Why I launched Work at AI Lab

I had a daily cron job watching AI-lab jobs for me, and it still wasted my mornings.

A friend and I had independently built the same workaround: a scheduled Claude Code monitor that checked AI-lab boards every day against our preferences — role, location, remote or hybrid, salary. Every morning it handed us the new matches. It felt like being automated. It wasn't.

I still spent 10–20 minutes a day reading specs, checking suitability, researching labs I had never heard of, and trying to remember whether I had already looked at a posting — or applied to it, or been rejected from it. The monitor kept resurfacing jobs I had already investigated. Claude would not remember. So neither did the system.

That is why I launched Work at AI Lab: every AI lab, every open role, in one search — with links that take you straight to the lab's own application page. No middlemen.

The problem is memory and discovery, not listings

The listings exist. They live scattered across dozens of applicant-tracking boards, one per lab. The part that breaks is everything around them.

First, memory. A job hunt is a weeks-long triage loop, and every tool I tried treated each visit as the first visit. Save, dismiss, apply — those decisions evaporated between sessions. I re-read the same postings. I re-researched the same labs. The most expensive part of my search wasn't finding jobs; it was re-judging them.

Second, discovery. Before automating anything, I had researched roughly 200 labs. That list is what made the monitor valuable — and it is also what made it painful. Most of the interesting roles were at labs I knew nothing about. Each one meant a new round of questions: what do they actually build, who is behind them, is my experience relevant, is this even worth an application? A bare title on an aggregator answers none of that.

And title matching alone would have filtered me out of AI entirely. I came from video technology. Nothing about my job titles said "AI lab candidate," but the skills transferred — and I ended up applying to, and interviewing with, labs I had never heard of before I started looking. My friend's search was the mirror image: he wanted a shortlist of prestigious, values-aligned labs watched closely. Mine needed the opposite: unfamiliar labs made legible. The product has to serve both.

What was already out there wasn't it

I tore down the existing boards before building. One covered only three labs with no search at all. Another covered eleven companies but carried zero compensation data, zero indexable pages, and data five weeks stale. The biggest one — ~19,000 jobs across nearly 700 companies — was fresh and well monetized, but deliberately broad: hardware, defense, legal tech, consumer startups. Everything AI-adjacent, which means the frontier-lab signal drowns.

The gap I kept hitting: nobody tracks the labs themselves as the unit of coverage — many of them, checked daily, with honest freshness and sourced context per lab. That is the product: lab coverage as the growth surface. Each onboarded lab unlocks its jobs. The direction is hundreds; what the site shows today is the actually monitored count, not the aspirational backlog.

Freshness has to be honest or it is worse than useless. A failed fetch must never masquerade as a wave of closures, and "first seen by this service" is not the employer's posting date. So the site shows its sources, its last successful check, and its unknowns explicitly. Three lines of footnote buy a lot of credibility — I learned that studying the one competitor whose pay-stats footnotes I actually believed.

What workatailab.com is today

Today it is a public discovery product, free, no account: search jobs and browse labs, open a job or a lab page, and follow the link to the original application. Remote and country browsing included. The landing page leads with search, then the latest jobs, then exploration by lab — because the point is the unfamiliar lab you find on the way to the familiar one.

What it is not yet matters too, so I will say it plainly: there is no personal inbox, no matching, no digest email, no account that remembers your decisions across devices. That is the roadmap, not the launch. I would rather ship an honest directory with fresh data than advertise memory I haven't built.

The goal: earn candidates first, charge employers later

The goal is sequenced on purpose: demonstrate a growing, returning audience of job seekers first — weekly active seekers, repeat visits, outbound apply traffic per lab — and only then test employer-paid sponsorships through direct sales. One time-bounded, clearly labeled sponsored placement, starting on a discovery surface, never touching factual research or overriding a candidate's dismissal. Price and format get tested with buyers against real referral evidence, not invented beforehand.

That order is the whole business thesis. Sponsorships only work once there is an audience worth sponsoring to. And the audience only returns if review stops being repetitive: new opportunities worth investigating, evidence for fit, decisions that persist.

If you are exploring AI labs right now — especially if you are coming from another industry the way I came from video — try the search and tell me which lab you discovered that you had never heard of. That answer is the metric I care about most.