Finding the right name was never the hard part. Turning that name into a vetted, interested, warm candidate — that’s what every hiring tool has failed to do. Until now, maybe.
Sebastian Scott was 17 when he started his first company. By his mid-twenties, he had built a tutoring platform that scaled to 15,000 users, developed AI agent systems for German manufacturers, and studied across three continents — Munich, New York, and Beijing. None of that, he’ll tell you, was the thing that clarified the problem he actually wanted to solve.
That came from watching recruiting up close and realizing that the industry wasn’t broken because of bad technology. It was broken because of bad incentives. Job boards make money on volume, not matches. Agencies charge 20 to 30 percent of a salary for the privilege of patience. And somewhere in between, the best candidates — employed, heads-down, not browsing job boards — stay completely invisible to the companies that need them most.
Clera, the company Scott co-founded and leads as CEO, now represents more than 60,000 professionals and serves over 500 startup clients. It raised a $3 million pre-seed round backed by 1984 Ventures, Deel Ventures, and angels from OpenAI and LinkedIn. It runs entirely over WhatsApp and iMessage. No app. No dashboard. No portal to log into.
In this conversation, Scott explains why AI sourcing tools solved the wrong problem, what the data has revealed about why good candidates and good companies keep missing each other, and why the direction of the ask — company reaching out to candidate, not the other way around — changes everything.
Excerpts from the interview;
Recruiting has been “broken” for decades. What made you believe AI could actually fix it this time, and not just add another layer?
Recruiting isn’t broken because nobody built enough software. It’s broken because of the incentives.
Job boards make money on volume, so they’re built to generate applications, not matches, which is why a company posts one role and drowns in 400 resumes to find the two that matter. The other option, an agency or headhunter, gives you a human who actually vets people, but it’s slow, and it costs you 20 to 30 percent of a salary. So companies have been stuck choosing between cheap noise and expensive patience.
What’s different now is that AI can finally do the part that made a great recruiter worth it, and I want to be precise about which part. There’s a wave of AI sourcing tools now, the natural-language search engines that scan hundreds of millions of profiles and hand you a ranked list with fit scores. Those are genuinely useful, and they mean finding names is basically a solved problem. But that was never the part that was broken.
The broken part is everything after the list: actually reaching those people, earning enough trust that they reply, understanding what they want, and fit-checking them with a human so the company gets a real, pre-vetted candidate instead of another pile to work through. The thing worth building is what turns a name into a vetted, interested person, at a scale and cost that lets every company get the short list instead of the pile.
Clera works over WhatsApp and iMessage, not an app or dashboard. What does that tell you about where agentic AI works versus where it gets overcomplicated?
Agentic AI works when it removes a step, not when it adds a surface.
Every app asks the person to come to you, sign up, learn the interface, and remember to check it. Most people just won’t. Texting is where they already are. A capable agent doesn’t need a dashboard to prove it’s smart; it just needs to be reachable and to actually do the thing.
These products get overcomplicated when a team builds a genuinely good agent and then wraps it in a whole UI, as if the user wants to manage the AI. A candidate doesn’t want to manage anything. They want a job. The simplest path to that wins, and it turns out the simplest path is the thread already open on their phone.
That’s also how we keep a real relationship with people who would never log into another portal, which is exactly what lets us hand companies candidates who are warm and ready, not just cold names.
Also Read: AI-Generated Code Needs a New Testing Model
You’re sitting on data about what candidates want and what startups need. What’s the most surprising thing it’s shown you?
How often both sides want the same thing and still miss each other.
Candidates count themselves out of roles they’d be great at, so they never apply. Companies write job descriptions for a person who doesn’t exist, then tell me the pipeline is empty. The gap usually isn’t ability. It’s confidence on one side, an unrealistic wish list on the other, and a good human in the middle closes it out in about a day.
The thing that surprised me most is how quickly that confidence gap closes when the direction of the ask reverses. When a candidate hears that a company has reviewed their background and specifically asked to talk to them, they show up as a completely different person than the one who’d been firing off applications into the void. Being requested to have a company recognize them first changes how people see their own worth.
Most AI hiring products are built for the employer. You flipped that and built for the candidate. Value decision, market decision, or both?
I’d tweak the premise, because the company is very much our customer. The difference is what we point the AI at.
Most hiring products help an employer get through the pile they already have: screening, ATS add-ons, and automated interviews. They’re machines for saying no faster. The problem is that the pile is the wrong place to look. The best engineers almost never apply. They’re employed, heads-down, building something, and not browsing job boards, which makes them effectively invisible to a company posting a role.
The usual fix is a headhunter, but a single recruiter can only hold a few searches in their head at once and can only work the slice of the market they personally know. That’s a tiny window onto an enormous pool, and the person you actually want is usually outside it. Reaching across the whole market used to be the hard part, but it isn’t anymore.
AI sourcing tools, Juicebox, and the rest of that category already search hundreds of millions of profiles and hand a recruiter a ranked list in seconds. A long list of the right names is now a commodity. The hard part just moved one step down: a name on a list is still a stranger who hasn’t been contacted, hasn’t agreed to anything, and might not even be looking. What we built lives in that gap. We actually reach those people, keep a real, ongoing relationship with them over text, and fit-check them with a human, so when a company needs someone, the people who were just rows in a database are already identified, vetted, interested, and warm.
We sell that to companies because that’s where the pain and the budget are. But it only works because we’re obsessive about the candidate side. And there’s a piece of this that people underrate. When a company asks to meet a candidate, that request is itself a form of recognition. Most job seekers spend months sending applications into the void, never hearing back, never knowing if anyone even read them. Flipping the direction so the company reaches out first gives people something the old process almost never did: the feeling of being wanted rather than screened. That recognition is part of the product, not a side effect. People who are genuinely good, actually open, and ready to talk are a scarce thing in this market now that the lists themselves are cheap, and you get to them by building a real relationship, a search tool never will, and by treating them like someone worth pursuing rather than a row to export. So it’s both.
Serving companies and being radically good to candidates aren’t in tension with each other. The second is how you deliver the first.
Also Read: “Disheveled,” “Not Coherent” — The Bias Is in the Notes, Not the AI
If Clera works at scale, it changes how the startup talent market operates. What does that world look like, and what’s the risk that AI makes hiring faster but not fairer?
If it works, the default flips. Companies stop posting into the void and start getting handed a small, vetted list of people who are actually a fit and actually open, and a good hire takes days instead of months. The candidate’s experience flips, too. Instead of applying and waiting, they get approached, and being requested by a company is a kind of recognition that the old process rarely gave anyone. For startups, especially, who you can hire stops depending so much on who you already know, which is the part I care about most. AI search has already made finding people cheap, so the thing that stays scarce, and the thing worth building around, is trust and genuine interest rather than another list.
But the risk is real, and I won’t wave it away. AI can absolutely make hiring faster and not fairer, because if you train it on who got hired before, you just rebuild the old gatekeeping behind a cleaner interface. Faster is the easy win. Fairer only happens if you design for it on purpose: judging people on what they can actually do, putting overlooked candidates in front of companies that would never have found them, and being honest about what the system optimizes for.
Speed is a side effect. Fairness is a deliberate choice, and it’s the only reason this is worth building.
