Tools That Almost Work for Freelancers and Knowledge Chaos

1. Notion templates vanish when pages are duplicated too quickly

So there I was, trying to clone a client hub in Notion to reuse the layout for another retainer gig. I duplicated the main dashboard, jumped tabs mid-load to fix something in Google Calendar, came back — and somehow half the linked databases were now plain text. No filters. No tabs. Just static text blocks where dynamic views used to be.

Turns out, Notion occasionally fails to preserve linked databases during fast duplication if you navigate away from the tab during the process. It’s not documented anywhere (yes, I looked). This only seems to happen with dashboards connected to multiple synced databases, especially those pulling from different workspaces.

If this happens, your best bet is to go into the original page’s history, restore a version that still has tables intact, and copy/paste sections manually. Fun workaround. Or waste of time. Hard to tell sometimes.

Side note: if you’re using Notion and juggling multiple database views with shared filters, don’t trust that duplicating your neatly set-up dashboard will be a clean copy. You might have to rebuild everything from scratch. Again. Just like I did before lunch.

2. Obsidian sync gets grumpy with iCloud Drive structure changes

I loved Obsidian right up until Apple decided to reorganize how iCloud Drive sorted app folders mid-update. Suddenly, two devices syncing the same vault labeled it differently: one said “ObsidianVault,” the other “Obsidian Vault.” One tiny space broke the entire sync. Half my notes went missing on the phone. Desktop had duplicates with git conflict markers.

This isn’t exactly Obsidian’s fault—it’s more like an iCloud naming quirk that doesn’t resolve unless you unlink and relink the vault manually. But there’s something scary about watching your entire graph network disappear between devices.

If you ever see note slugs in Obsidian suddenly get replaced by weird hashed timestamps, check your folder sync structure before assuming it’s corruption. It might literally just be iCloud getting cute with spacing again.

My fix: I moved my whole vault to Dropbox. Less sexy, far more stable. I kept aliases in iCloud Drive but disabled sync from within Obsidian mobile. It’s ten extra steps and shouldn’t be necessary but that’s kind of life with personal knowledge tools right now — friction disguised as optionality.

3. Airtable interface designer is still not ready for client handoff

I really wanted to love interface designer. I built a client project tracker, wired it into their content pipeline, set up status pills, and even color-coded Kanban views by team. Looked great. Then I tried to let the client edit the priority field… and they couldn’t.

Airtable hides critical field permissions inside the table base — not in the interface itself. So if you give your client access to the interface but not the base, they can’t interact with relational fields at all. And even if you grant them editor rights, most multi-select fields from a linked table show up as read-only unless they have access to the linked records too.

{
  "field": "Status",
  "editable": false,
  "reason": "field is from another base with restricted access"
}

That message never surfaces in the UI, but you’ll find it if you pop open dev tools. I found it by accident while inspecting an unclickable dropdown.

So yeah: if your client’s confused because they can see the button but clicking it does nothing — that’s why.

I ended up tossing Interface Designer for that project and moved to Coda instead, where user permissions are less opaque and deeply integrated. Took longer, saved my sanity.

4. Readwise Reader limits highlights per item based on source type

I hit this one while importing a longform WebMD page (no shame). I was highlighting generously — symptoms, studies, sidebars, source citations — then bam: highlight limit reached. Warning message appeared once. Then every new highlight just silently failed.

What I didn’t know: Reader distinguishes batches of highlights depending on whether the content’s a true uploaded file, browser article, or scraped newsletter-style email. Web articles scraped via extension (rather than via URL paste) can sometimes have artificial highlight caps applied because they’re bucketed into a legacy content ingestion method under the hood.

What actually worked

I re-imported the exact same link using “Paste link” instead of “Save to Reader” and suddenly all highlights were accepted. Gorgeous. No limit. So now I audit every link I save by deleting the extension version and readding it manually. Because of course I do.

This is nowhere in Readwise’s docs, but someone on Reddit confirmed the same edge case using JSTOR articles. Apparently there’s a cap to prevent malformed documents from crashing sync pipelines. Hidden feature or defensive design? Depends on who’s asking.

5. Tana works but not until you stop treating it like Roam

The first time someone pitched Tana to me they called it “Roam plus Notion with AI.” That should’ve been my first red flag. But hey, freelancers are masochists. So I spent four hours pulling all my client SOPs into a complex node tree. Tags, linked references, smart fields — the works.

And then the queries broke. Panels pulled in items from unrelated trees. #tags matching too broadly. Comments surfaced as results. It felt like a bad salad bar where everything shared the same spoon.

What finally clicked: Tana assumes your tag is a data schema unless you explicitly tell it not to be.

I had been tagging reference notes with #project-name just to keep them organized — not realizing that every instance of #project-name was turning into an auto-type with implied slots and references. It wasn’t just metadata… Tana thought I was defining an object class.

Once I toggled all those tags to “just tag” modes in the tag manager, the queries stopped returning garbage. So yeah: if Tana makes zero sense to you at first, it’s likely that your mental model is bringing in too much Roam noise without accepting Tana’s strict preference for structure.

6. Google Docs comments don’t copy between shared template files

This is dumb but somehow still true. I built a proposal template in Drive with time-saving comment suggestions meant for internal reference. Stuff like “Add their LinkedIn bio here” and “Double-check invoice terms.” Then I duplicated the file, shared it with a new client—and all comments were gone.

Turns out if you duplicate a Google Doc using “Make a copy” from inside a shared folder but the destination is outside a folder you own—or if the copy lands in the recipient’s Drive instead—inline comments get dropped. No warning. Nothing in the changelog. Silent fail.

Confirmed it in a few setups. If both files live under your own Drive and the recipient is an editor, comments persist. Toss in a Shared Drive with quirky permission levels? Comments vanish.

For now, I manually export my templates as .docx, reupload them, and then re-add the comments in a side panel using a linked Google Sheet as a comment reference mapping. Incredibly inefficient. But at least I don’t look like I forgot to add critical notes when things get sent out.

7. Slite’s AI search often ignores older threads or archived pages

If you’re using Slite as your async home for client knowledge, be aware: archived content is indexed but can sometimes be fully absent from AI query results. I ran into this trying to find a systems diagram we made last summer. Searched exact phrase, got nothing. Pulled up the archive manually, found it in five clicks.

This seems to happen when:

• The document was archived more than a few weeks ago
• No updates have been made since before the AI indexing rollout
• You’re not using an Enterprise plan with recurrent indexing enabled

In one support thread, a Slite rep said old archived items are queued for “heat-based reindexing” — meaning they only get reindexed if someone opens them manually. So to make your old stuff searchable again: click on it. Seriously.

I now run a monthly “poke the archive” ritual where I command-click through 10 old documents just to make sure the AI remembers they exist. It’s ridiculous, but kind of like feeding sourdough starter. Same energy.