It is also the practice where the wrong claim is easiest to make. Every firm in search now says it recruits in AI. What we can point to is a record — one engagement in particular, because at seed stage the record and the company are nearly the same thing.
Talk from knowledge
In AI, the vocabulary is free. Anyone can learn the words — fine-tuning, retrieval, agents — in a weekend. Talk shows up in the first answer. Knowledge shows up in the follow-ups: what did you try that didn't work, where did it break, what tradeoff did you choose and what did it cost you, what would you not do again?
Telling the two apart is the whole job in this practice, and a résumé can't do it. A working conversation can — but only if the person asking the questions can follow the answers.
We can, because we don't come to this work as readers. We run AI in our own firm. We build our own agents and the harnesses that hold them together, and we live with what they get wrong. That is roughly once-a-decade work here — this firm has always built its own tooling rather than buying someone else's — and the AI rebuild is the current one. It is what lets us sit across from an engineering professional, hold a real discussion about the work itself, and hear the difference between someone who has done it and someone who has learned the words.
That difference is what we are actually selling in this practice. Anyone can forward a résumé with the right keywords on it.
It is also not a new judgment for us. We built a travel company's entire data science team before data science was known as a practice — starting with its very first true scientist, placed instead of another software engineer, who set up a lab to run A/B testing on the company's SEO. The names of the disciplines change. Telling a scientist from an engineer, and knowledge from talk, is the same work it was then.
Four people, and most of the seats were ours
In early 2022, an AI company of four people found us by word of mouth, through the startup community. The work itself had begun in 2015, as an R&D lab under a different name, building large language models into websites years before the rest of the world had language for what that meant. When the technology became concrete enough to be a product, it turned toward commerce: an AI co-pilot that lives inside an online store.
They came looking for a vice president of engineering. We ran the search and placed the vice president. Alongside it, we placed a principal engineer.
A company that size doesn't hire by department. It hires by survival, one critical seat at a time — and over the following years, most of those seats were ours to fill. Software engineers came next, in 2023. A marketing leadership search that winter. And a product leadership role we worked through with the founder for the better part of a year and a half — the job changed shape three times on paper, from product delivery to product-led growth, before it was right — filled in 2025 from our own network of prior candidates, not from any posting.
The product started working
In A/B testing across tens of thousands of shopping sessions, the co-pilot lifted store revenue by an average of forty percent. In the last sixty days of 2023, more than fifteen hundred stores signed up.
That is the moment a search firm's work gets tested. The four-person company suddenly had demand to meet, and the hires already in place were the core it had to meet it with. Today the company has grown to roughly twelve to fifteen people, is growing rapidly, and is preparing to raise its next round — in a segment that has a name now, agentic commerce, whose rules its founder is one of the people writing.
When a company starts with four people, the vice president of engineering, the principal engineer, the engineers, and the vice president of product are not a sample of the company. They are the core the rest of the company grows around.
The roles
- VP of Engineering / Head of AI & Machine Learning
- Research Scientists
- Machine Learning Engineers
- LLM Infrastructure Leads
- AI Safety Researchers
- Agentic Systems Architects
- Principal Engineers, AI/ML
- Product Leadership for AI products
How an AI/ML search runs here
It starts the same way every search here starts: not with the role, but with the question — if this placement is a success a year from now, what will we be pointing to that really worked? In AI that question does real work, because the titles are young and the jobs underneath them differ wildly. A research scientist at one company is a production engineer at another.
The product leadership search above took a year and a half partly for that reason: the job had to be defined before the right person could be found for it, and the definition changed three times. We would rather take that time than fill the wrong shape quickly.
And the relationship outgrew the searches, as it tends to here. We brought the company a customer of its own — and when the installation stalled in the dead week between Christmas and New Year's, we chased it down ourselves until it went live. No search agreement calls for that. It is most of why the phone keeps ringing.
Start with the question
Every AI/ML search we run starts with one question: a year from now, what worked? Tell us where you're headed and we'll tell you honestly whether we're the right firm to get you there. If we're not, we'll tell you who is.
Current openings, in this practice and otherwise, live on the jobs page — that is the one place they stay current.