Accelerator as a network protocol
I keep coming back to an old scribble: an accelerator as a protocol for selection, learning, and network effects. Free, remote, 0% equity. The idea was clear. The shape wasn’t. Is it one program or a partner mesh? 🚀👇
Found this in my notes from Aug 13, 2017 while cleaning the old folder: “Oswald accelerator in partnership with nobility accelerator and I focus accelerators.” In 2025 I added: “YC for the next billion founders, 0% equity, self‑taught” (yes, really!). In 2017, equity and in‑person cohorts ruled. Remote was fringe. Even the note was fuzzy on structure - network of partners or one wrapper?
What changed by 2025 is the bundle unraveled into four Cs.
• Curriculum went open and cohort based, with projects as the unit.
• Community moved to always‑on chat and async repos. Global mentorship feels normal.
• Credential shifted from brand to proof of work. Public code, shipping cadence, real users.
• Capital diversified: grants, credits, prizes, pre‑sales, small flexible checks.
Selection followed the same arc, from pedigree to proof‑of‑work. LLMs now draft docs, review code, and sketch experiments so humans lean into judgment and intros.
Zero‑equity, remote accelerators work when they modularize.
• Selection: open calls, light screens, public artifacts over essays.
• Learning: open curriculum, office hours, project milestones.
• Community: peer circles, alumni compasses, rotating experts.
• Capital: grants, credits, prizes, rev‑share pilots, optional checks later.
Tradeoffs: sustainability and who funds the commons. Signaling and whether non‑equity credentials matter. Depth at scale. The enduring moat is network density, not content.
Durable principles I’d keep: proof over pitch. Modular over monolith. Public by default. Network density beats content. Teach people to fish.
Open questions:
- Can we score selection with transparent signals without recreating gatekeeping?
- What funding stack keeps 0% equity resilient across cycles?
- How do we codify mentor judgment so it scales without losing taste?
Here’s the original note from 2017: https://github.com/AnandChowdhary/notes/blob/main/notes/2017/oswald-accelerator-in.md