Sam Saffron co-founded Discourse over a decade ago because he thought internet forums had stopped getting better.

In this episode, he talks about what's actually changed now that AI is inside the development loop, why he thinks most teams are misusing it, and how Discourse is building AI into moderation, search, and community management without breaking what makes communities work.

Joan and Sam get into the mechanics: how Sam codes with AI tools day-to-day, what MCPs are and why context management is the real problem nobody's solving well, and why programming languages matter less than they used to. They also talk about what it takes to run a platform that has to serve a 10-person forum and a 10-million-post community on the same codebase.

Takeaways

• Most AI hype in engineering teams is real, but the gains are unevenly distributed
• Moderation at scale requires AI, but trust requires humans
• Context and provider flexibility matter more than which model you pick
• Delegation and planning are harder to automate than writing code
• A community platform that does everything is harder to build than it sounds, and probably worth it

Chapters

• 00:00 Sam on AI hype: what's real and what's theatre
• 05:22 How AI changed Discourse's engineering team
• 13:00 AI moderation: where it works, where it breaks
• 19:05 The reporting problem: what communities flag vs. what they should
• 24:22 Scale, activity, and moderation load
• 29:42 Why provider flexibility matters more than model choice
• 54:27 One platform for every community: the product vision