Ten years of hiring by hand, and why we built TalentOS

For ten years I did this job by hand. Recruiting across legal and tech, reading hundreds of thousands of candidates, and, the part people forget, selling and scaling the services businesses doing the hiring. I carried the business-development number as well as the delivery. That combination is unusual, and it is the whole reason TalentOS looks the way it does.

What a decade of doing it manually actually teaches you

Not that recruiting is slow. Everyone knows recruiting is slow. What it teaches you is that a small service business loses the same work over and over, in four places, and never sees it happen.

A lead arrives on a Friday afternoon and nobody answers. A search starts, and the rubric someone built last quarter (the real one, the one the hiring partner meant rather than the one in the job description) is gone. A great process lives in one person's head until they resign, and then it does not live anywhere. Growth stops the week the owner gets busy delivering, because the owner was the only one selling.

The diagnosis is not “slow.” It is stateless.

Every search, every intake, every process starts from nothing. You rebuild the context, use it once, and throw it away. That is why hiring the fifth person is exactly as hard as hiring the first, and why a service business with plenty of demand still cannot absorb it.

Slow and expensive are the complaints everybody makes. Stateless is the one that actually explains the pain, and almost nobody says it out loud.

We tried a copilot first. It was not enough.

The 2023 answer was a recruiting copilot, and it is still most of what the public record says about us. A tool answers when it is prompted and starts fresh every time you open it. That is the same statelessness wearing a nicer interface.

Saying that plainly is not an admission. It is the reason for the pivot. A worker is different from a tool because it keeps the context: what you corrected last month shapes what it does next month.

Four functions, four workers, one expert in the loop

Every service business runs on the same four functions, and they are rarely anybody's actual job:

• Sales and marketing: feast and famine, because the owner is the only rainmaker.
• Intake: leads rot, because nobody answers at 4:47 on a Friday.
• Hiring: capped by who you can personally find, and every search restarts.
• Operations: knowledge lives in heads, so a resignation is a data-loss event.

A digital worker owns each one: Mark on sales and marketing, Ira on intake, Sara on hiring, Ted on operations. TalentOS is the system underneath them, holding the rubrics, SOPs, candidate history and intake rules, which is the state a small business normally loses. That is the part you keep when an engagement ends.

And there is a person attached. Most AI fails in a service business for one boring reason: nobody owns it. It gets bought, gets a login, and goes quiet in five weeks. A forward-deployed operator runs the function for about ninety days with AI carrying the volume, then the routine moves to the worker while your COO-in-the-loop keeps the judgment. The expert-in-the-loop is the product, not a caveat.

The one thing I want you to test rather than believe

Every rating Sara gives cites the source document word for word. When she cannot find the evidence, she caps the score instead of guessing.

That is deliberately a different kind of claim from “more accurate.” Accuracy is something you have to take on trust, and this is a category where a vendor was recently caught overstating badly enough that every buyer now assumes inflation. Checkable is the position worth having. Bring a real role to a call, watch her read up to 300 applications in under a minute, then open any candidate and check the quote against the resume yourself.

Where this goes

The vision is not more software for a business that already has too much of it. It is that a five-to-fifty-person service business should be able to run all four functions properly without hiring four people it cannot afford, and should own the system that makes that true rather than renting it.

The numbers at the top of this page are what the manual version produced. The point of the product is to make that repeatable without a decade of somebody's life attached to it.

Frequently asked questions

What is TalentOS?

TalentOS is an AI operating system with an expert-in-the-loop for small service businesses. Four digital workers own the four functions a service business runs on, meaning sales and marketing, intake, hiring and operations, and the system underneath holds the rubrics, SOPs, candidate history and intake rules that a small business normally loses when someone leaves.

Where do the numbers on this page come from?

They come from a decade of agency recruiting done by hand: 160+ placements across legal and tech, 85% retention on those placements, and an average saving of $12,000 per hire against traditional agency fees. They are agency history, not output from the TalentOS product, and they are presented here as the background to why the product exists.

What does an expert-in-the-loop actually do?

A forward-deployed operator runs one of your four functions directly for roughly ninety days, with AI carrying the volume. After that the routine work moves to the digital worker and the operator stays on as a COO-in-the-loop for the judgment calls. It exists because most AI in a service business fails for one reason: nobody owns it.