Half my team are agents

Our team at Superpower is about 40 people. Twenty of them are human. The other twenty are AI agents.

That's not a metaphor. That's not "we use AI tools." I mean we have agents with names, roles, and responsibilities - sitting alongside the growth team in our shared systems, doing real work, every day.

And honestly? The line between the two groups is getting blurry.

Here's the setup.

We have a shared repo. Everyone on the team (the human everyone) creates agents, updates agents, and pushes to main. It's the same workflow you'd use for code. Branch, build, push, merge.

I built a medical reviewer agent. It checks health content for clinical accuracy before anything goes out the door. Someone else built a sales agent. Someone else built a creative reviewer that tears apart ad copy with more honesty than any human on the team is willing to deliver. They all have names. They all have specialties. And once they're merged into the repo, the entire team has access to them.

Think about that for a second. I spend a few minutes building an agent that solves a problem I keep running into. I push it. And now twenty other people can use it instantly without me training anyone, without a meeting, without a Loom video. The agent is the knowledge transfer.

It's like hiring a specialist, onboarding them in a day, and then cloning them across the whole company.

Okay so, as you know I don’t really do many sponsorships. This is a company that is run by my friend Omar and I think has such elite design and thoughtfulness around their product.

Now that ajays.quest is live, I need to know when something breaks — but I'm not going to sit there watching dashboards all day. That's where Adora comes in. Their AI Insights feature acts as a product-aware agent that monitors my site 24/7 and catches issues I'd never find on my own - failed flows, error loops, empty states — before users even complain.

I can’t even talk about how much time it saves because it’s literally just something a human can’t do.

Here's where it gets weird. In a good way. Individual agents are useful. But when you get multiple agents to spin up and work on a task together, the output is on a different level.

We call them agent teams. You take the medical writer, pair it with the creative reviewer and the sales agent, point them at a single task - say, writing a landing page for a new health product and let them collaborate. One drafts. One checks the science. One optimizes for conversion. They argue with each other (sort of). They iterate. And what comes out the other side is better than any one of them would produce alone.

It's the same principle that makes human teams work: diverse perspectives applied to a shared problem. Except these teammates don't have calendar conflicts, don't need context-setting, and can spin up in seconds.

I'm not saying agents replace the humans on our team. The humans decide what to build, who to talk to, and why something matters. But the agents handle a growing share of the how and they do it fast enough that the humans can focus on the work that actually requires taste, judgment, and relationships.

I pointed them at the concept and let them run. The meeting kept going. We kept talking. And about thirty minutes later, Mira and Marcus had a working prototype ready for us to look at. Not a slide deck. Not a brief. A prototype with user flows shaped by behavioral psychology and medical insights already baked in.

We pulled it up on screen and the room went quiet for a second. Not because it was perfect. But because we'd gone from "hey, what if..." to something tangible in the time it took to finish the meeting. The team could react to a real thing instead of debating an abstract one.

Here's what I think most companies are missing.

Everyone's talking about "adopting AI." They buy seats for ChatGPT, maybe they set up a few automations. That's not a strategy. That's a toy.

The real shift is when agents become shared infrastructure. When you treat them like team members that live in a repo versioned, tested, accessible to everyone. When building an agent isn't a side project but a core part of how you contribute to the team.

Our repo is becoming something like an org chart for agents. You can look at it and see: here's our sales agent, here's our creative reviewer, here's our outreach agent. Each one has an owner (the human who built it), a history (every commit), and a purpose. When someone has a new use case, they don't start from scratch. They fork an existing agent, adapt it, and push.

The workflow looks like this:

It's collaborative, it compounds, and it means the team gets smarter every single week without hiring anyone.

That’s it this week. I know I’m an absolute AI agent obsession but I feel like this inflection has been as big as when chatGPT first came out for me.

Until next time,

Ajay

A few tools and collections I’ve built (or obsessively curated) over the years:

These are tools and services I use personally and professionally — and recommend without hesitation: