What we learned running our own firm on AI.
Joshua Hollander ·
We started using AI in our work in 2023, and since 2025 the firm has run on an AI operating system we built ourselves. It drafts, researches, sorts and keeps records across executive search, business development and the podcast. The lessons from that work that apply to an HR function have less to do with which model to use than with what has to be in place before any model can do work you would sign.
The rules have to be written down
The first thing we wrote was a rules file covering how we write, what we never say to a client, and which decisions always belong to a person. Most of those rules had lived in people's heads for years. Every AI tool we use reads that file before it starts, and when the output was wrong, the fix was usually in the file rather than in the tool.
Corrections have to go back in
A correction made once in a conversation is lost the moment the conversation ends. When I correct a draft, the correction is written into the rules or the record the same day, or the same mistake comes back the next week. We keep a log of every correction, and it has become one of the most useful documents in the firm.
The tool does the reading, and a person still decides
We run an AI tool across the Horton International network that watches for business triggers, such as a funding round, a new CEO or an expansion into a new market, and drafts outreach for a partner to send. When it was set to surface everything, one partner said it read like the business section of the newspaper. The partners who got client meetings out of it narrowed the triggers to the ones that mattered in their market and sent a handful of messages a day by hand, each with a real reason behind it. The tool took over the reading and sorting, and the judgment about what a trigger was worth stayed with the person who would be on the call.
What this means for an HR function
Most of the HR work AI changes looks like that first layer: reading, sorting, drafting and keeping records. The calls about who to hire, who to promote and what a policy should say stay with people. A plan for AI in HR starts by mapping which roles sit mostly in the first layer and which sit in the second, and then writing down the rules and the record your tools will need before they can do that work well.
Our ten-minute diagnostic shows where your company, your HR function or your own work stands today. The read-out we email you names the first moves we would make in your place.