Skip to content

Knowledge base

The junior learning path

AI absorbs the small, low-risk tasks juniors once learned on, so teams have to create learning work on purpose. Hiring shifts with it: interviewers now test whether a candidate questions and audits flawed AI-generated code.

Written by

Updated:

Getting Started 2

Our position. We hire on judgment under ambiguity and level on decision quality, not on how much work someone supervises (opens in a new tab). Experience matters for the judgment it produced, not for the title it earned.

The cost shows up in review first. A junior who cannot see what the agent missed ships it, and a senior absorbs it.

How should teams protect junior learning?

  • Point juniors at the agent as a copilot, not an autopilot, so they keep contact with the problem domain.
  • Bake AI into onboarding, for example as a skill that tours the codebase and opens an example pull request.
  • Test systems thinking and business-context translation, which twelve CTOs now rank above prompt engineering.

What happens to entry-level engineering work?

The transition is hardest on the people still building those foundations. AI absorbs exactly the well-scoped, low-risk tasks juniors once learned on, which threatens the pipeline that turns juniors into seniors (The AI-Native Software Engineer, Addy Osmani (opens in a new tab), 5 December 2025). That same talk names the trap as the seventy-percent problem: agents get roughly seventy percent of the way, and a junior cannot see the missing thirty percent, so they ship it in a pull request and a senior absorbs it in review. The Mechanical Survival blog puts the loss in sharper terms, that pointing juniors at autopilot rather than copilot divorces them not only from skills but from the problem domain, because stripping the practice strips the senses through which they would have perceived the work (Team dynamics after AI (opens in a new tab), 2 October 2025).

Field accounts confirm the gap is real and unsolved: by February 2026 a large share of tasks once handed to juniors was monopolized by AI, and few teams have yet adapted how they train them (Industry synthesis – AI’s impact on the state of the art in software engineering in 2026 (opens in a new tab), 6 February 2026), and the market that results is U-shaped, highly leveraged seniors at one end and AI-native juniors at the other, with the middle squeezed (video essay on the AI-native talent market (opens in a new tab), 13 January 2026). One workable counter-move is to run the pipeline in reverse, baking AI into onboarding so new hires arrive fluent and pull their senior teams forward (Bloomberg talk – AI adoption at Bloomberg (opens in a new tab), 16 December 2025), some teams packaging this as an onboarding skill that sets up the environment, tours the codebase, and opens an example pull request (Industry synthesis – AI’s impact… (opens in a new tab)).

The market data has started to show the squeeze: new-graduate hiring at the fifteen largest US tech companies is down 55 percent since 2019, and UC-system undergraduate computer-science enrollment fell for the first time since the dot-com bust (SF Standard – AI Writes the Code Now, What Is Left for Software Engineers? (opens in a new tab), 19 February 2026). The advice that follows for anyone still building foundations is not to bet a career on the specific skills being automated but on what sits above them, judgment, problem framing, and the ability to look at AI output and know whether it is right (A Country Full of Geniuses (opens in a new tab), 13 February 2026).

Hiring signal already rewards exactly that: asked what skill they now prize that they did not before AI, twelve CTOs converged not on prompt engineering but on critical thinking, systems thinking, and business-context translation, the ability to spot the confident-but-wrong answer, to reason about how parts scale together, and to turn an ambiguous requirement into the right thing to build (Final Round AI – CTOs Reveal How AI Changed Software Developer Hiring in 2025 (opens in a new tab), 8 July 2025).

Their interviews shifted accordingly, handing candidates slightly flawed AI-generated code and watching whether they pause, ask clarifying questions, and audit it rather than accept it, because the developers in demand are the ones who can clean up the mess AI creates, not the ones who generate more of it, as the CTOs Final Round AI interviewed put it.

A practitioner interview proposes a concrete apprenticeship for a team that already runs agents. A new hire first writes code by hand, for about a year, to know what the agents are dealing with. Then the lead engineer runs them as one more agent: the same task size, the same deterministic gates, for several months of low output and fast learning. Only after that do they run agents of their own. The skill that path builds is recognizing when an agent thrashes, changing one thing and breaking another. The practitioner spotted it only because he had struggled the same way himself. He offers the path as a guess, not a tested program (LIVE: Software Fundamentals in the Age of AI (opens in a new tab), August 2026).

Resources

Articles

Posts

Talks and videos