Behind the tools everyone uses and nobody really looks at.
A company publishes a job posting.
The role requires “leadership”, “change management”, strong execution capabilities, strategic thinking, several years of experience in complex environments. A few minutes later, a candidate copies the job posting into ChatGPT along with their resume and writes:
“Can you rewrite my profile to better match this role.”
Then a conversation begins.
“I don’t see the PMP certification.”
“I don’t have it.”
“You still managed several complex projects. We can reframe your experience to place more emphasis on cross-functional coordination, deliverables and program management.”
The resume shifts slightly, not enough to become false but just enough to become more compatible with the system that will read it.
And slowly, something begins to move.
For a long time, recruiting still relied on a relatively simple relationship: one person trying to describe their real work to another person, with all the limitations, omissions, awkwardness and simplifications that naturally come with that process.
Today, multiple successive layers of systems are beginning to rewrite, optimize, compare and interpret those representations before a human being ever truly looks at the work behind them.
The candidate learns which words increase the chances of appearing in search results. The AI learns how to reformulate experience to improve ATS compatibility. The system then extracts the formulations closest to the job posting. Another AI may summarize the “most relevant” profiles before a recruiter even opens the applications.
And at every step, language becomes slightly less a description of real work and slightly more an interface of compatibility between systems.
The strangest part is that nobody is necessarily lying.
The candidate simply wants to remain visible, the recruiter simply wants to manage an impossible volume of applications, the AI simply wants to improve the match and the system simply wants to compare readable elements.
Everyone acts logically according to their mandate.
And yet, the more compatible the representations become with one another, the harder reality becomes to see.
Two candidates can now produce almost identical profiles while having gone through completely different professional realities. One person may have led a complex transformation for several years while another mainly learned how to reformulate their experience correctly, and both can progressively begin producing very similar signals for the systems reading them.
The ability to pass the filter slowly begins separating itself from the ability to actually do the work.
And the shift becomes almost invisible because recruiters themselves progressively learned how to work inside the limits of the system instead of seeing beyond it, keywords, standardized titles, shortlists, lexical matching, while part of the real work disappears precisely because it resists simple categories.
You still believe you are reading profiles while you are often looking at representations already optimized to produce sufficient compatibility with other representations.
For years, technological limitations made that simplification almost inevitable. Systems could not truly reconstruct complex context. So humans learned to simplify their language in order to become compatible with them.
The difference today is that systems are finally beginning to function differently.
With AI, it progressively becomes possible to reconstruct architectures of experience, connect trajectories that do not carry the same titles, understand context instead of simply comparing words, detect comparable professional realities even when the formulations used are completely different.
And yet, many organizations still use this new capability to accelerate the exact same old mechanics, faster matching, faster sorting, faster shortlists, while systems are finally beginning to do something else entirely: reconstruct context, connect complex experiences, detect human coherences that go beyond keywords and bring investigation back into recruiting. The problem is no longer truly technological. For the first time, the tools are beginning to have the capacity to work with far more intelligence and nuance than previous systems ever could. The question becomes much simpler and much more uncomfortable: do we actually want to understand human work more deeply, or do we simply want to accelerate the old methods of sorting even faster.
The real always leaves traces.
