Open Source Your SaaS Before AI Dooms It
Despite cloning and security risks, open-sourcing lets AI empower users to fork and customize, shattering incumbents' feature bloat moats and fueling growth—as seen in T3 Code's 1,500 forks from 16K weekly users.
Risks of Open-Sourcing Are Real but Manageable
Speaker acknowledges major downsides: competitors can point AI agents at your repo to clone features ("Rebuild this"), users self-host to avoid paying, and security vulnerabilities explode as agents scan open code. Cal.com struggles with massive security reports and exploits due to its open-source nature. T3 Chat remains closed because a small insecurity could cost millions on their edge infrastructure; their tiny 2.5-person team can't handle the fallout. Yet, these risks pale against closed-source decline: companies ship low-quality updates unchecked by AI hype, eroding trust.
"Agents are able to like decompile and figure out holes in systems relatively well, but they are much better at it for open-source projects." (Context: Explaining post-Mythos security surge; highlights why even advocates delay full open-sourcing.)
Incumbents Win Via Feature Bloat—Open Source Breaks It
Historical winners like AWS, Salesforce, and Retool dominate because they amass 1,000+ features over years, trapping customers. A typical user needs 50, but 25 are core (used by 80% of customers), 25 are rare (<1%), and 950 are niche. Migrating fails over that one bespoke feature used by 1%—10,000 customers for giants, but 1 for startups. Plugins help (Retool's integrations), but they're hell: unstable, support-heavy, lock you in. Vercel succeeds by hosting your code atop AWS, missing 95% of features but excelling at web app deploys with GitHub. Users plug in Supabase, Convex, or Cloudflare via code—no plugins needed.
This modular code-first model scales without bloat. AWS's failed all-in-one attempts (Amplify) prove modularity wins.
"If you have a million customers, a feature that's used by 1% of them is still used by 10,000 customers. If you have a 100 customers, a feature that's used by 1% of them is used by one team. That doesn't work." (Context: Why giants retain customers; reveals economic barrier startups can't match without open source.)
AI Lowers Forking Costs, Sparking User-Led Innovation
AI flips the script: non-devs fork and customize easily. T3 Code (open-source AI coding GUI wrapping Claude/Codex CLIs; bring-your-own-sub) hit 42K installs, 16K weekly users, 9K GitHub stars—and 1.5K forks (10% of weekly users). Users like Emanuel fork into "DP Code," adding multi-terminals, split chats, queuing, plugins from CMUX, even mobile support and handoff features. He praises it as "a skeleton to play with and build from."
This "magical" sharing loop feeds PRs back to core. Pre-AI, forking was dev-only and high-cost; now, everyone iterates. Imagine PostHog (open analytics): users add custom charts without 2-3 week support tickets or self-hosting ClickHouse clusters. HashiCorp thrives on Terraform forks/customs.
"Do you know how crazy that is? 10% of our users have forked and made some customization." (Context: T3 Code metrics; quantifies AI-driven forking explosion, proving demand for tweakable bases over rigid apps.)
Build Modular Skeletons—Let Customers Fork the Weird
Future: Ship extensible skeletons where users/AI handle bespoke needs. Don't build plugins; host their code. T3 Code proves it: users love the base, fork for edges. Advise portfolio companies: go all-in open source. Vercel hosts your app code; apply to apps like Retool/Salesforce by open-sourcing core, letting forks integrate weird data sources.
Closed source dooms you to quality decline and unmatchable bloat. Open wins by crowdsourcing the long tail via cheap AI forks.
"Simplest way I can put it, you got to let your customers make their weird shit." (Context: Core thesis while Ubering to YC Demo Day; reframes competition as empowering user customization over feature arms race.)
Speaker sponsors RWX CI: agent-ready with local run loops (CLI runs full CI with caching pre-commit), parallel caches (22s runs post-setup), dashboards. Beats GitHub Actions for AI dev.
Key Takeaways
- Acknowledge risks (security, cloning, self-hosting) but prioritize: open source anyway or face obsolescence.
- Target core 25 features perfectly; let AI forks handle the 5% long tail incumbents hoard.
- Build code-first platforms like Vercel: host user code for seamless integrations (Supabase, Convex).
- Measure success via forks: T3 Code's 1.5K forks (10% users) show AI unlocks mass customization.
- Ditch plugins (unstable, costly); forks + AI = user-led innovation without support burden.
- For apps like PostHog/Salesforce, open core enables custom charts/integrations atop your infra.
- Use tools like RWX for agent CI: local loops + caching guarantee fixes pre-commit.
- Invest in open source: speaker pushes portfolio companies here for moat in AI era.
- Track metrics: 42K T3 Code installs → 16K weekly → 1.5K forks signals product-market fit.
- Evolve from bloat to skeletons: users build the weird, you maintain the reliable base.