Look at how AI is being sold to recruitment agencies right now and a pattern shows up immediately. You are not offered a capability. You are offered a staff.
There is a sourcing bot. A screening bot. An outreach bot. A scheduling bot, a note-taking bot, a compliance bot. Each has a name, a face, and increasingly a job title. The pitch is that you are building a team, and the number keeps going up, because more bots reads as more product.
We went the other way deliberately, and it is worth explaining why, because it is not a branding decision.
A roster of bots is a management problem
Agencies are not short of tools. They are short of attention. The binding constraint on a recruitment desk is how many things a person can hold in their head while three clients are waiting on feedback.
Every additional bot spends some of that constraint. You have to know which one to open. You have to learn where each one thinks your data lives. You have to configure each one separately, and check each one separately, and remember which of them is allowed to send things. A team of thirty-six anything requires management, and management is precisely the resource the desk has least of.
The pitch says you are getting employees. What you are actually getting is direct reports.
The technical cost is worse than the human one
Split the desk across many agents and something more damaging happens underneath: each one only sees its own slice.
The note-taking bot hears the client say they have gone off the candidate because the last two hires from that background did not stay. That is a strong, specific signal about how this client actually decides. The sourcing bot never hears it, because it is a different product with a different context window and no reason to be told.
So the next shortlist you get is built as though the conversation never happened. You correct it by hand, again, and the correction teaches nothing, because there is nowhere for it to be stored that the sourcing step will read.
This is the quiet reason bot rosters plateau. Each bot is individually reasonable and the system as a whole cannot learn, because learning requires that what you observe in one place changes what you do in another.
A note taken at 9am should change the shortlist at 2pm. That only works if the same thing did both.
What one agent buys you
Kit is a single agent across the whole desk. It takes the note, files it on the right record, drafts the follow-up, ranks the shortlist and chases the feedback. Because it is one agent, the sentence the client said this morning is available to the ranking this afternoon without anyone having to move it.
The compounding effect matters more than any individual feature:
- Every shortlist you cut teaches what a good candidate looks like on your desk, not on the market's.
- Every draft you rewrite teaches how your agency actually sounds to a client.
- Every rejection is signal, and rejections are the thing recruiters produce most of.
None of that requires you to configure anything. It is a by-product of working, which is the only kind of learning that survives a busy week.
The honest trade-off
One agent is not strictly better on every axis and it would be dishonest to claim otherwise.
A specialist product will usually beat a generalist at its one task on the day you compare them side by side. If your entire problem is outbound sequencing, a sequencing-first tool will out-feature us on sequencing. The single-agent bet is that the compounding beats the specialisation over a year, because your desk is not thirty-six separate problems. It is one problem observed from thirty-six angles.
There is also a control question, and it deserves a direct answer. One agent doing everything sounds like more autonomy, not less. It is the opposite in our design: Kit proposes, and a person accepts or rejects before anything reaches a candidate or a client. Nothing sends on its own. A single agent makes that easier to enforce, because there is one place where the rule lives rather than thirty-six.
We would rather ship one thing that gets to know your desk than a roster you have to run.