Migrating off Actions is a pain, but the real lock-in is the network effect. Moving to something like Codeberg means losing all those "drive-by" contributors who won't bother signing up for a new instance just for one small fix

Though given the current flood of AI slop, filtering out the casuals might actually be a feature, not a bug

The asymmetry is brutal: generating slop costs pennies and five seconds, while reviewing it costs senior engineers hours. The natural filter of complexity is broken, so the only way out is to artificially reintroduce friction

Projects like Godot need to enforce a strict Proof of Work. Want your PR reviewed? Attach a video of the fix running or a profiler log. If you can't prove you ran the code locally, a bot should auto-close the ticket. Otherwise they're just going to drown

You don't need frameworks for one-off scripts, but in prod, you're going to need RAG, proper memory, tools, and orchestration anyway. Without standards you'll just end up writing your own janky framework on top of requests. LangChain is definitely a bloated mess, but it provides structure. The beauty of Elixir is that this structure (OTP) is baked into the language, not duct-taped on the side

Spot on. BEAM is great at surviving process crashes, but if the whole cluster goes down or you redeploy, that in-memory state evaporates. It's not magic. For agents that might hang around for days, pure Elixir isn't enough, you still need a persistence layer. The ecosystem is catching up (Oban Pro, FLAME), but in reality, we're still building hybrids: fast actors for active chats and a good old DB for history and long-running processes

The Let it crash concept is perfect for deterministic bugs, but does it work for probabilistic errors?

If an LLM returns garbage, restarting the process (agent) with the same prompt and temperature 0 yields the same garbage. An Erlang Supervisor restarts a process in a clean state. For an agent "clean state" = lost conversation context

We don't just need Supervision Trees, we need Semantic Supervision Trees that can change strategy on restart. BEAM doesn't give this out of the box, you still code it manually

The problem isn't the abstraction level, it's the loss of the incubation period. Moving from Assembly to C didn't remove the need to think through data structures. Prompt engineering, however, triggers skipping the reflection stage entirely. Constantly managing bots leaves zero bandwidth for the "are we building garbage?" question. AI scales typing speed, not the speed of architectural decision-making

That's the hidden price of fast development

Validation is always harder than generation. Writing yourself means building context brick by brick. Reviewing AI means reconstructing someone else's - often broken - logic from zero. Juggling three projects in parallel fragmentizes mental context like a 90s hard drive. After a few hours, more energy goes into task switching and hallucination hunting than actual engineering. Deep work has been replaced by high-speed bot micromanagement

Character matters but so does having people around you who are willing to call it early, before you've rationalized yourself into ignoring it

But less about personal brilliance and more about how social power actually works when money, status, and weak accountability intersect