What's a resume even for when both sides use AI to write and read it?

The shift

Writing a tailored, fluent, role-matched resume and extracting fit from one both drop to near-free. The document becomes an AI-to-AI handshake — machine-drafted on one side, machine-parsed on the other — that the human on neither end necessarily wrote or read, so the artifact itself carries almost none of the information it used to.

The axioms

• A resume is a self-authored summary that stands in for the person until an interview — the cheapest way to preview a candidate at scale.
• The cost of producing a good resume (writing, tailoring, formatting, error-checking) signals effort, literacy, and how much the applicant wants this specific job.
• A reader extracts signal by reading the document — the resume is where fit is first judged, so it has to be legible to a human eye.
• Resume conventions (keywords, reverse-chron format, one page, action verbs) exist because a human parser is slow and scarce and needs the information pre-arranged.
• The resume is the gating artifact: it's the entry ticket that decides who gets a first conversation.
• Claims on a resume are taken provisionally on trust — nobody verifies most of them, on the bet that fabrication is effortful and gets caught later.
• Tailoring a resume to a specific posting demonstrates genuine, targeted interest, because doing it well cost the applicant real time per application.

Invalid axioms

1. The cost of producing a good resume signals effort and how much the applicant wants this job. Fluent, tailored, keyword-aligned, typo-free copy is now free per application; production cost no longer tracks effort or interest. The habit-trap: screeners still read polish, tailoring, and "clearly customized to us" as proxies for a candidate who cared enough to try.
2. A reader extracts signal by reading the document, so it must be legible to a human eye. Neither end reliably involves a reading human anymore — a model drafts it, an ATS or model parses it. The habit-trap: teams keep spending on resume-writing UX, formatting rules, and human first-pass reads for a document whose two actual users are both machines.
3. Resume conventions exist because a human parser is slow and needs information pre-arranged. One-page limits, action verbs, and reverse-chron layout were compression for a scarce human reader; a model ingests the raw record just as well unformatted. The habit-trap: ATS keyword scoring and format-compliance rules persist as gatekeepers, so applicants now optimize the document for a parser instead of a person — and AI does that optimization for them for free.
4. Tailoring to a specific posting demonstrates targeted interest. Per-application customization was the costly tell that separated a real applicant from a mass-sender; it's now a checkbox in the generation prompt. The habit-trap: "clearly tailored to the role" survives as a positive screening signal after the cost that made it meaningful is gone.

Unchanged axioms

1. Someone has to stand behind the claims in person, and the interview is where the accountable version of the story gets told. A model can write "led a team of eight through a replatform"; only the candidate can be pressed on it live and be caught if it's hollow. The document deflates; the accountable, in-person account of one's own work does not.
2. Verifying that the claimed history is real stays scarce. Employment dates, titles, credentials, and whether the work actually happened need ground-truth checking against an external source — a model can draft the claim and can't self-certify it. AI makes fabricated history cheaper to produce, which raises the value of the check, not lowers it.
3. A reputation and a real referral still carry weight the document can't manufacture. "A person I trust vouches for this candidate" rests on a relationship and a reputation someone is spending — not on tokens. That signal moves in the opposite direction to the resume: as the document deflates, the vouch inflates.
4. Demonstrated capability on real work still discriminates between candidates. What someone can actually do — shown in shipped work, a work sample done under observed conditions, a reference to concrete output — remains scarce information the résumé only ever gestured at. (Fast-moving: "unsupervised take-home as proof of capability" is already eroding, since AI assistance on anything done off-camera is now hard to rule out — the signal survives only where the work is verified or observed, not merely submitted.)

New axioms

1. When the document is machine-written and machine-read, what is it actually transmitting? If the intended reader is an ATS/model and the author is a model, the resume is two systems negotiating over keywords with the human's history as raw material — and no one has established what, if anything, that handshake still tells a hiring team about the person.
2. Signal has migrated off the artifact, and the pipeline isn't rebuilt around where it went. The load the resume used to carry — proof of effort, interest, capability, honesty — now lives in verified credentials, referrals, and observed work samples. Which of those the field standardizes on as the new entry ticket is unsettled.
3. Who bears the cost when the entry ticket stops meaning anything? The resume still gates the first conversation while no longer discriminating, so the cost lands somewhere — on applicants who can't tell what to send, on screeners drowning in indistinguishable documents, or on good candidates filtered out by a parser tuned for a signal that's now noise. Nobody owns that reallocation.
4. Detecting a fabricated-but-fluent history is a live problem when generation is free and per-application. A model will confidently produce a plausible role that never existed; distinguishing an embellished machine draft from an accurate one, at intake volume, is not a solved practice.

Where it breaks

"The resume is the gating artifact — it decides who gets the first conversation" (invalid) collides with "the signal has migrated off the document onto verified work, referrals, and reputation" (new): the pipeline still routes every candidate through the one artifact that no longer discriminates, while the information that would actually sort them sits in channels the funnel doesn't read at the gate. A team that keeps the resume as the filter is screening on the emptiest object in the process and deferring the real signals — the vouch, the credential check, the observed sample — until after the gate that was supposed to use them.

ai-axioms.vercel.app/axioms/hr/recruiting-resume-signal-ai-axioms