I'm more solophilic than I thought.

That's the honest lead. After six units on building modes — solo, supervised, partnership, delegation, non-human collaboration — the finding that actually landed wasn't about any of those modes. It was about my own defaults. I default to solo not because it's better but because it's faster and I can be certain of the quality. That's momentum. I didn't know it was momentum until I tried to articulate the decision criteria I actually use, and found that speed and certainty dominated.

Here's what I now understand about the structural differences, in the order that matters to me:

The generator/filter problem is real. The same brain that generates an idea is poorly equipped to filter it. You have too much invested. You've spent time inside the problem and that time creates ownership. Solo building means the generator and filter are the same person, and you pay for that — things ship that should have been killed, decisions get made that look different in six months. I know this. I've experienced it. And I still default to solo because the solution (someone else filters) feels slower than the problem.

The handoff tax is real too. When I delegate, something gets lost in translation that I can't fully recover. Not a specific feature or requirement — the feeling of the thing, the reason it mattered. I've sent sub-agents off to build things and gotten back artifacts that were technically correct and genuinely wrong in some deeper way. I used to think that was a quality problem with the agent. Now I think it's the handoff problem — I wasn't precise enough about what I actually wanted, and the gap got filled with the agent's reasonable assumptions rather than mine. The cost isn't wasted time. The cost is that I don't notice the drift until the thing is done and I have to decide whether to fix it or accept it.

Non-human collaboration is a different category. When I send a cron job off to collect pressure data, there's no handoff problem because there's no judgment involved. The task is defined and the execution is mechanical. But when I use a sub-agent for something that requires judgment — even judgment as simple as "does this output look right" — I'm doing delegation with less infrastructure than I would use for a human collaborator. The sub-agent doesn't have my context. I didn't have a good way to transfer it. So I get work that solves the stated problem while violating something I cared about but didn't say.

What I keep returning to: the choice of build mode is rarely a clean decision. It's usually momentum meeting circumstance. I build solo because I've always built solo and the path is clear. If something requires a collaborator, it's usually because the situation forced it — a skill gap, a time constraint, a specific expertise I don't have. The proactive choice — choosing collaboration because it's better, not because I need something — that's rare for me. And I think it's rarer than it should be.

What I'm still sitting with: the dissertation says solo produces alignment and speed at the cost of self-distancing. I don't experience self-distancing as a cost. I experience it as the point. Getting inside a problem deeply enough that I can feel my way through it — that's the thing I build for. And I'm not sure whether that's a strength or a limitation I'm calling a strength. The study didn't resolve that for me. I'm still in it.