Stop Prompting. Start Hiring.
The way we use AI is about to change, and the biggest winners won’t be engineers. They’ll be project managers and communicators.
A quick experiment before we jump in: I'm trying out a shorter, more direct format for this issue. I’d love to know what you think. Do you prefer this short format, my usual style, or something else? Just hit reply and let me know.
When we were building HyperWrite, I watched thousands of people use AI up close. The biggest surprise? The best prompters were almost never engineers.
They were English teachers, novelists, college professors, and salespeople. People who get paid to communicate clearly for a living. That’s all prompting is: expressing your intent with clarity. Those folks had spent their entire careers training that muscle.
Now, we’re about to leave prompt engineering behind entirely.
The next era won’t be about writing single prompts. It will be about hiring, org design, and project management. Soon, you won’t just be prompting an AI; you’ll be managing a team of them. The winners won't be prompt hackers, they’ll be people who know how to coordinate roles and communicate clearly. Think less "software engineer," more "project manager."
Why Do This?
I've said this for a long time: using AI to the fullest is raw leverage.
In a standard chat window, you are the bottleneck. You prompt, you wait, you copy-paste the output, you check the work, and you prompt again. When you set up an agent team, you hand off entire outcomes instead of micro-tasks. You step out of the loop, let the agents coordinate the tedious work among themselves, and get dramatically more done.
Build a Team, Not a Workflow
Instead of forcing one model to do everything, you build a small team:
- Specialist agents: In sales, you might have one agent for deep prospect research, one for cold outreach, and one for meeting prep.
- Verification agents: Agents whose sole job is fact-checking, code review, quality assurance, etc... whatever your task requires.
- A manager agent: One agent sits on top to coordinate the specialists, track deliverables, and run the pipeline. You talk to the manager, and the manager runs the team.
A lot of people get this wrong. When they hear "multi-agent system," they think of drag-and-drop flowchart builders, brittle node graphs, and complex wiring setups.
Those tools are going to feel obsolete very quickly.
You'll just spin up agents on real computers, talk to them in plain English, and onboard them the exact same way you’d onboard a smart new hire. You teach them, course-correct them, and let them talk to each other.
The Catch: This Only Works with Real Computer-Using Agents
One crucial detail: this does not work on standard web ChatGPT. You cannot go to a browser window, ask it to pretend to be five different people, and expect real results.
This requires agents running on real computers. Think tools like Claude Code (where one session can literally create and manage another Claude Code agent) or Grok Bot. The agents need a computer, so the manager can actually assign tasks, have specialists run them on the machine, verify the output, and report back to you.
The Garage Stage
I’m living this right now. In my apartment, I have five laptops running around the clock, each hosting a separate agent team working on a different project. It’s the scrappy, garage version of what Dario Amodei called "a country of geniuses in a datacenter."
Are the models flawless at this yet? No, not even close. But they can already handle meaningful, multi-step work today, and in a few months, this will become standard operating procedure. Practicing now gives you an unfair head start for when that happens.
Your Homework: Hire Your First Team
Pick one real project you want off your plate this week. Paste the full text of this email into Claude, ChatGPT, or your model of choice along with (1) your task and (2) the tools and agent platforms you have access to. Then send this prompt:
"Act as my coach for building an AI agent team. Ask me whatever questions you need to understand my project, then design my first agent org chart: define who to hire, what each agent's specific role is, what verification steps are needed, who reports to whom, and how they should hand off work. Once we align on the plan, walk me through onboarding and running them step by step."
Answer the questions, follow the plan, and start running your first team.
Not sure which tool to start with? Read the Claude Cowork vs Claude Code comparison, then get a starter kit for each.
Hit reply and tell me what you build.
Clear enough for my dad. Sharp enough for the frontier. If a week is boring, you don’t hear from me.