My friend's position is that collaboration doesn't really pay anymore, at least not for the kind of stuff he and I actually build on the side. He'll tell you AI made him fast enough alone that a collaborator is mostly overhead now — someone to sync with instead of someone who helps. He's not wrong exactly, but something about it didn't ring totally true to me.
Like yes, I can sit down on a Saturday and have a working prototype by lunch of something that would've taken a team a week two years ago. No standups, no Slack threads, no "let me loop in Sarah." Just me and the agent, moving.
So my friend is right. Collaboration is slower and sometimes - even often - not necessary.
But he's making one mistake. He thinks the beginning is the work. The truth is that shipping something is just the beginning.
The part you can see is the part that got easy
I have dozens of those Saturday prototypes. My repo's kanban board is a graveyard of 80%-done ideas. My production projects constantly trip up or have silent fails I have to address.
It's the classic iceberg image all over again. Mocking the prototype, scaffolding the app, writing the first draft - even shipping those things doesn't necessarily mean its over.
Below the waterline is everything that actually decides whether the thing matters: distribution, reliability, depth, access, or just the taste to know what's worth finishing. AI definitely helps you build the thing. It won't tell you if anyone should care, and it won't carry it the last 20% when carrying it is boring and hard.
So my friend's solo-speed advantage is real. It's just an advantage at the easy part.
AI will say yes to everything
AI is the most agreeable collaborator you'll ever have. Ask it to build your idea and it builds your idea. Ask it if your idea is good and it'll find a way to tell you yes. It executes your taste faithfully, and that's exactly the problem. A human collaborator will tell you nobody cares about the feature you're in love with.
Your taste needs friction to get better.
And we are still, almost always, building for humans. Whether it's a product or a family newsletter, the question that matters is will this land with the people on the other end, and that's a question other humans answer better than any model. Building with someone is a shortcut to that calibration. You're not just adding hands, you're adding a second nervous system that's wired to other people.
A collaborator is also an accountability buddy
Solo + AI gives you infinite optionality. You can start anything, anytime, and get it to working on your own.
Infinite optionality is a problem.
When you can start everything, you may ironically finish less (or ship more slop which is even worse).
Someone else counting on the thing existing means you can't just let it slide into the backlog when no one's looking. They also, weirdly, give you permission to kill the things that aren't worth it, because alone, every half-idea stays alive too long, clogging your mental capacity because you haven't consciously closed the loop.
So what does collaboration even look like now?
This is where I get less sure.
The old model (divide and conquer, sync meetings, "you take the frontend, I'll take the backend") is more fluid and less useful than it was.
It'll be fun to see (and participate) in the next type of collaboration.
The closest I've gotten so far is two AI-pilled people in a shared repo moving absurdly fast. When both people actually know how to work with the tools, the coordination overhead decreases since each person's AI harness can carry a lot of that load. So the human collaborative work is in thinking and reviewing and improving not executing.
What I'm really curious about right now is how to build a real-time collaborative environment where humans and agents are co-equal participants in one live space. Not async branches and pull requests. More like a multiplayer game than a git workflow. Several people and several agents in the same room, building at once, in the moment. I don't know if anything like this actually exists yet. As far as I can tell, the primitives are showing up (shared agent sessions, collaborative canvases) but nobody's built the full substrate where humans and agents just work together live.
Somebody should. Somebody will. It'd be a hell of a thing.
Where I land (for now)
The bar for "worth collaborating on" went up. Pretending otherwise is just nostalgia for meetings or status.
But the things above that higher bar, the ones with real distribution problems, real taste problems, real "will this matter to anyone" problems, real maintenance and slog problems, those benefit more from collaboration than they used to, not less.
All I know is, it's still fun to collaborate with high-agency people.