The firm

The Zelcast thesis

Why we only work with the top 1% of engineers and startups under 100 people, and what operating taught us about how the best teams actually get built.

Zelcast  ·  July 27, 2026

Zelcast exists because the market for engineering talent fails exactly where (and when) it matters most.
The best engineers do not apply for jobs.
The best early-stage startups cannot afford a single wrong hire.
The recruiting industry, which was built on volume and big-company signals, cannot tell the difference between someone who screens well and someone who can carry a company. We can, because we have operated. We keep the aperture narrow on purpose: the top 1% of engineers, startups under 100 people. Nothing else.

For years, founders we knew would ask the same question: who should I hire and how do I actually get them? Strong engineers we knew would ask the mirror of it: which of these companies is worth a bet?

The people behind Zelcast have founded companies, hired hundreds of engineers, and backed a few hundred startups, and after enough years of that you simply know who is good on both sides. The introductions turn into hires and hires turn into the best people at those companies. The reason is not the size of anyone’s rolodex. It is that operating teaches you what to look for.

Everything we believe about hiring comes from doing it badly, and then less badly, at our own companies. A few things hold up over and over.

Engineering output follows a power law. The gap between a good engineer and a great one is not twenty percent. On the problems that decide whether a startup lives, it is closer to ten times, because the great ones choose the right problem before they solve it.

The first ten engineers set a ceiling. A-players recruit A-players. One mediocre early hire will lower the bar for every hire after and no one will notice until it is expensive to fix. Talent density is the only startup advantage that compounds completely on its own.

Interviews are a mediocre signal at best. The strongest signal in hiring is a person who worked beside someone for two years and will stake their name on them. Networks carry information that resumes cannot.

The cost of a wrong hire is not symmetric. Google can absorb a thousand hiring mistakes a year without noticing. A twelve-person company that hires the wrong senior engineer loses a year. Most seed-stage companies do not have a spare year.

Why the open market fails at the top

Job boards and application pipelines are a market for people who are looking. The best engineers are mostly not looking. They are employed, paid well, and buried under recruiter spam they do not read. The open market will systematically show companies everyone except the people they actually want. We wrote about this selection effect on its own, but the short version is that a great applicant pool and the great engineers are two different populations.

Recruiters will screen for the wrong game

The deeper problem is that the recruiting industry learned its craft inside big companies, so it screens on big-company signals: the brand on the last job, the years in the title, keywords in the stack. Those work when the role is a well-specified lane inside a system that catches mistakes. At an early-stage company they barely predict anything, because the job is very different in kind.

Half the work is deciding what the work is.

The things that decide the outcome, what someone builds unprompted, how fast they move through ambiguity, whether the people around them get better, never appear on a resume, and a keyword screen will not see them. A recruiter can match a profile to a req. Very few will tell you whether a person can carry a company, because most have never been in one small enough to know.

AI breaks the volume model

The AI era makes this much, much worse for the old model. Not better. Outreach is now free. Any recruiter can send ten thousand personalized messages before lunch, which is exactly why none of them get read. Sourcing tools all pull from the same indexes, so everyone is looking at the same lists of the same people. When volume costs nothing then volume is worth nothing. The only assets left are the ones that have always been scarce. Knowing who is actually great and being someone both sides trust.

At the same time, AI has raised the stakes of every hire. A great engineer with today’s tools does work that took a small team five years ago, so the best companies are staying smaller, longer. Fewer seats, more leverage per seat, and each hire carrying a massive share of the outcome. AI’s arrival has concentrated the war. The fight is no longer over headcount but over the small number of people who can actually wield the new leverage.

Quality is the entire game

Recruiting is also built as a volume business. Submittals per week, seats per quarter, a fee on fill. The math works because at scale hiring is statistical and a big company can afford to be a little wrong a thousand times. An early-stage company gets maybe ten swings, and every one of them is ten percent of the company. Fifty resumes is less progress at that stage and more fifty distractions from the one person who matters.

We run the opposite model. Fewer introductions with high conviction. We live with both sides of every one we make, which keeps us honest in a way no process can. We would rather show a founder one right person than twenty possible ones.

Relationships are the product

We know our people. We get to know engineers before the right role exists: what they want to build, what they care about, what kind of founder they would follow, what would actually make them leave a good job. Most of the engineers we work with are not looking and we are not trying to place them. We are keeping up a relationship, so that when the right thing appears, we both recognize it in the first conversation.

We do the same on the other side. Before we put anyone in front of a company, we learn how the company actually works. What do the founders believe, how do decisions get made and what does the culture reward. At an early-stage company that context is 99.78% of the match. The same brilliant engineer is a great hire at one startup and a disaster at the next. The difference is never in a JD.

Why only startups under 100 people

We do not work with Microsoft, Salesforce, Oracle, Google, or OpenAI. This is not because they are bad companies, but because they do not need us. Past a certain size, hiring is a process problem. It is much more about pipelines, panels, committees, calibration. Big companies are good at process and the market is full of firms that serve them.

For startups with under a hundred people, hiring is much more a judgment problem. There is not a committee to average out a mistake. Every hire will visibly move the company’s ceiling, and the person has to be right on dimensions a pipeline will not measure: how do they handle ambiguity, can they raise the bar around them, as a founder can I build on them. Judgment does not (and will never) scale through job boards. It will come from having sat in the seat and paid for the mistakes.

Why only the top 1%

Matching is only worth anything when both sides are scarce. Most engineers can find a fine job without us and most roles can be filled without us. That part of the market works well enough. The narrow slice where it truly fails is at the top: the small number of engineers who change a company’s trajectory, and the small number of companies intense enough to deserve them. That slice is the entire business.

Where this goes

Our ambition is specific: we want the first ten engineers at the most consequential startups of the next decade to have been chosen for the things that actually matter, by people who knew how to see them. Software widens how much of the market we can cover and AI has made that coverage wider still. The judgment at the end is still human and that part does not automate.