Most people use AI the way they'd call a smart friend. You ask a question, get a good answer, hang up, and then do the work yourself.
An agent is closer to making your first hire. You give them a job description, a desk, a laptop, and a filing cabinet, and they get on with the job while you work on something else. That desk, laptop, and filing cabinet are what people mean by a harness: a program that gives the model tools, files, memory, and a loop, so it keeps working until the job is done.
Later today at Singapore University of Technology and Design (SUTD), I'm teaching that shift to students in the StartSomething student entrepreneurship program. They're working on startup ideas they'll pitch for a chance at the Baby Shark Fund, with the accelerator program run by WaveSparks, a founder first entrepreneurship program.
In the session, called "Your first AI agent," each student signs up for an LLM gateway, creates an API key, installs Hermes Agent, and picks a model. Then they make their first hire, a lead gen bot. It searches the web for companies and decision makers that fit their customer profile, checks each one against public sources, and saves the verified leads with source links to a spreadsheet on their desktop. For a student about to pitch, that's customer discovery they can start tonight. We'll also compare Hermes with five other harness agents: OpenClaw, OpenCode, Claude Code, Pi, and Amp.
Most founders I meet still treat AI as the smart friend and use it to write posts, emails, and decks. Their hours go into research: lead lists, competitor checks, grant rules, vendor comparisons. They search, check, write it down, and do it all again next week. That's the work a first hire takes off a founder's plate, and an agent can do it.
Running an agent well looks a lot like managing a new hire. A vague brief gets vague work. "Find me leads" goes nowhere, while "20 Singapore companies with 50 to 500 staff and a named head of ops, with a source for each" gives the bot something it can finish. If you can't write that brief, you may not know your customer well enough yet.
You'd also check a new hire's work before it reaches a client. Agents fill gaps with confidence, so the students' bot has to cite a public source for every lead and flag anything it can't confirm. And you'd never let a new hire send emails or spend money in your name on day one. Agents get the same limits until they've earned your trust.
Good staff leave something behind. A chat answer disappears when you close the window, but a workbook or a CRM entry stays, and your team can sort it, share it, and build on it next week.
Your first hire can hand you 50 leads. Deciding which five will buy, and going to talk to them, is still your job.
So, if your first hire started tomorrow, what would you hand them first? Write that job description, because that's the first agent you should build. If you can't think of anything, look hard at where your week actually goes.

