Prove AI Can't Do This Job. That's Your New Hiring Process.
A hiring manager we work with wanted to bring on a mid level backend engineer last month. Before the req could even go up for approval, she had to write a memo explaining why an AI agent could not do the work instead. That is not a hypothetical anymore.
17% of companies now require exactly that kind of justification before they will approve a new hire. We are seeing it show up everywhere right now, and not always as a written policy. Sometimes it is just the question everyone expects you to answer before a req goes anywhere.
The Memo Nobody Trained Anyone to Write
What this actually looks like inside a company varies. A form, a slide, sometimes just a Slack thread with a VP or a finance partner. But the shape is always the same. Someone above the hiring manager wants a clear answer to one question. What does this person do that the AI tools you already have cannot?
That question is harder than it sounds. Most job descriptions were never written to answer it. They describe responsibilities, tasks, a list of things the role touches. Responsibilities are exactly the part AI is getting good at, which means the old job description does not help you here at all.
Why This Happened Faster Than Anyone Expected
A few years ago this would have sounded paranoid. Then AI coding tools got genuinely good, a real share of implementation work stopped requiring a dedicated person for it, and finance departments noticed.
75% of companies now factor AI capability into a hiring decision in some form. That is not a hiring freeze. It is a new question sitting in front of every headcount request that did not used to be there. The roles have not disappeared. The assumption that growth automatically means more people has.
Most of These Justifications Fail for the Same Reason
We talk to a lot of hiring managers working through exactly this right now, and the same mistake keeps showing up. People default to defending the job title instead of the work. Senior engineer, five years of experience, knows the stack. None of that answers what the person reviewing the request actually wants to know.
The ones that work do something different. They point at a specific moment where a judgment call has to be made under real pressure, and they explain why that call cannot be handed to a model. A production incident at two in the morning where three systems are giving contradictory signals. A client who changes the requirement halfway through a build and needs someone to push back, intelligently, in real time. A legacy system where what the documentation says and what the system actually does have quietly drifted apart. None of that is a task. That is the actual reason the role exists.
What Convinces Someone, in Plain Terms
If you are the one writing this memo, stop listing what the role does and start naming what goes wrong when it is done badly. AI gives a confident, wrong answer and nobody catches it before it ships. A commitment gets made to a customer that the system underneath cannot actually support. The one person who remembers why a decision was made two years ago leaves, and nobody else can explain it. Each of those has a name. Put the name in the memo, not the job description.
This is also the real test hiding inside the phrase AI fluency, and most companies are still testing for the wrong version of it. The question was never whether someone can operate the tools. Almost everyone can, at this point. The real question is whether they know when to stop trusting what the tool just told them, and that instinct does not show up on a resume no matter how it is worded.
This Is Not Going Away, So Get Good at It
Treating this as a temporary annoyance is the wrong read. The companies that get fast and specific at making this case are going to out hire everyone still writing vague justifications, simply because they will know exactly what they are asking for before the req ever goes up.
We build this case with hiring teams constantly, on both sides of it. We help identify what a role genuinely requires that AI cannot cover yet, and we place engineers who have already proven they can carry that judgment instead of just describing it in an interview. If you are staring at a headcount request you cannot quite justify yet, let’s talk.