Yes—but not automatically. AI can take over routine coding tasks that once gave junior engineers practice, and some early-career employment data show worrying declines. The answer is not to preserve every old task unchanged: teams need to deliberately give newer engineers meaningful work in specifying behavior, testing, reviewing generated code and diagnosing failures. Where a mistake could cause serious harm, they also need to preserve practice without AI. That is a practical proposal, not a proven guarantee that employers will retain junior roles.
What is at stake for junior engineers?
Early-career software work is more than a queue of tickets. It is one way engineers learn to turn vague requests into precise requirements, recognize edge cases, understand unfamiliar systems and notice when a plausible-looking change is wrong. If AI handles more implementation, it may also remove some of the occasions on which newcomers build that judgment.
As an Amazon Associate I earn from qualifying purchases.
The question is not simply whether AI can write code. It is whether organizations will make room for people to acquire the knowledge needed to specify, test and take responsibility for software—including software partly produced by AI. Without those learning paths, the industry risks weakening the route from junior contributor to experienced engineer.
Free tools Windows power users keep installed
One-click scans. No signup required.
What do the early-career employment figures show?
A broader pattern in AI-exposed occupations
A 2025 Stanford Digital Economy Lab paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen analyzed monthly ADP payroll records through September 2025. It reported a 16% relative employment decline for workers ages 22–25 in occupations exposed to AI, controlling for firm-level shocks, while employment among experienced workers remained stable. The authors described the pattern as consistent with generative AI disproportionately affecting early-career employment, but cautioned that other factors may contribute. The estimate is not proof that AI alone caused the decline.
#1 Best Overall
The paper also found that patterns differed across occupations according to whether AI use was more automative or augmentative: declines among younger workers were more evident in occupations measured as more automative, and less evident in augmentative-use occupations. These are associations in the paper’s measures, not a forecast for every job, employer or engineering team.
The software-developer case study
In a separate occupation-specific case study, the paper reported that employment for software developers ages 22–25 had fallen nearly 20% from its late-2022 peak by September 2025. That figure describes a change over time in the studied data; it is distinct from the paper’s broader 16% relative estimate. The authors discuss limitations in their data and exposure classifications, as well as possible explanations beyond generative AI.
Together, these results make the concern credible without settling its cause or future scale. They show why it is worth asking whether early-career opportunities are shrinking; they do not establish that every company is cutting junior roles, or that AI is the sole reason for the changes.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Why can automation weaken an apprenticeship?
Automation can remove a task and the practice that came with it. Matt Beane’s research summary from UC Santa Barbara describes this problem in robotic surgery: when an experienced surgeon could operate without relying on a trainee’s hands, residents sometimes received fewer opportunities to practise and less effective instruction. Beane also documented residents building capability through simulator practice and other informal routes, and argued that work and technology can be designed to keep apprentices involved.
The point applies beyond surgery. A routine coding assignment may be inefficient, but it can also teach a newcomer how a system behaves, what a requirement leaves unsaid and how to find a defect. If AI takes over that assignment, a team should ask what will replace the learning—not assume that exposure to finished code, or access to a chatbot, provides equivalent practice.
Beane summarized the risk this way: “That system of learning — which we all take for granted — is under a new kind of threat from the way we’re handling new technologies. It doesn’t have to be that way.” His observation is about how work is organized, not a claim that technology makes apprenticeship impossible.
Rank #3
What can chess grandmasters teach software teams?
A 2024 peer-reviewed study analyzed more than 11.6 million decisions by elite chess players. Its authors reported that human decision quality improved steadily across four decades, without distinct bursts of improvement around the two AI advances they studied. That finding supports a limited lesson: stronger AI and continuing growth in human skill can coexist.
It does not show that chess became more popular because of engines, that software engineering will follow the same pattern, or that companies will keep junior jobs for the sake of training. Chess offers a reason not to treat AI capability and human development as inevitable opposites—not proof that the labor-market problem will resolve itself.
How should teams redesign junior work?
Start with a direct question: What could happen if the AI is wrong? A mistake in a disposable prototype is not equivalent to a defect in software whose failure could harm people. The consequence should shape both the amount of AI assistance and the amount of unaided practice a team preserves. The following framework is a recommendation, not a validated universal policy.
| Software context | Suggested junior work | Practice to preserve | AI’s role |
|---|---|---|---|
| Prototype or other lower-risk commercial software | Clarify requirements, write tests, review generated code and investigate failures. | Give juniors enough implementation and debugging work to understand the system, rather than limiting them to approving output. | Use it to accelerate routine implementation or act as a coach, while expecting engineers to check its work. |
| Mission-critical software where failure could cause harm | Combine supervised implementation with test design, code review, diagnosis and explicit reasoning about failure modes. | Schedule structured debugging and problem-solving without AI, and simulate failures so engineers practise responding. | Use it as a coach or simulator as well as a tool under review; do not let it replace all practice when errors matter. |
Managers can make the distinction operational: identify the consequence of failure, decide which work requires human sign-off, and plan when newer engineers will practise without AI. Then give them feedback on the reasoning behind a test, review or diagnosis—not just whether the final code passes. This turns “learn by doing” into something a team can inspect and improve.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can AI help teach the skills it may displace?
It can be used as a coach or simulator: for example, to generate a debugging scenario, suggest edge cases for an engineer to assess, or explain a code path after the engineer has first tried to understand it. That is different from treating a generated answer as evidence that someone has mastered the task. The learning value depends on whether a person still has to make and explain decisions, receive feedback and sometimes work without assistance.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →A Built In article recounts an anecdote about an intern feeding client meeting transcripts to an AI coding agent and being told not to understand what clients say. The anecdote captures a tension, not a sound general rule. Delegating routine transcription or implementation does not remove the need to understand the client’s domain: engineers still need enough context to judge whether the result meets the real requirement.
Best Value
Where will the next generation of software developers come from?
There is no guarantee that the market will preserve today’s junior roles or create an equivalent path on its own. One possible response is to move entry-level work away from routine production and toward requirements, test creation, review and failure diagnosis, while retaining enough hands-on and unaided practice to build technical understanding. Whether employers adopt that approach at scale remains an open question.
The central choice is organizational: use AI only to reduce the need for junior execution, or redesign work so newcomers can use it while still learning how software fails and how to verify a result. Chess shows that human skill and AI can advance together; the employment evidence and apprenticeship examples show why that outcome cannot simply be assumed for engineering.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
