LLMs text processing capabilities will remove the limitations needed by traditional resumes allowing for more context and better matching between jobs and candidates.
This is the first article in a three-part series on AI’s impact on the hiring process. This first article focuses on how the resume itself will evolve given the incredible text processing capabilities of LLMs. In the second part, we see how the same technology will change the dynamics of interviewing. The series concludes by understanding how these two changes, along with other newly unlocked capabilities will alter the hiring process. While there are potential issues for bias, if done right, AI can make the entire job market much more efficient benefiting candidates and companies alike.
I remember once when I was a young child, my father was looking for a new job and typing up his resume on a typewriter. Once finished he would make copies and mail them off to companies along with cover letters. To find and apply to a job was significant work, so most people would apply to a limited number of them. Fast forward to today and you can find hundreds of jobs in mere seconds on any number of job websites. The job application itself can be done in seconds, with only a few button clicks.
At first blush, this sounds great for applicants, since things got much easier (ask anyone who ever needed to use Whiteout on a typed resume). Unfortunately, it also created a vicious cycle. Because of the ease of applying, candidates apply to more jobs, including stretch jobs, or jobs for which they might not be qualified. As more people apply, the percentage of people who get an interview goes down. Knowing that the odds are low, candidates are told, “it’s a numbers game, apply to more jobs.” More applicants lower the odds further, so job seekers need to apply to yet more jobs. (See the article “How Many Job Rejections Are Too Many?”)
Moreover, hiring managers, who are now getting mountains of resumes need to use an ATS (Applicant Tracking Systems) to sort (not just manage, but proactively select) from the ever-larger set of resumes. But an ATS is going to filter out resumes just because they don’t have some keyword, lowering the odds that the right candidate will be selected since even the right candidate may not have the ideal combination of right keywords that a less-qualified resume does. Again, this encourages applicants to just apply to more jobs. And the more jobs to which you apply, the less time you have to customize the resume to the job, making it less likely you pass the keyword or other first-pass filter.
Ultimately, the better the software automaton that lowers the cost of applying (time to discover and apply to jobs), the worse it gets for candidates, and for companies. This can’t continue.
Heterosexuals who follow social dating norms (the man makes the outreach to the woman) and use dating apps in major cities like New York have encountered the same problem. A woman (“job poster”) can go on the app and get hundreds of right swipes within a few hours (or on older platforms, scores, or even hundreds of email messages a day). It’s overwhelming. A man (“job applicant”) will know it’s a numbers game and so will treat it as such and focus less on crafting the right message to two or three women who seem ideal, but will instead swipe right as quick as possible, paying limited attention to the profile. As with job boards, it’s a vicious cycle that is discouraging to both sides, following a tragedy of the commons pattern.
Instead, we need to fundamentally shift how job applications work (and, for the record, online dating, too). Here’s how. (I’ll just focus on hiring, but a very similar solution will work for dating.)
First, we need to move off of traditional, short form resumes. The standard resume today is typically only one or two pages and has short sentences or bullet points. It’s a Joe Friday style document, leaving out a lot of color and just focusing on the facts. (Joe Friday never actually said “Just the facts ma’am,” but it’s a line very much associated with him.) Such short resumes are useful for when a hiring manager must quickly review scores of candidates manually and needs a consistent summary of each. Most resumes are scanned for a handful of seconds and by following a standard format the hiring manager can efficiently assess them.
As we move towards more automation using generative AI (LLMs), that constraint is no longer needed. Recruiters today tell candidates that your resume needs to be two pages, but your LinkedIn does not. The reason is because LinkedIn is used for search and the more details and context you have, the more likely you are to be found in a search. We need more long form work records. LinkedIn and similar websites are a start, but not the end. What is really needed is to have additional prose behind each job (and for the candidate overall), providing more context and depth to who the person is.
Ideally, there will be a standard for these extended resumes (so candidates applying don’t have to type their entire work and education history over and over again on each job application). Consider a digital version of your resume file, not simply a text Word doc or PDF, as they are today but a new format. It may be as simple as an XML standard. Within that file there’s a two-page version similar to what you have today with bullet points or a summary for each role. There’s also a way to add more text, such as a few paragraphs, with additional information per role and about the candidate in general. The AI agent can read through all the extra text. The human who eventually interviews you can choose to just get the two-page version for the interview.
Even better, the interviewer (or interviewee) can ask AI to provide a two-page version specific to the role. Or other aspects of the candidate can be easily extracted. It may be something as simple as, “Show me the sales figures this candidate has delivered in his last five roles.” Or perhaps it’s, “Explain the type of leadership challenges she faced in her last three roles and how closely it compares to the one we need our new hire to address.”
This is becoming more important because careers are less likely to follow linear trajectories. Unfortunately, resumes, as with most writing, is linear, organized chronologically, but the relevant parts to the question (in this case, "Should I hire this person?”) may not be a linear path through the resume. (Curious readers can dive more into this concept in the article “The Future is Context Dependent Content.”)
Second, we encourage candidates to have multiple resumes. Some say you should tailor a resume for each job, but that's a bit much. Still, you may have a few versions for slightly different roles or industries. This isn’t about lying on the resume, but having a different angle, emphasis, or story with each one. Any online “resume” (again, LinkedIn, or other job board profile) should allow for different angles. This doesn’t have to be different versions, but maybe sub-resumes from this enhanced resume with more information, or different facets of the enhanced resume. Someone looking for a certain background can start with a resume facet best suited for that role.
Consider someone with both product and product-marketing experience who has worked in manufacturing, health tech, and medical manufacturing. The version of the resume for an industrial manufacturing business would likely be different from the version for a health tech software business. The job history would all be there, but the level of detail for any given job, and what those details are, would be different for each resume facet, oriented to that type of role for which it was designed. You can have many resumes facets, since no one will need to look at all of them (and AI can potentially help you generate them).
In fact, most people won't look at any. Pretty soon it will be AI agents who do the first pass. And this is why the two-page restriction won’t make sense in the future. I’m not saying we should all be writing entire books about who we are, succinctness is a skill, and eventually a human will see your resume (or the future equivalent). However, needing to condense everything you’ve done into a hard space limit won’t matter because there’s no real cost to having an agent read a little more about you in a longer document. We can’t pass 2,000 words to a human and ask her to scan it in a few seconds, but we can ask that of AI. (Yes, there’s some marginal extra cost for AI to process ten pages instead of two, but it’s on the order of pennies and seconds. Monday morning an AI agent can read 1,000 ten-page resumes and have a short list for you by lunchtime, if not sooner.)
Another advantage is that we can now match on more criteria. All too often candidates are rejected because some detail that got cut from the resume knocked them out (often a keyword). Now all details can be included. Even better, important details, like leadership style can start to come across in the resume’s additional prose, which is not used to find matches today. As such many candidates may fit some simple metrics, like years of experience, but not match on culture or other intangibles. (See “The Streetlamp Effect in Hiring” for a more detailed discussion of this challenge.).
In the article “What Luggage Can Teach Us About Innovation” I noted that while most people point to the innovation of wheels on luggage being added only in the twentieth century as an example of missing something obvious, in reality, it’s an example of environmental factors limiting the use of technology. We’ve always been able to create more detailed resumes, but the cost of creation (typing and mailing initially), and processing (a hiring manager reading through scores of resumes), required us to limit the resume. It’s not unlike the Qwerty keyboard. Many argue its design was intentionally inefficient to prevent jams; some researchers suggest it was due to standards from the telegraph industry. Whatever the reason, its continued use stems from historical tradition even though today everyone agrees the Dvorak keyboard is objectively better. We’re generally stuck with the Qwerty keyboard, but we don’t need to be stuck with limited-scope resumes.
Today a search by a candidate across open jobs is generally based on keywords (as well as location and salary). Unfortunately, keywords don’t convey much information. For example, a “Director of Marketing" job at a small company may mean you’re a one-man show. At a Fortune 500 company that could mean you’d oversee a team of twelve. The same title refers to very different jobs. But today’s job alerts will match both roles to a candidate even if she is looking for or qualified for only one of those roles and not the other. Likewise, recruiters looking for a technical candidates may see a keyword like “Kubernetes” (if they don’t miss it altogether because the candidate used the standard industry abbreviation of “k8s” instead) but not know if this was used only for a few months on one project during a five-year role or if it was part of her daily responsibilities over the entire five years. Again, two candidates with the same keyword look very different once we have more information. The extra prose can provide that context allowing for much more targeted searches. Applying can still be easy, but the filtering process can help candidates and companies alike narrow down the option pool, reducing the noise.
I don’t think this will happen overnight. And there will likely be a new standard for how the short (two-page) facet of a resume should be formatted as well as the longer version of it. We can also include other details like a more detailed candidate objective, even getting into specific things like amount of travel, corporate culture, number of days in office, etc. Again, more details aren’t a problem for LLMs. Perhaps the current job boards will evolve to support this (be sure to let them know about this series of articles), or maybe new job boards will rise up to replace them. I do think sometime in the next 10-15 years we will see things change.
(Note: A few days before the first article in this series got posted, OpenAI announced its own job platform. It’s short on details but what they did say made it sound more of a traditional job board that just focused on AI-skilled candidates, rather than a disruption to the job search process using AI.)
One word of warning, bias is well established in AI training data. Legacy bias in hiring will translate to bias in AI reviews of resumes and other parts of an application (see Redlining in the Twenty-First Century: Everything Everywhere All at Once). For this to work well, we need to address this issue.
Technology has lots of benefits but there are often secondary effects. As we lowered the cost of finding and applying to a job, we created new problems as the number of applicants became overwhelming. Thankfully the ability of technology to process larger amounts of information will allow us to better filter, on both sides of the hiring table. To do so, however, we need to break out of the constraint set by old, physical processes. I’ll cover this more in part 2. In 2024 I predicted the decline in cover letters which is already coming to fruition, “RIP Cover Letters 1956 - 2024.” Resumes, you’re next.
It’s critical to learn about corporate culture before you accept a job offer but it can be awkward to raise such questions. Learn what to ask and how to ask it to avoid landing yourself in a bad situation.
Investing just a few hours per year will help you focus and advance in your career.
Groups with a high barrier to entry and high trust are often the most valuable groups to join.
