Why Vibe Coding Makes Behavioral Sourcing Finally Work

Finding Signals With machines

For a long time, our sourcing practices revolved around the same gravitational center: LinkedIn, Boolean strings, Google X-ray … overall, keyword precision. It works. It scales. It reassures. But it also reduces people to titles, buzzwords, and static CVs, which is not ideal.

Think behavioural vs Think ROI

Imagine you are seeking for a particular type of Ml Engineer, working under Search, Information Retrieval cathegory. Some of those guys write solid public repositories. They publish models on Hugging Face. They comment, debate, and share ideas across platforms, networks, and dedicated Slack channels. They contribute meaningfully, inside the company and far beyond it.

But only a few of those guys have taken the time to carefully fill their LinkedIn profile. Their title are perfectly up to date, nice listed projects, cool buzzwords. Their summary reads like your hiring manager’s wishlist.

So what happens next?

You’re a sourcer. You open LinkedIn. You plug in your Boolean string. And let’s not be naïve here. LinkedIn’s AI copilot, Juicebox, and all those shiny new sourcing tools are still built on the same foundation: LinkedIn data. They may look smarter, faster, more “AI-powered,” but they ultimately retrieve the same kind of signal. If it’s not visible on LinkedIn, it simply doesn’t exist. So… you plug your cool key-words, and, you, along with every other recruiter working on similar roles at competing companies, fall straight into that trap. You retrieve that few guys that filled properly their LinkedIn. And you all reach out to those same person. You all call the same candidates. You all feel like your job is done.

And that’s the flaw.

Not in the tool, but in the way we’ve been trained to think about sourcing. We don’t miss talents because it isn’t there. We miss them because it doesn’t speak Boolean.

And a big part of us already knows this in theory. But in practice, it still feels like looking for a needle in a haystack. Let’s not romanticize it.

If we’re thinking in terms of ROI and measurable impact , if we’re trying to be ready for the inevitable “So what have you actually done all week, Diane?” question, then spending hours manually digging through unconventional platforms for potential candidates probably isn’t the smartest path. 100% agreed.

And that’s the ugly truth: my sourcing pace would be painfully slow, and LinkedIn would always win the quantity battle (even if the quality-to-conversion ratio isn’t always there). And when the pressure is to convert more, we start emailing the wrong people, people with fewer relevant keywords in their profiles, because we’re trying to hit volume and we’re basically guessing they might fit. And little by little, we lose our credibility and reputation as a technical recruiter.

But that was before vibe coding entered the picture.

What LLM can do for you

At the time, the “behavioral sourcing” page in my playbook was only focused on some specific manual slack channel extraction for Information Retrieval and some signals on reddit for Lispers. Since then, I’ve been quietly filling it, experiment by experiment, signal by signal.

Here are some cool tools and reflections to widen our scope.

Apify: A Marketplace to Track Social Media Posts & Comments at Scale

Apify has a marketplace called Apify Store, where you can pick from hundreds of prebuilt automations called Actors. Actors = ready-to-run scrapers / bots

• Each Actor is like a mini-product
• Built for a specific platform or use case
• Has settings (keywords, URLs, filters, limits)
• Can be scheduled or triggered automatically

Why is it interesting for behavioural sourcing? Because Apify is built to track exactly that: actions, activity, signals, patterns.

I experimented last month with Apify to track LinkedIn comment activity and combined it with looping enrichment on company and location. It surfaced conversations that never appear in profiles, and people who would never show up in a Boolean search. Think about it, if you are tracking Sales or product people, (because there are better platforms for engineers) who actively talk about a specific product helps you spot real expertise and credibility that never shows up on a LinkedIn profile. And it also gives you immediate context for outreach.

Note that I’ll write about that Apify work separately in an other article, because it deserves its own space.

Screenshot of how the Apify connection is orchestrated with Zapier:

Now speaking of engineers: Reddit threads, medium, X posts, ResearchGate, meetups, discords… Apify Store has a product for each of those needs. So, ideas & trials can be endless. All of this can be automatically retrieve with Apify (and I’m not earning anything by saying this, but you get $5/month of free credits).

These implementations can be easily guided by an LLM. In my case, especially if, like me, you´re looking for enrichements.

Limitations: Only LinkedIn can loop with a profile enricher, which means you can retrieve posts, but not always connect them back to full profile data. So good luck figuring out how to reach the person, where they work, and where they’re based.

Codes run to source

GitHub repositories where proof-of-work is unavoidable. I have explained how it s achievable in a previous article. And most recently, Hugging Face models and datasets are reachable (Finaly managed to make it work this week!).

None of this was meant to signal that we had suddenly become exceptionally advanced. Quite the opposite. The goal was to open the door and say: “these are ideas”. Many of them are achievable by non-technical recruiters. But if we want them to last, to scale, to be safe, we needed engineers in the room. And now, with vibe coding, there’s a new twist: You can actually implement it, you just need a clear vision of what you want to achieve.

In a few prompts, you can spin up a script that retrieves a candidate’s public GitHub repositories, maps their contribution patterns, or pulls their publicly shared Hugging Face models and datasets. That alone is proof that these signals can be interesting: they’re measurable, repeatable, and increasingly easy to operationalize.

Some Takeaways

The takeaway is that we now have a huge rabbit hole open, ready for you, non-tech, to jump. The barrier to experimentation collapsed last year.

What used to require an engineer and a sprint can now be prototyped in an afternoon, what used to be manually done is now achievable at large scale. Meaning recruiting teams can bring engineering in where it matters most: turning prototypes into systems that are reliable, fair, and secure.

Signal-driven sourcing is becoming real.