Tools-for-thought apps promise the same outcome — a second brain that compounds — and ship wildly different mental models for getting there. We spent two weeks running Reflect, Tana, and Capacities side-by-side as a working engineer's notebook: standup notes, design docs, RFC drafts, on-call postmortems, and an ongoing reading log. This is what actually held up.
What "tool for thought" means when you write code for a living
The PKM category is crowded with marketing copy that doesn't survive contact with a sprint. Engineers don't need a graph view to admire; you need:
- Sub-200ms capture latency. If it takes longer to open a note than to write it, you'll fall back to a terminal scratchpad.
- Plain-text or markdown export. Your notes outlive any single SaaS. Lock-in is the killer.
-
Predictable linking. Backlinks that don't silently break when you rename, with
[[wiki-style]]links that resolve fast. - Keyboard-first navigation. A mouse round-trip per note breaks flow.
- Real search across years of notes. Full-text, ranked, with operators.
Everything else — AI summarization, kanban, calendar integration, embedded queries — is a nice-to-have. We weighted the comparison around those five.
Reflect, Tana, and Capacities at a glance
The pricing converged around $10-15/month, so cost is rarely the deciding factor. What matters is the data model.
Reflect — opinionated and quiet
Reflect bets on one workflow: open the app, land on today's daily note, type, links resolve to backlinks automatically. There's almost nothing to configure. Capture was the fastest of the three in our measurement — roughly 140ms from cmd-N to a blinking cursor. The AI features ("ask my notes," outline generation, transcription) are baked in rather than bolted on.
The tradeoff is rigidity. If you want structured fields per note (status, priority, tags with values), Reflect mostly says no. It's a journal that grew up.
Tana — power-user maximum
Tana is what happens when you take Roam's bidirectional links, fuse them with Notion-style databases, and let everything be both a node and a query. Supertags are the headline feature: tag a node #book and it inherits fields (author, status, rating) and shows up in any view that queries #book.
If you already write Notion databases by hand, Tana is faster. But the learning curve is real — we spent the better part of a weekend understanding how supertags inherit, when search-nodes differ from queries, and how to design a workspace that doesn't collapse under its own structure once you cross a few hundred nodes.
Capacities — middle path with objects
Capacities sits between the two. Everything is an "object" with a type — Person, Book, Project, Meeting — and each type has its own template. It's lighter than Tana, more structured than Reflect. Backlinks work, but the unit of thought is the object, not the daily note.
The free tier is genuinely usable for solo work. Pro unlocks AI features and unlimited storage.
None of these three has a stable public REST API as of mid-2026. If you want to script your notebook — pipe commits into a daily log, auto-create meeting notes from calendar events, feed an RSS reader into a reading queue — you'll be doing it through manual export or browser automation. Obsidian and Logseq remain the better picks if scripting is a hard requirement.
Where each one breaks down
After two weeks, the failure modes were specific.
Reflect broke down on structured projects. Tracking a dozen open RFC reviews across three repos, each with status and reviewer, devolved into prose that didn't filter or sort. We ended up exporting to a CSV.
Tana broke down on quick capture. The supertag prompts and field-completion modals that make queries powerful slowed every standup note by a few seconds. Small in isolation, painful at forty captures a day.
Capacities broke down on cross-object queries. Asking "show every meeting where person X is attendee and the related project is on hold" required either manual page navigation or query syntax that's less expressive than Tana's. The object model is clean; the query layer hasn't caught up.
If you already live in Notion and the affiliate disclosure doesn't bother you, Notion's AI plus databases approximate roughly 70% of Tana with less mental overhead. The remaining 30% — node-level linking, true outlining, supertag inheritance — is where dedicated PKM still wins.
Picking one without regret
The honest decision tree we landed on:
- You want a journal that auto-builds a graph and you'll never script it. Pick Reflect. It's the most likely to still feel good in 18 months because there's nothing to maintain.
- You'll spend a weekend designing a workspace and want database-grade queries. Pick Tana. The ceiling is highest; budget the onboarding.
- You think in entities — books, people, papers, companies. Pick Capacities. The object model maps cleanly to bibliographies and CRM-like notes.
- You need scripting, plugins, or local-first storage. None of these. Use Obsidian or Logseq, and revisit this category in a year.
Originally published at pickuma.com. Subscribe to the RSS or follow @pickuma.bsky.social for new reviews.
Top comments (0)