I Hired 20 New Employees on Saturday.

I hired 20 new employees on Saturday.

They were not human. They were agents.

Each one has a specific job inside the business.

One helps me think through financials. Another does sales research. One works on website copy. Another can code the website. Others monitor email and Slack, help with new-customer onboarding, organize projects, and keep an eye on work that is falling behind.

That sentence would have sounded absurd not very long ago.

The Old Way of Hiring

Hiring twenty capable people used to mean job descriptions, recruiting, interviews, salaries, onboarding, management, and months of finding out whether they were actually good.

On Saturday, it meant setting up twenty focused AI agents.

Not Twenty Copies of the Same Chatbot

The important part is not the number.

The important part is that they are not twenty copies of the same chatbot.

Each has a role. Each has instructions, a defined set of skills, and the context needed to do one job well. The financial agent should think differently from the copywriter. The project manager should see different priorities than the person watching customer onboarding.

But they are not isolated.

I built shared, cloud-based company memory: our business, marketing, sales effort, active projects, operating decisions, and the source material each agent needs to do useful work. The agents have specific areas of focus, but they can draw on the wider company context when it matters.

Context Changes Everything

A normal employee does not arrive on day one knowing the history of the company. They learn through conversations, documents, experience, mistakes, and repetition.

These agents are different. Before they begin, you can give them the accumulated context of the business: what matters, how decisions get made, what has already been tried, and where the work stands today.

Then you make their jobs narrow enough to be useful.

The copywriting agent does not need to manage my calendar. The project manager does not need to write website code. The Slack agent does not need to make a financial recommendation.

But when the project manager needs to explain a major initiative, it can pull together the relevant history immediately. It does not need to ask three people where the latest document is, find an old email thread, or wait for a meeting.

That is already changing how I work.

The Most Useful Agent So Far

The most useful agent so far is my personal project manager. I feel slightly disloyal saying that, because the others are doing valuable work. But it has quickly become the place where I begin the day.

Every morning, I ask: “Show me my task list for today.”

It brings together my priorities, overdue items, active initiatives, and calendar. It helps me see whether the day I have planned is actually possible. It can flag conflicts, help fit work into the calendar, and tell me what deserves attention first.

More importantly, it is more realistic about time than I am.

I tend to look at a list and see possibility. It looks at the same list and sees four hours of work, two meetings, a deadline, and an impossible afternoon.

That is useful.

I also talk to it like a person.

I will say, “Tell me more about that project,” or “What did we decide last week?” or “What needs my attention right now?” It does not replace judgment. It gives me a much better starting point for judgment.

The Rest of the Team

The other agents are beginning to do the same thing in their own lanes. They monitor. They organize. They draft. They surface things that need attention. They make it easier to move from “I know this needs to happen” to a usable first version of the work.

I am also changing the input side of my workflow. This week I am adding Wispr Flow because voice is the only realistic way I can work across this much context without spending the day typing instructions into boxes.

The Hard Part Is Not Creating Agents

None of this means the system is finished.

This is a new experiment. The hard part is not creating agents. The hard part is teaching them enough about the business to be useful without making them vague, overreaching, or wrong.

It is deciding what each agent can do, what it should only recommend, and when a human needs to step in.

That is where the real work begins.

They can create more drafts, more analysis, more reminders, and more activity. So can everyone else.

The question is whether they help the business act with more clarity, learn from what happens, and improve the next move.

That is what I am testing now.

And it may be the first time I have hired a team where the onboarding took one Saturday.