Managing an Alien Team

Intro

I didn't tiptoe into AI; I ran toward it. From the first week, it wasn't about replacing skill. It was about getting my spark back, turning ideas into things people can click, read, and buy. Over decades I've built in many corners of tech, but the audience of this story is you: business owners and agency leaders who want to move faster without losing your voice or your standards.

At some point I realized I wasn't "using a tool." I was coordinating a crew—helpers that don't share my assumptions but can move at breathtaking speed when briefed clearly and checked kindly. That's where the alien metaphor comes from, and we'll earn it together.

The Five States of Mind

I started with a simple dare: how far can I push one idea today? Not "Can AI code?" but "How quickly can I make this feel real for a customer?" The magic wasn't a breakthrough feature; it was building a daily habit. Small wins stack. Momentum builds on itself.

Here's how it actually looked one Tuesday: at 7:10 p.m. I wrote the idea for a new AI agent in plain words. By 7:35, I had a simple page—one headline, one box, one button. At 7:55, I texted the link to five customers and asked, "Does this help you today?" It wasn't polished. It moved the conversation forward.

When I'm stuck, I do three things

• Write one clear sentence a customer would say back to me.
• Ask for three different routes, each with one tradeoff.
• Ship the smallest slice I can show to one person today.

My biggest breakthrough wasn't a better tool; it was learning how to ask. AI helpers are fast and literal. I'm the one with context and judgment. When I speak in clear outcomes, I get work that feels "senior." When I speak in vague requests, I get vague results.

I use a five-line brief anyone can write:

A Tuesday I loved
At 7:10 p.m. I wrote five lines. At 7:35 p.m. I had a believable first draft. At 7:55 p.m. I was demoing it to a friend with zero apologies. Not perfect—present.

Speed didn't come from asking one helper to do everything. It came from giving each helper one clear job and keeping a simple beat.

Examples of one-job tasks

• Give 5 plain-English headlines for our AI bookkeeping service for solo founders, 6–10 words, no puns.
• Write the hero only for a one-screen landing page: headline (≤8 words), subhead (≤18 words), CTA label (2–3 words) for “Start free trial”.
• Write 3 follow-up emails after a demo request; 45–65 words each, human tone, one clear ask, no emojis.
• From these 10 customer replies about onboarding, list the top 3 themes; 1 sentence per theme. [paste replies]
• Define 5 metrics that tell me if this week is better than last for our paid newsletter; include a one-line definition and data source per metric.

Between helpers I add checkpoints: previews, small tests, quick thumbs-up or thumbs-down. Think bandleader, not one-man band.

When the flow arrives, protect it. I don't go slower; I add safety nets so the speed is safe to keep.

Midnight save
I once added a simple "preview" step and caught a polite but wrong email before it reached thousands. The win wasn't speed. The win was speed I could survive.

Great work rarely comes from one big launch. It comes from loops: one slice per day, one real reaction, one small polish, one new proof. Do this for twelve weeks and you'll look up at a product you couldn't have planned on a whiteboard.

My week when I'm on track

The metaphor I earned: the alien team

I didn't start with this line; I grew into it. Modern assistants feel like a crew of specialists who aren't from here. They don't share our defaults, but they bring remarkable stamina and speed. My job is to translate the mission, assign one clear task at a time, and certify the output.

They are brilliant. They are literal. They do not tire. They are not me. And that's perfect—because I bring taste, judgment, and stakes. I know what "done" means when someone's time or money is on the line.

Vision sets the rails. Vibes run the train. Compounding wins the race.

PM’ing agents is not managing people

With people, you rely on their experience; with agents, assume they know 100× more facts than you—instantly—but have zero lived context, zero taste, and zero stakes unless you impose them. Treat agents like systems, not employees. Your job is to design the work, certify outputs, and run the operation.

People vs agents (what changes)

• Context: People infer intent from history; agents need explicit goals, scope, and examples.
• Judgment: People weigh tradeoffs; agents follow rails unless you define escalation rules.
• Knowledge: Agents recall vast facts; you supply the constraints, priorities, and what “good” means.
• Learning: People absorb norms; agents learn via versioned prompts, data, and evals.
• Trust: People respond to values; agents respond to guardrails, SLAs, and rollback paths.

Operate like a COO: contractual, observable, test‑driven, versioned, cost‑aware, safety‑first, and relentlessly iterative. That’s how you PM AI agents.

If you copy one thing from me: envision with AI as the engine, keep guardrails like crazy, and fuse human judgment with machine scale.

Back to Real Wave Blog