different every day. I want people with ability to solve problems. I don't care how."
The 2026 equivalent: the candidate who reaches for Claude Code or equivalent AI tool and demonstrates they can use it effectively — rather than insisting on writing everything from memory.
Both hosts close with direct words for new entrants. Victor: "Dear Junior, there is a strong chance that you will replace all old farts like we are who are still doubting it." Their practical advice: spend 20 bucks, spin up Claude Code for a month, and build something non-trivial. Don't evaluate it at week one — evaluate it after you've built muscle memory with it. Think about your first week with Java.
"You're junior, let's say software engineer. You enter the company, you're not making decisions. You're not changing the architecture. You're doing what you're told. And what I just described, if you replace the word junior with agent, we get the same result."
"If every developer on the internet can do what you do, the model trained on the internet can too."
"I always valued capacity to learn and to adapt over actual experience. It's easier to teach something specific to a person than to teach them how to be playful, curious, how to learn."
"Your freerange organic code is not going to be there faster than the code that I just made a million dollars with."
"Don't think that the results you will get from it the first week are the results you will get from it a year later. Think about your first week with Java."