Forty-five minutes. Three tabs of X-Ray syntax open. A notepad full of Boolean variations. Zero candidates reviewed.
I've had that exact feeling roughly ten thousand times in fourteen years of recruiting. There has to be a better way to do this.
Of course, I never did anything about it. I'd been carrying this idea for a while — paste a job description, get a ranked list of real people, skip the part where I become a human search engine — but I assumed I'd need to hire a developer, or learn Python, or somehow become a different person. So I filed it under "someday" and went back to the Boolean strings.
Then I started using Claude Code.
WHAT SOURCERY ACTUALLY DOES
Sourcery takes a job description, extracts the key criteria using AI, generates X-Ray search queries, fires them across multiple platforms at once, and returns a ranked list of candidate profiles — each scored 1–10 with a one-liner explaining why it matched.
It runs in about fifteen seconds.
That's the whole product. Left side: paste a JD, pick your sources, hit run. Right side: profiles ranked by relevance, each with a score and a one-liner from Claude on why it matched. Click through to the actual profile. Export to CSV when you're done.
IT'S NOT JUST LINKEDIN
Most sourcing tools are LinkedIn wrappers with better branding. Sourcery searches across nine platforms and — this is the part I'm most proud of — recommends which ones based on the role.
Paste a software engineering JD, it points you toward LinkedIn, GitHub, and Stack Overflow. Design role? Dribbble and Behance show up. Marketing or content? Substack and Medium get pulled in.
A GitHub contribution history tells you something a LinkedIn profile flat-out can't. What someone actually builds. How they write code. Whether they've shipped anything. That's real signal, and most tools ignore it completely.
HOW IT GOT BUILT — HONESTLY
I wrote almost none of the code.
I described what I wanted in plain English, gave Claude Code a detailed spec, and it built a working React + Node.js app from scratch. Not a prototype. A real deployed web app — backend API, live search execution, AI scoring, server-sent event streaming, CSV export button. The whole thing.
The flow under the hood:
My job was describing the problem clearly and giving feedback. Claude Code installed packages, wrote tests, fixed its own errors.
The whole build — v1 and v2 — took one evening and the better part of the next morning.
THE OBVIOUS QUESTION
Fourteen years in, people want to know if I'm automating myself out of a job.
No.
Sourcing was never the valuable part. It was the tax you paid to get to the valuable part — the conversations, the reads on candidates, the close. The hours building Boolean strings were overhead. Pure overhead. Sourcery compresses that to near-zero, which means the time goes somewhere it actually matters.
The recruiter who figures out how to use these tools is going to outperform the one who doesn't. Not because AI does their job — because AI handles the parts that shouldn't have been their job in the first place.
WHAT'S NEXT
Sourcery's a personal tool right now. A few dollars a month in API costs, deployed on Railway, accessible from any browser. I use it on real searches at Hard Rock Digital and it's already changed how I work.
V3 is probably saved searches, maybe Proxycurl for profile enrichment, maybe an outreach sequencer so the whole top-of-funnel lives in one place.
But honestly, I keep thinking about the Tuesday afternoon. Three tabs of X-Ray syntax open. Forty-five minutes of search strings. Zero candidates reviewed.
Took me fourteen years to fix it.
Took about twelve hours to build the fix.