Why finding the right job has never been harder — and why the answer might not be a better filter, but a better imagination
There is a particular kind of despair that sets in around the fourth week of a serious job search. You have updated your LinkedIn headline three times. You have tailored your résumé to the point where it no longer feels like yours. You have applied to roles you were overqualified for, underqualified for, and perfectly qualified for — and heard back from almost none of them.
The frustrating part is not the silence. The frustrating part is the sneaking suspicion that the right job exists. You just can't find it.
_This is not a personal failure. It is a structural one_
The Matching Problem Is Older Than the Internet
For decades, the dominant theory of job searching was essentially a logistics problem: get your information in front of the right people. The newspaper classifieds gave way to Monster.com, which gave way to LinkedIn, which gave way to an ecosystem of platforms, aggregators, and ATS systems so complex that entire consultancies now exist to help candidates navigate them.
But more pipework has not solved the underlying problem. If anything, it has obscured it.
The core dysfunction is this: job seekers search within the boundaries of what they already know they are looking for. We type in our last job title. We filter by industry. We scan the first two pages of results and, finding nothing that resonates, conclude that the market is bad. What we have actually done is searched a very small corner of a very large space — and called it thorough.
Hiring, viewed from the other side of the table, suffers from the mirror image of this problem. Recruiters write job descriptions that describe who they had last time, not who they need next. They filter resumes using keyword systems that reward people who know which words to use, not necessarily the people who can do the work. Both sides are searching for each other using maps drawn from memory.
The Vocabulary Problem No One Talks About
There is a concept in information retrieval called the vocabulary mismatch problem: the words a user uses to describe what they want are often not the words a database uses to describe what it has. In job search, this mismatch is catastrophic — and deeply personal.
A solutions architect with six years of enterprise field experience might never think to search for "technical customer success" or "value engineering" or "AI solutions consultant" — roles that would suit them precisely, roles that are actively hiring, roles that simply don't appear in the mental model they carry into a search box.
_The skills transfer. The language doesn't_
We are, in other words, limited not by what we are capable of, but by what we can imagine ourselves doing. And imagination — particularly about one's own professional identity — turns out to be a surprisingly narrow resource when you're under the pressure of an active search.
If You Want One Good Idea, You Need a Hundred
There is an old principle in creative problem-solving, attributed variously to Linus Pauling and Alex Osborn, that the way to have a good idea is to have many ideas. Quantity, counterintuitively, is how you find quality. You cannot edit your way to an insight you never generated in the first place.
Job searching has never had a version of this. There has been no mechanism for systematic idea generation at the top of the funnel — no way to ask "what else might fit me?" and get a serious, considered answer back.
Until now, possibly.
The LLM as Career Mirror
Large language models are not magic. But they do one thing with unusual power: they hold an enormous, associative map of human work — its titles, its functions, its adjacencies, its history — and they can traverse that map in ways that keyword search cannot.
Ask a language model to reason about a person's experience, and it will not return the ten most popular jobs with a matching keyword. It will reason about transferable patterns. It will surface roles the candidate never considered, roles that existed before they started searching, roles in adjacent industries where their particular combination of skills would be genuinely rare and valuable.
This is not personalization in the shallow sense — showing you more of what you already clicked on. This is expansion. It is the difference between a search engine and a thinking partner
An Experiment Worth Watching
A new platform called kumiin.io is testing exactly this proposition. The premise is deceptively simple: rather than asking candidates to search, it asks them to be understood — and then surfaces jobs they would not have found on their own.
The design philosophy is rooted in the "hundred ideas" principle. Most of what the platform surfaces won't be right. Some of it will seem strange. But somewhere in the noise is a signal — a role, an industry, a function — that the candidate had genuinely never considered, or had considered years ago and filed away. The platform's bet is that surfacing that possibility, even once, is worth the exercise.
It is early. But the underlying insight is sound: the bottleneck in job matching is not information volume. It is conceptual range.
_We know more about what we've done than what we could do.
We search in the past tense when the opportunity is,
by definition, in the future_
What This Means for Talent Strategy
For HR leaders and talent acquisition professionals, the implications extend beyond the candidate experience. If the best hires are the ones who bring capabilities an organization didn't know it needed, then the hiring processes optimized entirely around job description matching are selecting against exactly those people.
The homogenizing pressure of keyword-based ATS systems, combined with candidates who search within narrow self-defined lanes, creates a market that looks efficient while missing enormous amounts of value on both sides.
Better matching is not just good for candidates. It is a competitive advantage for organizations willing to hire based on potential rather than precedent.
The Search Box Was Never the Answer
The job market does not have a data problem. It has a translation problem — between what people can do and how work gets described; between who someone has been and who they might become; between the roles that exist and the imagination needed to find them.
_Language models, used well, are translation engines.
They don't just retrieve.
They interpret, reframe, and expand_
The résumé is not broken. The search is. And for the first time, there is a tool capable of searching the way a great career advisor would — broadly, associatively, and without the constraint of what you already know to ask for.
That is not a small thing.
A new platform called kumiin.io is testing/experimenting exactly this proposition.*
If you're building something solo, figuring it out as you go, or just want to say hi — I'd love to hear from you. Find me at humiin.io
Top comments (0)