Questions
How do I evaluate an LLM API before committing to it for production?
Build a 50-100 example eval set from your real worst-case inputs, score each model on it, and look at variance not just mean. Don't trust benchmarks — they're not your workload.
Notion AI: is the Q&A over my workspace actually useful, or a gimmick?
For 'where did we decide X' or 'what did the meeting about Y say', Notion AI Q&A is genuinely useful. For complex reasoning across many pages, it's hit-or-miss.
Is Linear worth it for a 5-person team, or is it overkill?
Linear is great for 5 people, but you can also use GitHub Projects at that size. The right time to pay for Linear is when the friction of GitHub Projects starts costing you real time.
How do I get my engineering team to actually use the design system in Figma?
Make the right thing the easy thing: link the Figma components to a code library (shadcn, Radix, etc.), document variants with code snippets, and add design tokens to the codebase.
Why does Redis need so much memory? Can I get away with less?
Redis stores the working set in RAM. If your working set is bigger than your memory, you need either more memory or to evict old data with an LRU policy.
Is Postgres on Neon production-ready, or is it still risky for serious apps?
Yes, Neon is production-ready. The architecture (separated storage and compute, branching, serverless) is a real advantage for modern workloads, and their SLA is real.
Do I need LangChain for a simple RAG app, or can I do it with raw API calls?
For a simple RAG over one document type with one embedding model, raw API calls are simpler. Use LangChain when you have multiple data sources, models, or want a real agent.
Is GPT-4o good enough for production, or should I wait for the next model?
GPT-4o is good enough for most production workloads today. Wait only if you have a specific capability the new model unlocks.
Will Drizzle work with my existing Prisma migrations, or is it a hard switch?
Hard switch. Drizzle reads the live Postgres schema, not Prisma's migration history — but if your schema is already correct, you can adopt Drizzle without losing data.
Should I run Bun in Docker containers for production, or is bare-metal better in 2026?
Bun in Docker works well in 2026 — the official `oven/bun` image is small (~50 MB) and production-ready. Bare metal is faster on cold start but loses the deployment consistency.
Is Obsidian a good choice if I want to write a PhD thesis that compiles cleanly into LaTeX?
Yes — Obsidian is plain Markdown under the hood, and Pandoc handles the Markdown → LaTeX conversion well for most academic writing.