Nobody taught you how to delegate and that's why you're not using AI Agents
Josh Miller posted something this week that stuck with me. The tech for AI agents is ready. The models are insane. Every big tech company has an agents platform. And yet your friends outside tech, the ones glued to their phones all day, don't use any of it. ChatGPT and Claude are popular, sure, but most people use them like a glorified Google with a grammar checker bolted on.
He frames it as a puzzle worth a lot of money in 2027. I think part of the answer is simpler and less flattering than we'd like.
Most people have never had an assistant. And briefing an assistant is a skill.
The skill nobody practised
Siqi Chen put it well in a reply to that thread: to get value out of agents today, you're basically telling person-shaped things what to do all day, and most people aren't used to telling anyone what to do.
Think about who actually delegates in their daily life. Executives with assistants. Managers with teams. Maybe people who've hired contractors for a house renovation. That's a small slice of the population. Everyone else does their own work, their own errands, their own research. When you've spent your whole life doing things yourself, handing a task to someone else feels slower than just doing it. You have to explain the context. You have to say what you want. You have to define what "done" means.
That explanation step is the whole game. It's also where everyone gives up.
I see it constantly. Someone tries an agent, types "plan my trip to Rome", gets back a generic itinerary that ignores their budget, their kids, and their hatred of museums, and concludes the technology is overhyped. The technology did exactly what it was told. The brief was the problem.
A bad brief and a good brief
Here's the request most people write:
Find me a hotel in Rome.
And here's what the same request looks like when you treat the agent like a capable assistant on their first day:
Find me a hotel in Rome for 4 nights in February, two adults and a 7-year-old. Budget around 150 to 200 euros per night. We want to walk to the main sights, so somewhere central, Trastevere or near the Pantheon. We need a triple room or a family room, not two separate rooms. Breakfast included is a plus. Give me your top 3 options with prices and a one-line reason for each.
Same tool. Same model. Completely different outcome. The second version isn't clever prompting or some technique you need a course for. It's what you'd say to a travel agent on the phone in 1995.
Four things changed between those two requests, and they're the same four things that make any brief work, whether you're briefing an intern or a machine.
Context. Who is this for and what's the situation? The agent doesn't know you have a kid or that you've already been to Rome twice.
Constraints. Budget, dates, dealbreakers. Without them, the agent optimises for nothing in particular.
Examples. "Somewhere like the place we stayed in Lisbon" does more work than three paragraphs of adjectives. Show, don't describe.
Definition of done. "Top 3 options with a one-line reason each" tells the agent when to stop and what shape the answer should take. Skip this and you get either a wall of text or a single take-it-or-leave-it answer.
You already know how to do this
The strange part is that everyone briefs people all the time without noticing. When you ask a friend to pick up something from the pharmacy, you don't say "get medicine". You say which pharmacy, which brand, the backup option if it's out of stock, and to call you if the price looks wrong. Context, constraints, examples, definition of done. All four, delivered in one WhatsApp message, without ever thinking about it.
The block isn't ability. It's that talking to software this way feels weird. Twenty years of search engines trained us to type three keywords and pick from ten blue links. Search punished long queries. Agents reward them. The habit is exactly backwards.
There's also a psychological piece that I think matters more than anyone admits. Delegating means accepting that the result won't be exactly what you'd have done yourself. Managers learn this the hard way over years. A first-time agent user hits it in the first five minutes, doesn't have a name for the feeling, and quits.
Where I'd tell a friend to start
Not with anything important. Pick a task that's real but small and completely reversible. Researching a purchase you're going to make anyway. Comparing three phone plans. Summarising a contract before you read it properly yourself.
Then write the brief like a message to a smart friend who knows nothing about your situation. Say who it's for. Give the constraints. Point at an example if you have one. Say what you want back and how long it should be.
The first result will be maybe 70% right. Don't start over. Reply. "Option 2 looks good but the dates are wrong, and I said breakfast included." An agent takes correction better than any human intern, doesn't sulk, and applies it instantly. The back-and-forth is the workflow, and treating the first answer as a draft instead of a verdict is the single biggest mindset shift.
Delegation is a skill, and like any skill it feels awkward for the first week and invisible after a month. The people getting real value from agents right now aren't smarter or more technical. They just got through the awkward week.