Coinbase Named the "One-Person Team." LinkedIn Title Search Returns Zero.
Coinbase's May 2026 AI-native restructuring named a profile that doesn't exist in any ATS. Here's how to source it by skill and evidence, not title.
Brian Armstrong's May 2026 email to Coinbase gave recruiters a new archetype and no way to search for it. He named "one person teams" with engineers, designers, and product managers in a single role, supported by fleets of AI agents. If you typed that string into LinkedIn Recruiter this morning, you got nothing useful back. The profile is real. The title is not.
What Armstrong actually said, and why it matters for sourcing
The Coinbase memo wasn't a generic AI pivot. Armstrong announced a roughly 14% cut framed as a deliberate AI-native restructuring, flattened the org to five layers below CEO and COO, deleted pure manager roles, and pushed leaders into player-coach seats with up to 15 reports. Then the line that broke recruiter brains:
"We'll also be experimenting with reduced pod sizes, including 'one person teams' with engineers, designers, and product managers all in one role."
The justification was a productivity claim: "I've watched engineers use AI to ship in days what used to take a team weeks. Non-technical teams are now shipping production code." That is not a description of someone who is good at Cursor autocomplete. It is a description of someone who orchestrates agent fleets, designs the thing, scopes the thing, ships the thing, and debugs the thing in production. Coinbase wants AI-native pods. The market calls these people vibe coders, or doesn't call them anything at all.
Either way, your ATS doesn't have a checkbox.
The title search is dead. Here's the proof.
Run the obvious queries on LinkedIn. "Vibe coder" returns hobbyists and a few job postings. "One-person team" returns founders venting on Sundays. "AI-native engineer" returns people who added the phrase last week. The people who actually fit Armstrong's spec are titled "Founder," "Co-Founder," "Software Engineer," or nothing at all. Their evidence lives in shipped artifacts, not headlines.
Meanwhile the demand side is screaming. Vibehackers.io has analyzed 1,304 Upwork postings tagged to this stack in 2026 alone, and ZipRecruiter listings now spell the requirement out verbatim: "AI coding tools (Claude Code, Cursor, Lovable, Bolt) to support development tasks." The salary range tells you who's hiring: $90K to $130K for mid-level, $150K to $400K+ at senior, with internship hourly rates of $40 to $50 from companies that don't usually pay interns that.
The job is real. The taxonomy hasn't caught up. If you keep sourcing by title, you will lose every search to a recruiter who learned to source by evidence.
What the profile actually looks like
Andrej Karpathy coined "vibe coding" in February 2025 as a half-joke about letting the model drive. By May 2026 Coinbase had industrialized the joke into an org chart. In between, an archetype hardened into something specific enough to source against.
The toolbelt is the signal, not any single tool
Anyone can list Cursor. The real archetype lists two or three tools in deliberate combination, because that's how senior practitioners work. Till Freitag's 2026 landscape review counted 138+ tools across seven crystallized categories and concluded that the best teams use two or three in combination. In practice that looks like:
- Lovable or Bolt.new for the prototype and the marketing site.
- Cursor or Claude Code for the production codebase, with subagents and MCP servers.
- One harness or framework for orchestrating agent fleets (think Andrew Wilkinson's Harbor on GitHub).
Lovable, by the way, is not a toy in this stack. It holds the record for fastest SaaS to $200M ARR (12 months), doubled to $400M in four more, and sees over 200,000 new projects created daily. Cursor sits at $2B in annualized revenue with 360,000 paying users. These are infrastructure-tier tools now, and fluency across them is the actual sourcing keyword set when title search fails.
Agent orchestration is the differentiator
"AI-native talent who manages fleets of agents" is the operative phrase in the Coinbase memo. Not "uses Cursor." Not "ships fast." Manages fleets. The differentiating skill is orchestration: Claude Code subagents, MCP servers, custom harnesses, evals. Andrew Wilkinson's setup at Tiny is the canonical public example. His CFO, with zero coding background, vibe-coded a replacement for Adapar (priced at $50K to $100K per year) in about two weeks. Wilkinson swapped headcount for a $40K-a-month Claude bill and called it cheaper.
That story is a sourcing template. Ask candidates what they've shipped solo, end-to-end, in under two weeks. Ask what the agent layer looked like. Ask what broke in week three.
The right sourcing question is not "are you full-stack?" It is "what have you shipped solo end-to-end in under two weeks?"
The archetype is unbundling, not consolidating
The "one-person team" sounds like a generalist. It isn't. In practice these people are hyper-specialists in velocity who outsource design, PM, and QA to agents. Lazar Jovanovic, a full-time professional vibe coder at Lovable, is the role made literal: his job is to build internal tools and customer-facing products purely using AI, and he doesn't have a traditional coding background. He's not a generalist. He's a specialist in a thing that didn't have a name 18 months ago.
How to source it when title search returns zero
You need three signals stacked: tool fluency, shipped artifacts, and post-ship iteration. Each one filters out a category of pretender.
1. Tool fluency, in combinations
Search for profiles, repos, and bios that name two or more tools from the canonical stack: Cursor, Claude Code, Lovable, Bolt. Solo Cursor mentions are noise. Cursor plus Claude Code plus a deployment trail is signal. This is exactly the kind of plain-English query that titles can't express and Boolean barely can, which is why we built Refolk: you describe the person ("engineers who ship solo SaaS with Lovable plus Claude Code and have a live production URL"), and you get a ranked shortlist across GitHub, LinkedIn, and the open web.
2. Shipped artifacts with live URLs
GitHub stars are a lagging indicator for this archetype. Live URLs are the leading one. Look for personal landing pages that list three to ten shipped products, each with a working link, ideally with revenue or user counts attached. The vibe-coder community has converged on this format because the tools make it cheap. If a candidate's portfolio is a resume PDF, they're not the profile.
3. Post-ship iteration, not demo videos
This is the recruiter's hidden filter, and it's the one that matters most for an enterprise like Coinbase. Security researchers found that 170 out of 1,645 Lovable-created web applications shipped with vulnerabilities exposing personal information to anyone. That's more than 10% of apps leaking user data on launch.
The hireable candidates are the ones who can show what broke after launch and what they did about it. Bug postmortems on X. Changelog entries with dates. Refactors. Agent eval suites. If everything in their portfolio looks like a launch demo, the cost of hiring them will land in your incident channel.
Where the pool actually concentrates
Refolk's index shows the obvious heavyweights (SF Bay Area, Austin) plus a notable long tail in Istanbul, Vancouver, Paris, Lithuania, and the UAE. The communities worth mining directly:
- Vibehackers.io job board and goodvibecode.com, which claims over 10K vibe coders.
- Vibecode Network and Lenny's Newsletter's vibe-coding interview series.
- Startup Ideas Podcast and the "Where It Happens" audience, where Wilkinson and others have walked through their stacks publicly.
- Lovable's own community, where Jovanovic and peers post builds.
Most of these people are not on LinkedIn looking. They're shipping. The sourcing motion is closer to scouting than recruiting, which is another reason a plain-English query layer matters more than another Boolean string. Refolk pulls evidence from GitHub commits, personal sites, and open-web posts in the same shortlist as LinkedIn profiles, which is what you actually need when 92% of US developers now use AI coding tools daily and 41% of all code written globally is AI-generated. The signal isn't in one platform anymore.
What to put in the requisition
Stop writing "5+ years of full-stack experience." It selects against the archetype. Try this instead:
- Required evidence: three or more live URLs the candidate built solo in the last 12 months, with at least one showing real users or revenue.
- Required tools (any two): Cursor, Claude Code, Lovable, Bolt.new.
- Required behavior: public post-ship iteration. Changelogs, bug fixes, eval suites, or postmortems.
- Bonus: agent orchestration. Subagents, MCP servers, custom harnesses, anything that proves they manage fleets and don't just prompt.
- Title field: leave blank, or write "Engineer / Designer / PM."
If you write the req this way and source against it with evidence-first queries (Refolk handles this natively, but the principle works anywhere you can search outside of LinkedIn titles), you will find people. If you write it the old way, you will spend Q3 explaining to your CEO why the one-person team is still a five-person team.
Coinbase named the role. The pool exists. The titles do not. Source accordingly.
FAQ
Is "vibe coder" really a job title companies are hiring for?
Yes, and increasingly as a serious one. Vibehackers.io has analyzed 1,304 Upwork postings citing the vibe-coding stack in 2026, and there are 223+ full-time roles citing Cursor, Lovable, Claude Code, or Bolt as core requirements. Salaries top out at $400K+ for senior positions at major tech companies. Lovable itself employs a full-time professional vibe coder. The label still sounds informal, but the spend is not.
Why doesn't LinkedIn Recruiter title search work for this profile?
Because the people who fit are titled "Founder," "Software Engineer," or nothing, not "vibe coder" or "one-person team." Refolk's index shows roughly 12,900 profiles citing the canonical tool combination globally, but almost none use the new labels in their headlines. The signal lives in shipped artifacts, tool combinations, and post-ship iteration, all of which sit outside LinkedIn's title field.
What's the single best filter to separate real hires from demo-video pretenders?
Post-ship iteration. Given that more than 10% of Lovable-built apps audited in 2026 shipped with public data exposure vulnerabilities, the candidates worth hiring are the ones who can show what broke after launch and what they did about it. Look for changelogs with dates, public postmortems, and agent eval suites. A portfolio of pristine launch demos with no follow-up is a red flag, not a green one.
Does Coinbase's "one-person team" framing apply outside crypto?
Yes. Armstrong industrialized something Karpathy named in February 2025 and that Wilkinson, Lovable, and dozens of small studios have been running for over a year. Any company experimenting with AI-native pods, agent-fleet workflows, or radically smaller teams is hiring against the same archetype. The sourcing problem (no title, evidence-only) is identical whether the logo is Coinbase or a 12-person seed-stage SaaS.