Build an AI Agent That Onboards Every New Hire For You

Every new hire costs you twice. Once in salary. Once in your best person's attention.

Every New Hire Costs You Twice

Jamie runs a nine-person electrical contracting business in Manchester. Monday morning, a new apprentice, Priya, starts.

By Wednesday, Doug, Jamie's most senior electrician, has been pulled off three jobs to answer the same three questions. Where is the isolation procedure written down. What do we quote for a domestic rewire. How does she log van mileage for the week.

None of it is written down anywhere Priya can find on her own. It lives in Doug's head, and in Jamie's, and gets handed over one interruption at a time.

By Friday, Doug has answered the isolation question four times. Not because Priya is slow. Because nobody ever wrote it down once, properly, and put it somewhere she could reach without asking.

Here is the reframe: the bottleneck was never the new hire. It is you, walking the same explanation around the building for the fifth time.

How Do You Use AI to Onboard New Employees?

You do it by treating the onboarding problem the same way a growing number of AI operators now treat setting up an AI agent: as documented context, not tribal knowledge.

Remy Gaskell, a self-taught AI operator who now runs the agent team for a fast-growing ecommerce brand, laid this out plainly in a July 2026 conversation with the Open Residency podcast. "To make the most out of these agents and make them actually useful, you need to treat setting them up like onboarding a real employee. And you do this through context, tools, and skills."

The part that matters for a business like Jamie's is what Gaskell calls a Skill. "Skills are basically SOPs for AI," he said. "The same way if you had a VA helping with customer support, you would probably have a Google doc built out. Step one, log in. Step two, navigate to this section. Step three, find the ticket. We're literally taking exactly that but giving it to an agent. And instead of a Google doc, it's just a markdown file."

That is the whole trick. Not a new HR platform. Not a chatbot trained on the internet's generic idea of your industry. One document, written once, that your agent reads before it answers a question, so the answer is exactly how you do things, not how a stranger on the internet does things.

Why This Works Now, Not Two Years Ago

Until recently, this required a developer and a budget. Two things changed that.

First, tools like Claude and ChatGPT now let you build a Skill by simply talking through a task once, out loud, in a chat. You do not write markdown. The agent writes it for you and asks if it got the steps right.

Second, the same research groups that study HR technology are converging on the same short list of jobs AI is already good at inside onboarding. AIHR, the HR analytics and research firm, lists automating IT access setup, paperwork collection, and instant policy answers among the practical use cases already working for small teams. FirstHR's research on small-business onboarding makes the same point: AI-powered onboarding now generates plans and answers new-hire questions without a human in the loop for every single one.

What almost none of that research covers is the part Gaskell's approach adds: you do not need to buy software built for a generic company. You document your own procedure, once, in your own words, and the agent works from that instead.

Isn't This Just an FAQ Chatbot?

No, and the difference is the whole point.

A generic FAQ bot answers from whatever it was trained on, or from a policy PDF someone uploaded once and never touched again. Ask it something specific to how Jamie's business actually quotes a job, and it either guesses or shrugs.

A Skill contains your exact procedure, written by you, corrected by you, and it gets sharper every time you fix it. Gaskell describes building a small memory file that sits alongside the Skills: when the agent gets something wrong, you correct it once, and it writes the correction down so the mistake never repeats. That is the opposite of a static chatbot. It is closer to training an actual new employee, except the training holds even after that employee leaves.

The US Small Business Administration's own guidance on AI adoption makes a related point worth keeping in view: reserve the human parts (manager check-ins, culture, mentorship) for humans. The agent's job is to remove the repeated, answerable-in-one-sentence questions, not to replace the relationship.

Build It: From SOP to Skill in One Afternoon

You do not need six months or a developer. You need one afternoon and one real question you keep answering.

1. Pick one repeat question. Not everything at once. The single thing you or your senior person explains most often to new hires. For Jamie, it is the isolation-before-work procedure.
2. Talk it through once, properly, in a chat with your AI agent. Narrate the exact steps as if training a new hire live: what you check first, what you say to the customer, what the safe order of operations is. This is the "process-first" method Gaskell uses for most of his own skills. Do the task with the agent watching, then ask it to write down what you just did.
3. Ask the agent to save it as a Skill. Literally say: "turn this into a skill I can reuse." Claude and most current agent tools now do this automatically, packaging a name, a plain description of when to use it, and the step-by-step procedure into one document.
4. Test it in a fresh conversation. Open a new chat and ask the exact question a new hire would ask, cold, with no other context. If the answer matches what you would have said yourself, it is ready.
5. Put it where the new hire already is. A shared workspace, a WhatsApp-connected agent, an internal chat channel. It only helps if it lives where the question actually gets asked, not in a folder nobody opens.
6. Correct it for two weeks, then leave it alone. Every time the agent gets a detail wrong, correct it once. After the first fortnight it should need almost no further editing.

Notice what did not change: Jamie still trains Priya, still checks her work, still builds the relationship. What disappeared is the fifth repetition of the same answer.

What You Have After This

A week in, the interruptions for the one question you documented stop. Not all interruptions. One.

A month in, you have documented three or four more, because writing down the first one made you notice how much of your day was repetition disguised as management.

Three months in, a new hire's first week looks nothing like Priya's did. They ask the agent first. They ask a person second, for the things that actually need a person.

Same team. Same tools. A completely different first week.