Vibe Coding Is in Adobe and JPMorgan JDs. Boolean Returns 7 People.

Adobe, JPMorgan Chase, TikTok, Salesforce, Lam Research, KLA, Guidehouse, and Palo Alto Networks now name "vibe coding" (or Cursor, Claude Code, Lovable, Replit Agent, v0) as an explicit requirement in active 2026 JDs. Indeed is tracking more than a thousand of them. And when you paste that same vocabulary into a LinkedIn Boolean, you get almost nothing back, because the words live on the employer side of the market, not the candidate side.

That gap is the entire story. The hiring market has a new skill, the candidates who have it leave a different kind of trail, and the sourcing playbook most teams are running was built for a world where people put their tools on their resume. They don't anymore.

The keyword is on the wrong side of the market

Andrej Karpathy coined "vibe coding" in early 2025 to describe how people were actually shipping with AI assistants. Eighteen months later it's a line item in enterprise JDs. TikTok San Jose is hiring a Product Designer (Design+Code/Vibe Coding) in a $108,000 to $220,400 band, asking for candidates who can "use coding and AI tools (e.g., Vibe Coding, AI coding assistants) to rapidly turn product ideas into interactive prototypes." Lam Research wants "experience with GenAI coding tools, vibe coding, or vibe engineering" on a Fremont Software Engineer Apps 5 req. JPMorgan Chase wants "vibe coding experience and conceptual knowledge while leveraging developer assistant tools like GitHub Copilot." ServiceNow's Moveworks federal team wants someone who can vibe-code with Claude Code to build FedRAMP Continuous Monitoring automation.

Optum, Okta, NICE, Illumina, Zeta Global, EY, Pinterest, Deel, HqO, and the Brooklyn Nets / Barclays Center are all running variants. Adobe Summit 2026 in Las Vegas literally ran an official session, L613, titled "Vibe Code an Author Experience for Edge Delivery Services." Adobe isn't just hiring for it. Adobe is teaching its enterprise customers how to do it.

Then you go look at candidate profiles.

7

Seven. The softer query "AI-assisted development Cursor" returns single digits. This isn't a market with no candidates. This is a market where the candidates exist and the vocabulary doesn't transfer. People who ship with Cursor and Claude Code every day don't put "Cursor" in their headline any more than a backend engineer in 2015 put "vim" in theirs. It's a tool, not an identity, and resumes still describe identities.

Where the actual signal lives

If the keyword isn't on the profile, you need a different surface. Three of them, mostly.

1. GitHub artifacts, not GitHub topics

The diagnostic files for serious vibe coders are .cursorrules, CLAUDE.md, and mcp.json. Each project a power user runs gets its own .cursorrules and CLAUDE.md that add project-specific context to the agent. These files sit in public repos. They're searchable. A candidate with six repos that each have a thoughtful CLAUDE.md is telling you something a LinkedIn skill tag never will, which is that they've internalized the workflow enough to operationalize it across projects.

Stars and followers don't matter here. Artifact density does. The Cursor power user looks like someone with a handful of small repos, all updated in the last 90 days, each carrying agent config files, often paired with a Vercel or Lovable preview link in the README.

2. Subdomain trails

Lovable publishes projects to a default xxx.lovable.app URL. v0 ships to *.vercel.app. Replit Agent projects sit on *.replit.app. These are googleable, indexable, and crucially, attributable. A candidate's "build in public" Twitter thread will usually link to two or three of them. Their personal site often lists them as case studies.

This is the part of sourcing AI engineers that breaks traditional Boolean. You're no longer searching for a job title or a skill keyword. You're searching for evidence of shipping. Three Lovable subdomains and a v0 fork tied to one email beats any self-reported "AI-assisted development" tag on a profile. This is exactly the kind of cross-surface query Refolk was built for: you describe the person ("solo builder, three or more shipped Lovable or v0 projects in the last 90 days, based in the US, design background") in plain English and get a ranked shortlist that pulls from GitHub, LinkedIn, and the open web at once instead of asking you to write three separate searches.

3. The community surface

The Cursor Community Discord has roughly 100,000 members per public reporting. The v0.dev community, Replit Bounties, Indie Hackers, ProductHunt launch pages, and the goodvibecode and vibehackers job boards are where this cohort actually congregates. Vibehackers analyzed 372+ vibe coding job listings in 2026 with named comp data from Google, Amazon, and Airtable, which tells you both that the market exists and that the candidates reading those boards are the ones with the skill.

The Cursor revenue tell

Here is the part most sourcers get wrong when they try to fix this with a broader Boolean. They expand "Cursor" to "AI coding tools" or "Copilot" and pull thousands of profiles. That's worse, not better.

GitHub Copilot has over 20 million users. Cursor has around one million. Cursor makes roughly four times the revenue. The JDs at Adobe and JPMorgan are not asking for "anyone who has ever touched Copilot." They are asking for the power-user tier, the people who pay for Cursor out of pocket, who have switched their default editor, who run multi-agent orchestration across projects. That's a much smaller pool than aggregate "AI tool user" queries suggest, and it's the pool the JDs are actually written against.

Customer evidence backs this up. Over 85% of developers at Box now use Cursor daily, driving a 30 to 50% increase in product roadmap throughput. Stripe's customer page says "more than 70% of our engineers now use Cursor." By February 2025, every Coinbase engineer had utilized Cursor. Shopify, Vercel, and Linear all switched per Orbilon reporting. When a JPMorgan recruiter writes "vibe coding experience," they mean someone whose day looks like a Box or Stripe engineer's day, not someone who tried Copilot once in 2023.

4x

"Vibe coding" is not an engineering-only skill

The other reason most Boolean searches miss is that recruiters narrow to "software engineer" and lose the whole rest of the surface. The 2026 JD set spans design (TikTok UX, Adobe Firefly), product (HqO), security and GRC (ServiceNow, Palo Alto Networks), audit (Guidehouse), sales engineering (NICE), BI analysts, and even motion design (Uniqode).

The TikTok req is a Product Designer. The Guidehouse req is in audit. Vibe.co in NYC writes the skill as "a force multiplier, not a shortcut, a craft." If you're sourcing only against the SWE title, you've cut yourself off from the majority of the openings and most of the qualified candidates. PMs who can ship a working Lovable prototype in an afternoon are in scope. Designers who can hand engineering a working v0 demo are in scope. GRC analysts who can vibe-code a FedRAMP automation in Claude Code are very much in scope.

The JDs are hiring for taste, not for tool installation. Source for what people ship, not what software they own. </p>

Source for discernment, not adoption

There's a 2025 randomized study, cited by skeptical engineering leaders, that found experienced open-source developers using AI tools took 19% longer than without. AI made them slower. This is the study every senior engineer who hates the hype cycle is quoting at you, and it's the reason the better JDs are screening for judgment.

Look at the language: "know when to vibe code vs architect manually" (multiple JDs), "force multiplier, not a shortcut" (Vibe.co), "discernment" (several Salesforce and Adobe variants). They are screening out the uncritical AI user. They want someone who can ship a Lovable prototype on Monday, throw it away on Tuesday because it's the wrong abstraction, and rewrite it in Cursor on Wednesday with proper types and tests.

That's why artifact velocity is a better signal than tool name. The diagnostic question for Cursor Claude Code recruiting is not "do you use Cursor?" It's "show me three things you shipped in the last 90 days and tell me which parts you wrote yourself versus reviewed from the agent and why." The first question has a yes/no answer that any candidate can game. The second has a portfolio attached.

This is also where a plain-English sourcing layer earns its keep. "Find designers with a shipped Lovable project in the last 60 days who previously worked at a Series B or later consumer company" is the kind of query Refolk handles in one shot, where the same intent would otherwise require a GitHub search, a Lovable subdomain crawl, three LinkedIn filter passes, and manual cross-referencing.

The 60-second sourcing checklist

When you next get a Lovable Replit developer hiring brief, or any vibe coding jobs req from an Adobe-tier client, run this instead of your usual Boolean:

1. Pull GitHub users with public repos containing .cursorrules, CLAUDE.md, or mcp.json files updated in the last 90 days.
2. Cross-reference against *.lovable.app, *.vercel.app, and *.replit.app subdomains that appear on personal sites or in tweet history.
3. Add the Cursor Community Discord, v0.dev showcase, Replit Bounties, and vibehackers / goodvibecode job board active applicants.
4. Filter by artifact velocity: three or more shipped projects in 90 days, not stars or followers.
5. Only then map to LinkedIn for company history, location, and outreach contact.

The order matters. Start at the artifact, end at the profile. The profile is where you close. It is not where you discover.

FAQ

Why don't candidates list Cursor or Claude Code on their LinkedIn?

Because power users treat these as editors and assistants, not as resume skills, the same way a 2015 backend engineer didn't list "vim" or "tmux." The tool changes every six months and the underlying skill (shipping working software fast with AI as a collaborator) is what they actually want credit for. Until LinkedIn adds a "shipping cadence" field, the surface is going to lag the skill by 18 to 24 months.

Is "vibe coding" only for early-stage startups?

No. The 2026 JD set is dominated by enterprises: Adobe, JPMorgan Chase, TikTok, Salesforce, Lam Research, KLA, Palo Alto Networks, Guidehouse, ServiceNow, Optum, EY, Illumina. Adobe Summit 2026 ran an official session teaching customers how to vibe-code their Edge Delivery Services. The skill has moved through the early-adopter cycle and into the Fortune 500 procurement-and-compliance phase.

How do I screen for "discernment" versus uncritical AI use?

Ask candidates to walk through three things they shipped in the last 90 days and identify, per project, which code they wrote, which code came from the agent, and which agent output they rejected and why. The 2025 RCT showing experienced devs got 19% slower with AI is the screening lens here. You want people who know when to use the tool and when to set it aside, not people whose entire workflow is prompt-and-pray.

What does Refolk do differently for this kind of search?

Most sourcing tools index LinkedIn-shaped data. Refolk indexes GitHub, LinkedIn, and the open web together and takes plain-English queries, so a brief like "find US designers with a shipped Lovable demo in the last 60 days, previously at a consumer Series B+" returns a ranked shortlist in one query instead of three tools and a spreadsheet. For an AI-assisted development skill that lives in artifacts more than in profile text, that's the difference between finding seven people and finding seven hundred.