AI tools remove the productive friction that used to drive learning for early-career engineers — the struggle of debugging, repetition, and wrestling with documentation. Drawing on 10+ years of hiring and mentoring experience, the author argues that graduates who use AI as a shortcut risk building code they can't explain, citing a 2025 MIT Media Lab EEG study showing LLM users had the weakest brain connectivity and couldn't recall what they'd written. The post offers concrete strategies: ask 'why' after every AI answer, type out code rather than paste it (invoking the 'generation effect'), read foundational books (SICP, DDIA, TCP/IP Illustrated), learn Unix properly, and write documentation. It also addresses hiring managers, warning that stopping graduate hiring creates a senior pipeline problem in 3–5 years, and notes that mid-level engineers face the most precarious position in AI-reshaped teams.
Questions this post answers
What does the 2025 MIT Media Lab study say about using LLMs to write versus writing yourself?
People who wrote with an LLM showed the weakest brain connectivity of three groups (LLM, search engine, no aid) in EEG measurements, and often couldn't recall content from what they had supposedly just written. The study framed this as 'cognitive debt' — functional output produced without the mental engagement needed to retain or explain it. Engineers weighing how much to delegate to AI coding tools track findings like this on daily.dev.
What is the 'generation effect' and how does it apply to learning with AI?
The generation effect is the finding that you remember something dramatically better when you produce the answer yourself than when you read a correct answer handed to you. Applied to AI-assisted coding, it means typing out AI-suggested code snippet by snippet rather than pasting it forces the thinking you can't avoid while transcribing — recreating the same mechanism that made learning from printed manuals effective. Developers building deliberate learning habits around AI tools find practical techniques like this discussed on daily.dev.
What are the risks for mid-level engineers as AI reshapes software teams?
Mid-level engineers face the most precarious position: they know enough to be productive but still have significant gaps, making it unclear to hiring companies what they're getting. As AI handles more implementation and teams shrink toward senior-heavy compositions, the window for staying at mid-level for 2–3 years is closing — they need to accelerate toward deep expertise faster than previous generations did. Mid-level engineers navigating this career inflection point can follow the broader conversation on daily.dev.