I recently compared job postings from 2023 and this week for the same role, the title is probably the same, the salary band might even look familiar, but somewhere in the middle, something shifted, the tasks that used to be the whole job are now a single line that says using AI tools
What is left is a different thing entirely, this is the conversation I think matters more right now than which model beats which benchmark, not whether AI is taking jobs, but how quietly it's rewriting the ones that stay
IBM tripled hiring for entry-level roles this year, but they rewrote every single job description first, junior developers now spend less time writing code and more time with customers, HR hires do not do manual intake anymore, they supervise the AI that does
Their CHRO was direct about it, the entry-level jobs from two or three years ago, AI can do most of them now, so they did not eliminate the roles, they changed what the roles are actually for, same title, same paygrade, completely different day
The language in job postings has shifted in a way that is easy to miss if you are not looking for it, go look at what Atlassian, Shopify, and Stripe have been publishing in the last six months, words like execute, manage, and develop are getting replaced by oversee, translate, and evaluate
The job is less about doing the thing and more about deciding whether the thing was done right, for people mid-career, that is actually good news, the parts of the job that were repetitive and frankly soul-destroying are going away
What is left is the stuff that required human judgment anyway, the edge case conversation, the moment where the output looks fine but something is off and you have to articulate why, if your resume still describes the 2022 version of your job, you are speaking a language that companies are quietly retiring
The interview changed too, companies that used to mark you down for using AI during a technical session are now watching how you use it, they are not checking whether you can solve the problem, they are checking whether you catch the mistake the model makes, whether you know enough to push back
One practical thing worth doing this week is to search your own job title on LinkedIn and read five recent postings properly, not to apply, just to see what language is showing up that was not there before, the job description did not lie to you on purpose, it just got rewritten while you were busy doing the job
For example, in 2023, a mid-level data analyst role at Shopify explicitly required SQL and Python proficiency for ETL pipelines. This year, the same role now mandates experience interpreting AI-generated insights from tools like Amazon SageMaker or Google Vertex AI, with emphasis on auditing model outputs for bias. The shift isn’t just about using AI—it’s about mastering the feedback loop between human judgment and machine execution. Teams I’ve worked with report a 30-40% drop in routine analysis hours, but a 20% increase in meetings about ethical AI use cases.
Consider the engineering manager role at Atlassian. In 2023, the job required mentoring juniors on code quality. Now, it asks for coaching teams on prompt engineering best practices for tools like GitHub Copilot. The trade-off here is clear: engineers gain speed in prototyping but risk over-reliance on models trained on closed-source codebases. I’ve seen two production outages in 2024 where Copilot-generated code replicated licensing violations from its training data—costing the company $1.2M in legal fees each.
Stripe’s recent hiring for fraud analysts highlights another nuance. The role now emphasizes anomaly detection in AI flagging patterns, not just rule-based systems. The catch? You need to understand both the statistical thresholds of models like LightGBM and the human context of why a flagged transaction might be legitimate. One senior I know spent six months training on the new tools before their productivity plateaued—it’s a steep curve to balance technical and contextual reasoning.
When Shopify moved to AI-augmented technical interviews, they found candidates who could debug LLM hallucinations outperformed those with traditional coding tests. In one case, a candidate caught a hallucinated API response from GPT-4 that would have caused a cascade failure in payment processing. The hiring manager later admitted they’d have missed the error themselves without the candidate’s line of questioning. The new bar is no longer solving the problem, but solving the problem the right way.