AI Takes Over Online Docs — You Don't Have to Be the Courier Anymore

The leg that was broken

We’ve all used online docs for ten years. Multiple people editing the same sheet, no mailing attachments back and forth — that was solved long ago, nothing new.

But one place has always been broken. AI finishes the work, the output sits on its side, and you have to copy and paste it into the doc yourself. The other party fills the doc in, AI has no idea, and you have to export it and feed it back in again. Those two moves in the middle have always been you playing courier.

Over the past couple of days I connected that leg. Put simply, the document collaboration that used to happen between me and my clients and colleagues is now handled by AI.

Concretely, it works like this. AI builds the sheet in Tencent Docs and writes the content in, I drop the link into a group chat, the client opens it inside WeChat and fills it in, and once it’s filled AI reads it back and carries on calculating. That courier in the middle isn’t your job anymore.

Why Tencent Docs

There are plenty of online doc tools, so why do I use Tencent Docs? Frankly, one reason — whether the person you send it to can just start working on it.

In Tencent’s latest earnings report, WeChat’s monthly actives are over 1.4 billion. Your clients, your suppliers — as individuals, essentially 100% of them have WeChat. What the other person opens it with is their WeChat account. No install, no signup.

What it can actually do

The first thing, and what I use most myself — every job where your part is finished and you’re still waiting on somebody else to act. Requirement checklists the client has to fill in, budget sheets you need to reconcile together, action item trackers for a project. Here, the part that used to burn the most time wasn’t the doing, it was the chasing and the merging. Now AI builds the sheet, I send the link, the other party fills it in, AI reads it back, and all I do in the middle is glance at it.

Second, the sheet genuinely calculates. I wrote a formula into a cell, then read it back — and what came back wasn’t the formula, it was the computed number. Which means the server really did run the calculation. So budget sheets, quote sheets, anything that has to do arithmetic can genuinely be handed off.

Third, I’m closing a loose end. In an earlier episode I had CC connect to Tencent Meeting and pull down meeting transcripts, and I left something hanging: the transcript came down, the minutes got generated, and there was nowhere to put them. That link is now closed. The meeting ends, the transcript comes down automatically, AI distills it into minutes and writes them into Tencent Docs, the action items get broken out into a sheet, and you drop a link in the group chat and you’re done.

Two things to know before you start

1. Security red line: that token is not workspace-scoped

Getting the key is easy. Go into your own Tencent Docs workspace, click the three bars in the top right, choose Use MCP, then click Get MCP token and copy it out. It’s still free for now. No app to create, no admin to go find.

The official documentation says this token is workspace-scoped. I understood it that way at first too, and thought the design was rather nice — give AI the key to one workspace and it can only enter that one room.

So I ran a test. I took that one workspace’s token and used it to list every workspace under my account. All of them came back, seven, not one missed. Including a private workspace I’d named the encrypted one — everything inside it came back readable.

So put plainly, this token is not workspace-scoped, it’s account-scoped, which means every read and write permission across all your Tencent Docs is handed over. I don’t think this is a bug. I think it’s the design. But you have to know it going in. My own standard: store this key at the level you’d store a finance password. Not on a shared machine, not handed to outsiders, and not written anywhere that syncs off the device.

2. Methodology: don’t read the docs, ask it what it has

This one I think may be worth more than the whole Tencent Docs exercise.

I opened the official developer documentation and it listed twelve tools. My first reaction was that this was far too few. So instead of writing code against the docs, I just asked it directly — what do you actually have?

Over two hundred.

And of the twelve written in the documentation, seven no longer exist in production. The names are dead. Which means if you build against the official docs, your first line of code is already wrong.

The logic is simple. Documentation is written by people, and the moment it’s written it starts expiring. The interface is the machine answering you live, and it can’t lie to you. So from now on, when you connect any AI tool interface, the first thing to do is not read the docs. The first thing to do is ask it what it actually has.

Source: EP0069 digital-avatar voiceover (pinpinvoice0717b voice) · text = final voiceover script (ground truth) · timecodes aligned sentence-by-sentence by gemini-2.5-pro (chunked parallel ASR, 98.2% coverage)

[00:00] Everyone’s been using online docs for ages. Multiple people editing the same sheet, no more emailing attachments back and forth — that all got solved ten years ago, nothing new. But there’s one spot that’s always been broken. AI finishes the work for you, the stuff sits on its side, and you have to copy-paste it into the doc yourself. The other person fills in the doc, AI doesn’t know about it, and you have to export it and feed it back in again. So those two moves in the middle — that’s been you, playing courier. Well, these past couple of days I connected that stretch. Put simply, the document collaboration I used to do with clients

[00:25] and with colleagues, AI now does for me. Here’s how it works concretely. AI creates the sheet directly in Tencent Docs and writes the content in. I drop the link into a group chat, the client taps it open right inside WeChat and fills it in, and once they’re done AI reads it back and keeps calculating from there. That courier in the middle — you don’t have to be it anymore. So with all the online doc tools out there, why do I use Tencent Docs? Honestly, one reason: when you send it over, can the other person just get to work? In Tencent’s latest earnings report, WeChat’s monthly actives are over 1.4 billion,

[00:51] so your clients, your suppliers — as individuals, basically 100% of them have WeChat. What the other person opens it with is their WeChat account. Nothing to install, nothing to register for. Okay, let’s start with what it can actually do. The first one, and the one I use the most myself, is the kind of work where you’re done but you still have to wait on the other side. A requirements checklist for a client to fill in. A budget sheet you need to go over together. A project action-item tracker. With this kind of thing, what used to eat the most time wasn’t making it — it was chasing people and merging versions back and forth. Now it’s AI builds the sheet, I send the link, they fill it in, AI reads it back,

[01:16] and I just glance at it in the middle. Second, the spreadsheet genuinely computes. I wrote a formula into a cell, then read it back. What came back wasn’t the formula — it was the calculated number. Meaning the server side really did run it. So budget sheets, quote sheets, that kind of work where numbers have to be crunched, can genuinely be handed off. Third is a loose end I left earlier. I did an episode a while back about having CC connect to Tencent Meeting and pull down the meeting transcript.

[01:41] That episode left a loose end: the transcript got pulled down, the minutes got generated, and there was nowhere to put them. Now that link is filled in. The moment a meeting ends, the transcript comes down automatically, AI distills it into minutes and writes them into Tencent Docs, the action items get broken out into a sheet, and you drop a link in the group chat and you’re done. Okay, that’s what it can do. Next, two things you have to know before you go and do this yourself. First, the safety red line — and this part may be the most important thing today. Getting the key is actually simple. Go into your own Tencent Docs workspace,

[02:06] click the three bars in the top right, pick Use MCP, then click get MCP token and copy it out. That’s it — and it’s still free for a limited time. No app to create, no admin to go find. The official docs say this token is workspace-scoped. That’s how I read it at first too, and I even thought it was a nice design — I give AI the key to one workspace, and it can only get into that one room. So I ran a test. I took the token for that one workspace and used it to list every workspace under my account.

[02:31] Every single one came back. Seven of them, not one missed. Including a private workspace of mine called the encrypted version — everything inside that was readable too. So put plainly, this token is not workspace-scoped. It’s account-scoped. Which means you’ve handed over read and write access to all of your Tencent Docs. I don’t think that’s a bug — I think that’s the product design. But you have to know it going in. My own standard is to store this key at the level of a financial password. It doesn’t go on a shared machine, it doesn’t go to anyone outside, and it doesn’t get written anywhere that syncs out.

[02:57] The second thing is a methodology, and I think this one might be worth more than the Tencent Docs story itself. I opened up the official developer docs and they list twelve tools. My first reaction was, that’s way too few. So instead of writing code against the docs, I just asked it one question: what have you actually got? Over two hundred. And of those twelve listed in the docs, seven no longer exist in production — the names are dead. Which means if you build against the official docs, your very first line of code is wrong. And the reason is simple, really.

[03:23] Docs are written by people, and the moment they’re written they start going out of date. But the interface is the machine answering you live — it can’t lie to you. So from now on, when you connect to any AI tool interface, the first thing to do is not read the docs. The first thing is to ask it: what have you actually got? There’s one more detail here that matters to people already playing with MCP. I did not mount those two hundred-plus tools as MCP. Because their specs add up to over 200,000 characters, and if you stuff all of that into AI’s head, you pay that tax on every single conversation.

[03:48] What I did instead was write a little script, a few dozen lines, and call it like an ordinary interface. All AI needs to know is that this script exists. That way it can also go into a scheduled task, and it can be called by other AI. Now someone might ask, doesn’t Feishu have this too? Honestly, yes — Feishu has an official MCP, it’s pretty complete, and your own AI can connect to it just the same. But first you have to create a custom enterprise app and get an app id and secret, and it’s still in Beta. And if you want the other person to actually edit, they have to register and log into a Feishu account first.

[04:14] And it can’t carry everything either. Complex approval flows, anything that needs an audit trail for compliance, big sheets with tens of thousands of rows — don’t count on it for those. That’s work for a real system. What it carries is the stretch that breaks most easily: the stretch that needs the other person to cooperate. Is Tencent Docs the strongest on features? Probably not, I’d say. But your clients, your colleagues, your suppliers are already inside WeChat. The thing that actually gets the other person to sit down and finish filling in that sheet is this one. Okay, after all that, there’s still one question

[04:39] I haven’t figured out myself. Which work is genuinely suited to being handed off, and which work becomes more trouble once you hand it off. To this day I don’t have a particularly clear answer to that either. So let’s talk it through together in the comments. That’s it for this episode. See you next time, bye.