A few weeks ago I saw a post on LinkedIn from someone I used to work with. He is very good at what he does. The post said he and about a hundred other people on a government project had been told, without notice, that the work was over and they were done. No transition. No runway. They went home that day and woke up the next morning looking for what came next.
I have seen some version of that post more times than I can count in the last year. The names change. The agency or the contractor changes. The project changes. The pattern does not. Capable people, most of whom built their careers before AI was anywhere near the hiring process, are being moved without warning into a job market they have not seen in a long time.
I keep thinking about what they are about to walk into.
Most of them learned how to look for work in a world where, at some point, another person would read their resume. They learned to choose a clean font. To keep the formatting consistent. To tailor a cover letter to the company. To make sure the resume told a human story. They learned how to be seen by a hiring manager.
That world is largely gone. The reader on the other side of a job application today is almost never a person, at least not first. It is software. And the software does not care about your font.
Before we go any further, one word. Because if we do not understand this word, the rest of the story will not land the way it should.
That word is algorithm.
An algorithm is a set of instructions that tells a computer how to make a decision. That is all it is. When a streaming service recommends a show you end up watching for three hours, an algorithm decided that. When your bank flags a purchase as potentially fraudulent, an algorithm decided that. When your navigation app reroutes you around traffic, an algorithm decided that.
An algorithm is not magic and it is not neutral. It is a set of rules, written by people, learned from patterns in the past, applied to make a decision faster than a human could. When those rules are applied to something that matters in your life. Whether you get a job, whether you qualify for a loan, whether your child is admitted to a program. The algorithm is making a consequential decision about you, often without your knowing it is happening.
The Case That Made It Visible
To understand what the people I just described are about to encounter, it helps to look at one of the first cases that made this system visible to the public.
On a weekend night in 2017, a man named Derek Mobley looked at his phone and saw a rejection email. It had arrived at 1:50 in the morning. Less than an hour earlier, he had submitted a job application through a platform called Workday.
Mobley was an African American man in his forties with a degree from Morehouse College and nearly two decades of experience across finance, IT, and customer service. He had been laid off the year before. He had applied to more than 150 positions and had not received a single interview.
He called the moment a watershed. Not because of the rejection. Because of when it arrived. No human recruiter was awake reviewing applications on a weekend night. The decision had been made by a computer. The algorithm had taken his resume, compared it against a learned pattern of what a good candidate looked like, decided he did not qualify, and pushed the rejection out to him in under an hour.
In February 2023, Mobley filed a federal lawsuit against Workday. The case is still working through the courts. It is the largest AI hiring discrimination case in American history. But the lawsuit is not really the point of this piece. The system the lawsuit revealed is the point. And that system is what the next round of displaced workers is about to meet.
What the System Is Actually Doing
When you submit a job application online today, you are almost certainly not being reviewed by a human first. You are being processed by a hiring algorithm. The pattern is roughly the same whether the platform belongs to Workday, Eightfold, iCIMS, Greenhouse, or any of the dozens of others.
Step one is reading your resume. Not a person reading it. A computer reading it. The system pulls out what it considers relevant: your job titles, your employers, the dates, your education, gaps in your work history, your listed skills. It converts all of that into a data record. Essentially a form with many fields filled in. Everything else, every line of context you wrote to explain who you are, is discarded.
Step two is scoring you. The system compares your data record against a pattern it has learned from studying thousands of past hires at companies like the one you applied to. It assigns you a number based on how closely you match the pattern of people the company hired before. If the company has historically hired people from certain schools, with certain kinds of experience, without gaps in their work history, the algorithm has learned those are the markers of a good candidate. Your score reflects how well you fit that history.
Step three is ranking you against every other applicant. If a company receives five hundred applications for one position, the algorithm lines everyone up from highest score to lowest. The company tells the system to forward the top twenty to a human recruiter. The other four hundred and eighty receive an automated rejection.
Step four is sending the email. That is the 1:50am message Mobley received. Not written by a person. Not reviewed by a person. Generated and delivered because his score fell below the cutoff.
Here is the part most people coming back into this market have not yet absorbed. The system is not reading your resume the way a human would. Font and formatting matter less than they used to. Tone matters less. The shape of the sentences matters less. What matters most is keywords and context. The specific terms that appear, in the right places, in the right relationships, against the specific language of the job posting. A resume that a hiring manager would read as elegant and clear can score poorly inside the system because it does not use the words the model has learned to look for.
At no point in what happened to Derek Mobley did a human being look at his resume and decide he was not qualified. A computer did. And that computer was not applying neutral rules. It was applying rules learned from decades of past hiring decisions made by people, in a country with a long and documented history of discrimination in employment.
Why the Past Follows You
The algorithm does not see your race, your age, or whether you have a disability. It is not allowed to. Those are protected characteristics under federal law. But it does not need to see them directly to discriminate against you.
The algorithm has learned that certain patterns predict success at this company. Those patterns carry hidden information about protected characteristics. People who build these systems call them proxies, because they stand in for the protected information without naming it.
The year you graduated tells the system roughly how old you are. A gap in your work history might suggest you took time off to care for a child or a parent, or that you experienced a health issue. Things that correlate with gender and disability. The zip code you live in correlates with race and income. The name of your school correlates with race and class. The specific technical skills on your resume reflect the era when you were trained, which correlates with age. None of those things are your race, your age, or your disability. All of them carry information about those things.
A system trained on decades of biased hiring decisions will learn to use those signals. Not because anyone programmed it to discriminate, but because that is what the historical pattern shows. This is the part of the work I find most uncomfortable as someone who builds these things. The math is honest. The data is not.
A 2024 University of Washington study ran more than 550 identical resumes through major AI hiring tools. Resumes with names typically associated with Black applicants were ranked lower than identical resumes with names typically associated with white applicants, 85 percent of the time. Same experience. Same education. Same words on the page. Different name. Different score.
A System That Does Not Add Up
A friend told me recently about a conversation that has stayed with me. He was talking with a hiring manager who said, with some heat, that he could not stand resumes written with the help of AI. Inauthentic. Lazy. The candidate was not really showing who they were.
Then, almost as a side note, the same hiring manager mentioned that the company he works for runs every incoming resume through an AI screening tool before any human sees one of them.
That is extreme misalignment.
A company that makes a machine the gatekeeper of every application cannot, in the same breath, fault its applicants for using a machine to get past the gate. The applicants did not write that rule. The company wrote it for them when it deployed the system. Telling the applicant they have failed the spirit of the test while administering the test by machine is not a coherent position. It is a convenience.
For a candidate to not use AI to prepare for a system that uses AI to judge them is folly. The screening tool will favor candidates who match its learned pattern. The applicant who knows how to align to that pattern is doing what the system asks. Refusing on principle is not principled. It is unilateral disarmament.
You cannot automate the judgment and keep the moral authority. Those two things cannot both belong to you.
The Requirement Nobody Posted in the Job Description
There is a related thing happening underneath all of this that almost no public commentary captures, and it is the part I think about most.
Knowing how to use AI well has quietly become a competitive advantage in the job market. Not because companies have started listing it as a requirement, although some have. Because the system that screens you is itself an AI system, and being able to align to it is now part of what it takes to get through.
A high-quality AI-assisted resume is not what most people picture when they hear that phrase. It is not asking ChatGPT to write your resume for you. That produces something generic and easy for a screening system to recognize as generic. A good AI-assisted resume is the output of real work.
Done well, it is typically a synthesis of three to seven documents. Your prior resumes. Your work history in long form, written out at length somewhere only you can see it. The specific job solicitation, read closely. Deep research output on the company, the team, and the role. Sometimes prior writing samples or project documentation. The applicant brings all of that to the AI tool, directs the synthesis, and revises the output until it represents them accurately and aligns to the language of the posting.
That is real work. It takes hours, not minutes. It requires the applicant to know their own history in detail, to understand how a machine will read a resume, and to do the translation between those two things deliberately. It is closer to writing a grant proposal than to drafting an email.
The second skill there, knowing how a machine reads a resume, is not a skill most people who built careers in the last twenty years have been asked to develop. They have spent that time becoming good at the work itself. Now, on the way back into the market, they are being told the work history is not enough. They also have to be fluent in the system that judges them.
This is a real equity problem, and it is not the one most discussions of AI and hiring are paying attention to. Two equally qualified candidates with different levels of AI fluency, or different access to people who have that fluency, can have very different outcomes from the same hiring pipeline. Almost none of that difference would be visible to the company doing the hiring. It would simply look like one candidate scored higher than the other.
That is a hidden cost of building a job market on top of automated systems. It does not only penalize the candidates the algorithm misreads on race or age or disability. It penalizes the candidates who do not yet know how to be read by it.
What This Means for You
If you are looking for work right now, or if someone close to you is, the rules have changed in ways that are not obvious from the outside. The resume that worked ten years ago, or even two years ago, is often not the resume the current system is built to recognize. Knowing this is the starting point. Not knowing it is one of the quietest disadvantages in this market.
If you are about to begin applying for jobs after a long stretch inside one, assume the first reader is a machine. Read the job posting closely and let its language inform your own. Use the AI tools available to you, but do not let them write for you. Synthesize, do not outsource.
The Mobley lawsuit is asking a question that belongs to everyone who applies for work in the United States, not only the people obviously displaced by a screening tool's bias. The question is whether the people being sorted by these systems have any right to know how the decision was made, and any path to challenge it if it was wrong.
In April 2025, an executive order instructed federal civil rights agencies to pull back from cases like Mobley's. The agencies that were supposed to hold these systems accountable have been told to stand down. That leaves the courts, and individual people, asking what Mobley asked at 1:50 in the morning in 2017. Who decided this, and how, and was it fair.
I write about this here because the people who build these systems are not the only ones who should weigh in on how they work. The people the systems judge, including the people about to be moved into a hiring market they did not last see, have just as much at stake.
Those are not technical questions. They are human ones. And they belong to all of us.