Loading...

AI reduced coding effort. Engineering talent is still scarce.

AI now writes nearly half of all new code and engineering hiring got harder, not easier. The market split into two markets moving in opposite directions, and the 2022 playbook for breaking in (or hiring well) no longer works on either side.

AI writes nearly half of all new code and 87.5 percent of tech leaders still call engineering hiring difficult or worse. The reason is that AI did not collapse hiring into one easier market. It split it into a junior side that has collapsed and a senior side that is overflowing with applicants no one can verify. The 2022 playbook no longer works for juniors trying to break in or for companies trying to hire.

The talent crunch that never ended

In 2022, a hiring manager hearing that AI would write nearly half of all new code by 2026 would have assumed the talent crunch was about to end. Four years on, the opposite has happened. A 2026 Lemon.io survey found that 87.5 percent of tech leaders now rate hiring skilled engineers as difficult or worse. Not a single respondent called it easy. ManpowerGroup's global survey puts AI skills at the top of the list of capabilities employers cannot find, with 72 percent reporting hiring difficulty.

What actually happened is that AI did not collapse engineering hiring into one easier market. It split it into two markets moving in opposite directions, and that split is what the aggregate numbers hide.

The market broke in half

Overall software developer postings are down roughly 70 percent from their early-2022 peak. Junior hiring at Big Tech has fallen to just 7 percent of new hires, more than 50 percent below pre-pandemic 2019 levels. Companies replaced entry-level grunt work with Cursor, Copilot, and Claude Code, then quietly stopped opening the door.

The senior side looks nothing like the junior side. Open roles for senior AI/ML, infrastructure, and security engineers have grown, not shrunk, and AI-fluent senior engineers now command compensation premiums of 50 percent or more over their non-AI peers. The shortage is narrow. Companies are short of the specific seniors they need, not of engineers in general.

Why AI made hiring harder

Recruiters expected AI tools to flood the market with capable developers. What the tools actually did was flood the market with developers who look capable. Resumes are AI-polished, take-home assignments are AI-completed, and screening tests pass cleanly until the candidate hits real production code, a legacy migration, or an ambiguous architectural call. Andela, a firm that vets engineering talent globally, reports a near-75 percent fail rate on basic AI skills assessments. Most candidates can prompt a model. Standard screening cannot tell the difference between candidates who actually understand the AI work and candidates who only prompt for it.

The signal-to-noise ratio in the applicant pool collapsed. Every additional applicant makes finding the real one harder, because the surface signals that used to separate competent engineers from weak ones (clean resume, working take-home, articulate phone screen) no longer separate them.

The bar for senior quietly moved

When AI handles the boilerplate, what is left for humans is system design and trade-off judgement. Debugging distributed failures in production. Reviewing AI-generated code for the subtle bugs that confident sounding LLM output can hide. A senior engineer in 2022 usually meant five years of writing solid apps. In 2026, it increasingly means someone who can architect a system, lead an AI-augmented team, and own the parts of the codebase that a model can produce but cannot yet be trusted to decide. The title stayed the same while the job description changed underneath it, which is why companies sitting on stacks of applications still cannot fill the role.

The pipeline problem

There is a slower-burning issue underneath the senior shortage. Seniors do not appear from nowhere. They start as juniors, work on real systems, make real mistakes, and learn from people who have been there. A company that stops hiring juniors in 2025 because AI is faster and cheaper does not have a senior bench in 2030. The same firms cutting entry-level headcount today will be writing increasingly urgent "Senior Engineer" job descriptions in a few years. Stanford data already shows employment for developers aged 22 to 25 has dropped nearly 20 percent from its peak. The pipeline that produces tomorrow's seniors is already breaking.

What to actually do about it

Diagnosis is the easy part. Juniors trying to break in and companies trying to hire are both playing by a 2022 playbook in a 2026 market. Both sides have specific, available moves.

For juniors, how to break in anyway

The hiring market collapsed for one specific kind of junior, the one whose main edge was being cheaper and faster to onboard than a senior. That edge no longer exists, because an AI tool plus one senior is cheaper still. The juniors getting hired in 2026 are competing on something else, and the patterns are visible across teams that are still hiring.

The first is provable AI fluency built on real fundamentals, not surface tool use. Anyone can install Cursor, Copilot, Claude Code, or Codex and ship a working take-home. The candidates who clear the screening understand why the tools behave the way they do. That means knowing tokens, embeddings, and sampling well enough to predict where a model will go wrong. Knowing why hallucinations happen (no grounding, training-distribution gap, ambiguous prompt) and what reduces them. Knowing why a RAG retrieval missed (chunk boundary, embedding model mismatch, top-k cutoff) and how to fix it. And reading an evaluation report without staring blankly at the numbers.

The candidates getting hired have public artefacts that prove both layers. A GitHub repo with iterative commit history. A blog post walking through a real bug the AI got wrong and how they caught it. Or a video walkthrough of building something non-trivial end to end, where they explain the prompts they used and why the model failed twice before it worked. Show the messy middle. The failed approaches, the rewrites, and the moments the model was confidently wrong and knowing the mechanism is what told you why.

The second is specialising toward what seniors are scarce in. The senior shortage is concentrated in a handful of areas. LLM evaluation and agent evaluation (trajectory scoring, tool-use correctness, multi-step task success). Retrieval and RAG systems. Agentic AI and multi-agent orchestration - LangGraph, CrewAI, AutoGen-style state machines, and the inter-agent handoff and trust-boundary problems that come with them. Then guardrails and AI safety - input and output filtering, jailbreak detection, PII redaction, and content moderation. Then fine-tuning pipelines. Then LLMOps and inference serving - vLLM or TGI deployment, rollback, latency and cost budgets, and model routing. And finally observability for probabilistic systems, security review of AI-generated code, and AI-aware infrastructure. A junior who learns one of these well, even at a basic level, is competing in a much smaller pool than junior full-stack. You will not be the best person in the room, but you will be one of a handful of people who can have the conversation at all.

The third is working in public. The signal-to-noise problem cuts both ways. Cold applications drown in the same noise everyone else generates. Public output (open-source contributions to projects companies actually use, technical writing that gets shared, talks at meetups, answers in niche forums) is the strongest signal a junior can send. One serious pull request to a project a hiring manager already trusts is worth more than fifty applications.

The fourth is entering through an adjacent door. Pure software engineering roles for juniors have collapsed. Adjacent roles that still hire juniors have not. SRE, platform engineering, data engineering, security engineering, developer relations, support engineering, customer engineering, and junior AI or GenAI engineering roles are all easier paths into senior software work today. For the AI roles the bar is often one shipped RAG, agent, or fine-tuning project rather than five years of CRUD experience, because the role itself is too new to have a five-year bar. The internal transfer into engineering from one of these roles, twelve to twenty-four months in, is far more achievable than breaking in cold.

The fifth is interviewing for the new bar. The take-home and Leetcode-screen interview is dying because it cannot distinguish AI-completed work from real work. The interviews that have replaced it focus on judgement, architecture, and debugging. Prepare by walking through real systems out loud. Take an open-source project you have read, explain why it is built the way it is, and where it would break under load. Most juniors cannot do this. The ones who can stand out within thirty minutes.

For companies, how to stop breaking the pipeline

Companies that need a senior bench in 2030 cannot stop hiring juniors in 2026. The cheapest fix is restructuring entry-level work around AI rather than eliminating it. A junior paired with a senior and an AI tool can take on production work earlier than the pre-AI playbook allowed. That needs two things, a senior with the bandwidth to review, and a company that treats the junior as a contributor from week one rather than a multi-year investment.

The interview problem is fixable too. Replace AI-defeatable take-homes with live architecture discussions, pair-programming sessions where AI use is explicitly allowed, and code review exercises on AI-generated code with planted bugs. Teams that have moved to this format report that they can tell candidates apart again inside a single conversation.

The last piece is putting AI fluency into the role definition in writing. Specify the tools, the workflows, and the review habits the role requires, then assess them directly. Vague AI literacy questions are why the 75 percent assessment fail rate exists. Specific questions, like how a candidate would debug a hallucinated API call or diagnose a RAG retrieval failure, separate real users from prompt-only users immediately.

AI raised the bar for engineering work, made the day to day harder, and changed what hiring screens for. The companies and candidates who adapt to the new bar will find each other. The ones still running the 2022 playbook will keep complaining about a shortage that is mostly a process failure.

Sources

Sources for every statistic cited above.