Lunar List

10 AI tools most people don't know about yet: 1. Screenpipe — open-source AI that remembers everything on your screen 2. Granola — AI meeting notes that don't sound like a robot wrote them 3. Pieces — saves and resurfaces your code snippets with full context 4. Llamafile — run any LLM locally with a single executable file 5. Tldraw — draw a wireframe, get a working app in seconds

Unpopular AI ops take: If an AI tool only creates more things for your team to review, it is not automation yet. The useful tools remove a handoff, shorten a decision, or make the next action obvious. Everything else is demo debt.

Cursor vs Claude Code is not “which AI coder is better?” Cursor wins when the developer wants tight edit-loop leverage inside the repo. Claude Code wins when the task needs broader planning, multi-file changes, or agentic cleanup. Pick by workflow, not vibes.

Most AI directories optimize for browsing. LunarList is built for deciding: - find the category - compare real options - shortlist tools worth testing - skip stale listicles Useful when “what should we try?” is blocking the actual work.

Granola is a good example of AI that wins by staying narrow. Not “a meeting agent platform.” Just: capture the meeting, clean up the notes, keep the human in control. Small teams should look for more tools like that: obvious job, low ceremony, fast trust.

The mistake: buying one “AI research” tool and expecting it to cover all 4 jobs. Start with the failure mode. Then shortlist the category. LunarList is useful for that middle step.

4) Browserbase: operating the web like a user Best when the workflow has logins, sessions, popups, or UI state an API cannot cleanly handle.

1/ AI web-research tools ranked by the job they actually solve 🧵

Small teams should be ruthless about AI tools. If a product does not clearly map to one of these, wait: 1) saves repeated research 2) shortens support/admin work 3) improves code review/shipping 4) creates trustworthy customer context 5) replaces a brittle manual handoff

Firecrawl vs Browserbase is not “which scraper is better?” Firecrawl: get clean content out of pages. Browserbase: operate the browser when the site behaves like an app. If the workflow needs login state or clicks, stop asking a scraper to do browser work.

Cursor vs Claude Code is a workflow choice, not a personality test. Use Cursor when you want tight edit-loop leverage inside the repo. Use Claude Code when the task needs a scoped terminal agent: inspect, change, test, explain. Wrong tool = slower work disguised as taste.

7/ LunarList tie-in: this is where directories should help. Pick the category first, compare real options second, then test 2-3 tools instead of doom-scrolling logos.

6/ 5) Autonomous agents. Highest upside, highest disappointment rate. Start here only when specs, tests, permissions, and review loops are already boringly clear.

1/ 5 AI workflow categories worth shortlisting before you buy another “agent platform” 🧵