# Sk · Verified Skill — element 55 of 58 > Proof you can do the thing. Turns claimed ability into credentialed ability. - **Group:** 12 · Human Elements - **Necessity:** Optional - **Price band:** $ · under $30/mo - **Maturity:** Emerging - **Edition:** v2026.Q3 · verified 2026-09-13 ## Leading tools (v2026.Q3) - **Public portfolio + receipts** — receipts beat résumés - **Live human technical assessment** — a human watches you work - **Vendor AI certifications** — cheap, proctored, recognized - **LinkedIn Verified Skills** — verified by usage - **Portable credential rails** — the rails, not the signal ## Our take In an AI-slop labor market, verified human skill is appreciating. Expect this element to matter more every year. ## Combines with Fx, Ts ## The top 5 — deep dossier (verified 2026-09-13) A public portfolio with receipts still wins — shipped product, live URL, paying customers, a repo history with your fingerprints on it — because it is the only route employers already know how to read and it costs nothing but the work. But 2026 broke its monopoly: 67% of HR leaders say AI-generated applications are slowing their hiring and 65% say AI-polished resumes have made skills verification harder (Robert Half, n=2,000+, released Mar 10, 2026), and Gartner expects one in four candidate profiles worldwide to be fake by 2028. So the portfolio now needs a second, adversarial layer. A live human technical assessment (Karat-class interview engineering) is the highest-verifiability proof anyone can buy, and it's free to you because the employer pays. Vendor AI certifications — AWS AI Practitioner $100, Google Generative AI Leader $99, NVIDIA NCA-GENL $125, Claude Certified from ~$99 — are the cheapest credible floor when you need a recognizable token fast, with the caveat that Microsoft just retired AI-102 and AI-900 outright on Jun 30, 2026. LinkedIn Verified Skills (launched Jan 28, 2026 with Descript, Lovable, Replit and Relay.app) is the interesting new shape: verification derived from actual tool usage rather than an exam — free, unfakeable in principle, and completely unproven in practice. Credly and Open Badges 3.0 are the rails underneath all of it, not a signal on their own. On the thesis — that verified human skill appreciates as AI-generated claims inflate — the evidence is real but asymmetric: the inflation is thoroughly documented, the premium is not. 1. **Public portfolio + receipts** (you (GitHub, a live product, named customers)) — Free · GitHub Free gives unlimited public repos · a domain ~$12/yr · the real cost is shipping something. Best for: Anyone who has actually built and shipped — the highest-recognition, lowest-cost proof, and the only one that also functions as marketing. Why: It is the route every hiring manager already reads without a decoder ring, and the substrate is enormous: 180M+ developers on GitHub, 36M of them joining in the 12 months to Oct 2025, across 395M public repositories (+19% YoY). Crucially it carries information no exam does — commit cadence, how you handle a bug report, whether the thing has users. LinkedIn's own framing of its 2026 verification push was that employers "want to know what you can actually do," which is a portfolio argument even when LinkedIn is selling something else. Watch: This is the entry AI degraded most. A convincing repo, README, landing page and demo video are now a weekend of agent time, and 65% of hiring managers say AI-enhanced applications have made skills verification harder (Robert Half, Mar 2026). Depth and provenance are the defense — long commit history, real users, public writing, and the ability to be interrogated live about a decision you made 14 months ago. A portfolio you cannot defend in conversation is now a liability, not an asset. [https://github.com](https://github.com) 2. **Live human technical assessment** (Karat (+ Byteboard) · CodeSignal · interviewing.io) — Free to you — the employer pays. Employer side: CodeSignal Hire from $79/mo annual (60 credits) · $479/mo Grow (420 credits) · $20/credit overage; Karat is enterprise-quoted. interviewing.io is free for candidates; Karat's Brilliant Black Minds gives free mock interviews.. Best for: Converting a claim into an observation — a trained interviewer watching you reason in real time, which is currently the only proof AI cannot sit in for. Why: It is the highest-verifiability route that exists, and 2026 made it structurally more valuable rather than less: 41% of enterprises surveyed say they have already hired and onboarded a fraudulent candidate (GetReal, n=668 security and risk leaders, Sep 2025), and 38% of employers are responding by adding interview rounds (Robert Half, Mar 2026). Karat has run 600,000+ technical interviews and serves 7 of the 25 largest public companies; its own guidance is blunt that live human interviews remain the strongest safeguard, because the tells are behavioral — instant optimized solutions with no iteration, explanations that don't match the code. Watch: You cannot buy it for yourself — it is employer-procured, so as a founder you're on the wrong side of the transaction unless you use the free channels. It's also the most expensive route in wall-clock time and the most stressful, and the anti-cheat arms race is live: an entire cottage industry of overlay tools now markets itself specifically against Karat, CodeSignal and Triplebyte assessments. Results are rarely portable across employers. [https://karat.com](https://karat.com) 3. **Vendor AI certifications** (AWS · Google Cloud · Microsoft · NVIDIA · Anthropic · Databricks) — Per exam: AWS AI Practitioner (AIF-C01) $100 · AWS ML Engineer Associate (MLA-C01) $150 · Google Generative AI Leader $99 · Google Professional ML Engineer $200 · NVIDIA NCA-GENL $125 · Claude Certified Associate ~$99, Developer/Architect ~$125, Architect Professional ~$175 (Anthropic prices are aggregator-reported — see notes). Prep courses are free on Anthropic Academy, Google Cloud Skills Boost and Microsoft Learn.. Best for: Buying a recognizable, proctored, third-party-verified token in a weekend for around $100 — especially when a procurement form, partner tier or enterprise buyer needs a box ticked. Why: This is the cheapest credible floor in the category and the only one with genuine identity verification baked in: all of these are Pearson VUE proctored, ID-checked, and badge-issued through Credly. The AI-native end has real traction — Anthropic reports 36,000+ certified consultants across 1,300+ organizations as of Jul 2026 and 400,000+ people through Partner Academy training, and it added three new role-based certifications on Jul 23, 2026. Cost-to-signal is unmatched: $99–$200 and 90–120 minutes. Watch: Shallow and churn-prone. Google's Generative AI Leader explicitly targets people "with or without hands-on technical experience" — it certifies vocabulary, not capability. Microsoft retired both AI-102 and AI-900 on Jun 30, 2026, replacing them with AI-103 and AI-901, so a 2025 Azure AI badge is now a dead credential. Validity is short: NVIDIA's is 2 years, AWS and Google 3. And Anthropic's proctored exams are gated behind a Claude Partner Network company email — independent founders and consultants are locked out regardless of skill. [https://aws.amazon.com/certification/certified-ai-practitioner/](https://aws.amazon.com/certification/certified-ai-practitioner/) 4. **LinkedIn Verified Skills** (LinkedIn (Microsoft), with tool partners) — Free — but you need a paid seat on a partner tool to generate usage (Replit, Lovable, Descript, Relay.app; GitHub, Zapier and Gamma announced as coming). Best for: AI-tool operators who want proof attached to the profile employers actually look at, generated by work you were doing anyway. Why: Announced Jan 28, 2026, this is the first mainstream credential derived from telemetry rather than a test: partner tools assess "real usage patterns, product outcomes, or demonstrated proficiency within the tool," and the credential updates automatically as you improve. It sits on the distribution nobody else has — the profile that recruiters open first — and it is structurally harder to fake than an exam, because faking it requires actually using the product. LinkedIn is pitching it against the degree directly: employers "want to know what you can actually do." Watch: Almost entirely unproven. LinkedIn has published no earner counts, no employer-recognition data, and no detail on what qualifies as proficient — and it has form here: its previous multiple-choice Skill Assessments were quietly retired. The signal is also vendor-shaped, not skill-shaped ("proficient in Lovable" is a tool fact, not an engineering fact), and the partner list is a short list of AI startups whose own longevity is uncertain. Treat it as a free garnish on a portfolio, not a substitute for one. [https://www.linkedin.com](https://www.linkedin.com) 5. **Portable credential rails** (Credly by Pearson · 1EdTech Open Badges 3.0 · Accredible · skill.re / Certificate.School) — Free to the earner. Issuer side: Accredible from $45/mo (50 recipients, unlimited credentials each) · Credly enterprise-quoted · skill.re free to browse and join, 195 of its 255 courses end in a serial-numbered verifiable certificate. Best for: Making a credential you already earned machine-verifiable, portable and revocable — and, if you run a program, issuing credentials that survive contact with a skeptical employer. Why: This is the plumbing that turns a PDF into something checkable: Credly has issued 123M+ credentials from 3,700+ issuers across roughly 200,000 distinct credentials, and Open Badges 3.0 (released May 29, 2024) makes badges W3C Verifiable Credentials 2.0-compatible — cryptographically signed, wallet-portable, and carrying the assessment criteria and evidence inside the badge itself. Every vendor cert above already lands here, so you get the rails whether you chose them or not. The skill.re-class layer — cohort courses ending in a serialized, publicly verifiable certificate — is the cheap on-ramp for skills nobody sells a proctored exam for. Watch: The rails are not the signal. A verifiable badge proves the issuer issued it, not that the issuer's bar was high — and course-completion certificates from small platforms carry close to zero employer recognition. Open Badges 3.0 adoption is also slower than the spec suggests: 2.0 remains the most widely used version in 2026, so the wallet/decentralized-identity story is still mostly forward-looking. Credly publishes no public pricing. [https://info.credly.com/](https://info.credly.com/) ### How to choose - If You have shipped something with real users → Lead with the receipts and stop buying credentials. Nothing in this table outranks a live product with named customers — but make the provenance legible (commit history, public writing, changelog), because 65% of hiring managers now say AI-polished applications made verification harder. - If You have the skill but no shipped artifact yet, and you need a signal this month → One proctored vendor exam — AWS AI Practitioner ($100) or Google Generative AI Leader ($99) — is the fastest $100 of credibility available. Budget one weekend, and check the retirement calendar first: Microsoft killed AI-102 and AI-900 outright on Jun 30, 2026. - If You sell AI implementation services to enterprises → Claude Certified via the Anthropic Partner Network — 36,000+ certified across 1,300+ orgs as of Jul 2026 makes it the emerging table stakes in that buying conversation. You will need a partner-org company email; a personal address won't get you into the exam. - If You live in AI tools all day and want free, always-on proof → Turn on LinkedIn Verified Skills through Replit, Lovable, Descript or Relay.app. It costs nothing and rides the profile recruiters already open — but treat it as garnish, not as your primary evidence, until LinkedIn publishes recognition data. - If You're the one issuing credentials — a course, a bootcamp, an internal academy → Issue on Open Badges 3.0 rails (Accredible from $45/mo, or Credly if you're already in the Pearson ecosystem) so the badge is cryptographically checkable. Assume most employers still read Open Badges 2.0 in 2026 and design for that. ### The field (29 more) GitHub, Karat, Byteboard (acquired), Triplebyte (acquired), CodeSignal, HackerRank, TestGorilla, Codility, Vervoe, interviewing.io, Credly by Pearson, Accredible, Open Badges 3.0, W3C Verifiable Credentials 2.0, BCdiploma, Certificate.School, skill.re, LinkedIn Verified Skills, LinkedIn Skill Assessments (dead), AWS Certification, Google Cloud Certification, Microsoft Credentials (renamed), NVIDIA Certification, Claude Certifications, Databricks Certification, Google AI Essentials / Career Certificates, Kaggle, Polywork (dead), GetReal Security Full dossier data: https://elems.ai/e/sk.json --- Source: [elems.ai](https://elems.ai/e/sk.html) — the periodic table of the AI-led startup. Data: https://elems.ai/elements.json (CC BY 4.0, cite elems.ai).