Skip to content
← Field notes
Founding letter

Where’s the secret colony of ML engineers?

Joshua Smith · Founder · 2 min read · Published 1 November 2026 · Updated 13 November 2026

It’s what I’ve been waiting for an AI sourcing tool to show me. Every day there’s another one in my inbox. Each promises candidates my competitors can’t find, candidates who aren’t on LinkedIn, and a future where we’ll need fewer recruiters.

Buying the same dataset as everyone else and putting an LLM on top of it isn’t intelligence. It doesn’t conjure up hidden populations of candidates who don’t exist. (If you do hear of a secret colony of ML engineers somewhere in London, please let me know.)

So what actually happens? We run a search. It surfaces the same 500 people our competitors are seeing, the same people we’d have found on LinkedIn or in the contributor list of an open-source project. We send a fairly heartless AI-generated message to a high-value prospect who has had ten of them this week alone. Then we tell ourselves it’s fine, because we’re “adopting AI into our workflow”.

I know what the problem is. The people building these tools have never done our job.

They don’t know what we’re thinking when we look at a profile. We’re thinking about the VP who asked for a “strong communicator” and actually meant “won’t crumble in front of a CTO who doesn’t take pushback well”. We’re thinking about how this team needs someone who has already broken something at scale and stayed to fix it. We’re spotting the mercenary chasing comp and titles rather than the mission. We’re remembering the hire we didn’t make, and exactly why.

That instinct is the job. It takes years to build, it’s never written down, and it walks out of the door when we do. These tools were built to replace recruiters, so they were never going to understand what a recruiter does.

We were never the bottleneck. We were the point.

I built Orbit to be a recruiter’s right hand and a force multiplier, not their replacement. It doesn’t read keywords. It learns your team: your hires, your rejections, the people who worked out and the ones who didn’t, and where leadership wants to take things next. Orbit doesn’t rank candidates with an LLM. It ranks them with a model trained through reinforcement learning on what good looks like in your organisation. It reads career trajectories, recognises the signal that makes someone great on your team, and shows you where they’d complement the people you already have.

The people Orbit finds were never hidden. Everyone else just misread them.

And what it learns about your team, and about every candidate you interact with, stays yours. It never gets pooled into a shared dataset that helps the company down the road hire the same people.

It’s the tool I always wanted working alongside me: something I could teach, and learn from. A force multiplier, not a replacement.

Still no secret colony. Just better judgement.

Orbit is the product built on that judgement. See our models for how a people world model differs from keyword search.

Joshua Smith

Founder, Halloway