Is a portfolio or resume still worth building when AI can produce a plausible one for anyone?

The shift

Producing the artifact — a tailored resume, a cover letter, a case study, a sample design, a code repo, a writing clip — goes from costly and skill-gated to free and instant. The document stops being a hard-to-fake proxy for the work behind it, because the correlation between "this looks like someone capable made it" and "someone capable made it" is what AI severs.

The axioms

• A polished resume or cover letter is a proxy for capability, because assembling a credible one used to require having done the work.
• A portfolio artifact — case study, design file, writing sample, code repo — proves capability more directly than a resume, because producing a convincing artifact required the underlying skill.
• Screening on artifact quality is efficient, because quality correlated with the effort and ability behind it, so it was a cheap first filter.
• The document is a pointer; the scarce referent is the actual track record — what shipped, where, with whom, verified.
• A reference or a vouch carries weight because it stakes another person's own reputation on the claim.
• Work demonstrated under observation — a live exercise, an on-site trial, a paid project — shows capability that can't be pre-fabricated.
• Building a portfolio is worth the candidate's time, because the artifact is scarce enough to be the thing that gets you seen.

Invalid axioms

1. A polished resume or cover letter is a proxy for capability. Fluent, tailored, error-free application material is now free for everyone, so its correlation with the underlying ability drops toward zero. The habit-trap: teams still read "well-written, well-formatted, clearly tailored to us" as evidence of a careful, capable applicant, and still coach candidates to spend their effort there.
2. A portfolio artifact proves capability more directly than a resume. This is the sharper casualty. A case study, a landing-page mockup, a short writing sample, a self-contained code snippet — AI now produces convincing versions of all of these without the author having the skill they imply. The habit-trap: portfolios and take-homes are still treated as the "show, don't tell" gold standard, precisely the tier most exposed now, because a static polished artifact submitted asynchronously carries no proof of who made it.
3. Screening on artifact quality is an efficient first filter. It was cheap because quality was a reliable-enough stand-in for effort and ability. When quality is free, the filter selects for access to a model, not for the trait it was meant to detect. The habit-trap: pipelines still front-load resume-quality and portfolio-polish scoring as the cheap early gate, spending real reviewer attention on a signal everyone can now max out.
4. Building a polished portfolio is a high-return use of a candidate's time. The marginal artifact was worth producing when producing it was scarce and it was how you got seen. Now the same artifact is one prompt away for every other applicant, so the polished-artifact-as-differentiator thins. The habit-trap: candidates still pour weeks into portfolio polish that no longer separates them, instead of into things that do (shipped work with provenance, a reference who'll pick up the phone).

Unchanged axioms

1. The scarce thing was always the track record, not the document. What actually shipped, at which company, with which outcomes, alongside whom — that referent didn't get cheaper to fabricate credibly. AI makes the pointer free; it doesn't make the thing pointed at real. The resume was always a lossy compression of this, and its collapse just exposes that the referent was doing the work.
2. A reference or a vouch stakes someone else's reputation. A named person willing to be called and to attach their own standing to your claim is expensive to fake — it requires a real relationship and someone with something to lose. This is trust and accountability, not synthesis, and it doesn't abundance away.
3. Work demonstrated under observation still shows capability. A live working session, a paired exercise, a paid trial project, a conversation where you have to defend and extend the thing in real time — capability shown under watch, without a screen to route answers through, remains hard to pre-fabricate. (Calibrate: this holds for observed and interactive work; anything submitted asynchronously through a screen is now inside the INVALID bucket, and the line between the two is where the whole system is moving.)
4. Provenance — verified, attributable evidence a specific person did specific work — stays scarce. A commit history tied to a real identity, a shipped product with your name on it that others can corroborate, a talk people attended: the value is in being verifiable and attributable, not in being polished. This is the ground-truth verification AI can't self-certify.

New axioms

1. When any artifact can be AI-produced, the signal is no longer the artifact but its provenance. The field has no settled, low-friction way to prove "this specific human actually did this specific work" — no widely adopted equivalent of a verified track record that travels with the candidate. Building that provenance layer is an open problem, not a shipped feature.
2. Harder-to-fake evidence doesn't scale the way resume-screening did. Work trials, live exercises, and reference calls are exactly the things AI can't fake — and exactly the things that cost real hours per candidate. The old funnel filtered thousands cheaply on the now-worthless signal; the signals that survive are expensive per head, so the funnel's economics break at volume.
3. Both sides running AI collapses the async pipeline to mutual noise. Candidates AI-generate artifacts; employers AI-screen them. When both automate, the entire paper-and-portfolio stage carries near-zero mutual information, and nobody has established what the async top-of-funnel is even for anymore.
4. The candidates who lose most are the ones for whom the portfolio was the on-ramp. Career-changers, early-career, no-network applicants used a strong portfolio to substitute for a track record they didn't have yet. If verified track record and being-vouched-for become the surviving signals, the ladder pulls up behind the already-connected — an access problem the artifact used to soften.

Where it breaks

"The portfolio is the gold-standard 'show, don't tell' signal" (invalid) collides head-on with "the surviving signals — trials, references, provenance — don't scale" (new): teams that have correctly stopped trusting the async portfolio have no cheap replacement gate, so either the top of the funnel becomes uninformative noise or hiring quietly shifts to referrals and trials that only fire for candidates who are already known — moving the bottleneck from "can you make the artifact" to "can anyone vouch for you," and pricing out exactly the people the portfolio was invented to give a shot.

ai-axioms.vercel.app/axioms/society/portfolio-resume-signal-ai-axioms